Merge branch 'master' of github.com:comfyanonymous/ComfyUI

README.md

@@ -29,6 +29,7 @@ A vanilla, up-to-date fork of [ComfyUI](https://github.com/comfyanonymous/comfyu
    - [AuraFlow](https://comfyanonymous.github.io/ComfyUI_examples/aura_flow/)
    - [HunyuanDiT](https://comfyanonymous.github.io/ComfyUI_examples/hunyuan_dit/)
    - [Flux](https://comfyanonymous.github.io/ComfyUI_examples/flux/)
+   - [Lumina Image 2.0](https://comfyanonymous.github.io/ComfyUI_examples/lumina2/)
 - Video Models
    - [Stable Video Diffusion](https://comfyanonymous.github.io/ComfyUI_examples/video/)
    - [Mochi](https://comfyanonymous.github.io/ComfyUI_examples/mochi/)
@@ -1307,7 +1308,7 @@ For any bugs, issues, or feature requests related to the frontend, please use th
 
 The new frontend is now the default for ComfyUI. However, please note:
 
-1. The frontend in the main ComfyUI repository is updated weekly.
+1. The frontend in the main ComfyUI repository is updated fortnightly.
 2. Daily releases are available in the separate frontend repository.
 
 To use the most up-to-date frontend version:
@@ -1324,7 +1325,7 @@ To use the most up-to-date frontend version:
    --front-end-version Comfy-Org/ComfyUI_frontend@1.2.2
    ```
 
-This approach allows you to easily switch between the stable weekly release and the cutting-edge daily updates, or even specific versions for testing purposes.
+This approach allows you to easily switch between the stable fortnightly release and the cutting-edge daily updates, or even specific versions for testing purposes.
 
 ### Accessing the Legacy Frontend
 
@@ -1338,7 +1339,7 @@ This will use a snapshot of the legacy frontend preserved in the [ComfyUI Legacy
 
 ## Community
 
-[Chat on Matrix: #comfyui_space:matrix.org](https://app.element.io/#/room/%23comfyui_space%3Amatrix.org), an alternative to Discord.
+[Discord](https://comfy.org/discord): Try the #help or #feedback channels.
 
 ## Known Issues
 

comfy/__init__.py

@@ -1 +1 @@
-__version__ = "0.3.11"
+__version__ = "0.3.15"

comfy/api_server/routes/internal/internal_routes.py

@@ -1,13 +1,9 @@
-from typing import Optional
-
 from aiohttp import web
-
-from ...services.file_service import FileService
-from ...services.terminal_service import TerminalService
-from ....app import logger
-from ....cmd.folder_paths import models_dir, user_directory, output_directory, \
-    folder_names_and_paths  # pylint: disable=import-error
-
+from typing import Optional
+from folder_paths import folder_names_and_paths, get_directory_by_type
+from api_server.services.terminal_service import TerminalService
+import app.logger
+import os
 
 class InternalRoutes:
     '''
@@ -19,35 +15,19 @@ class InternalRoutes:
     def __init__(self, prompt_server):
         self.routes: web.RouteTableDef = web.RouteTableDef()
         self._app: Optional[web.Application] = None
-        self.file_service = FileService({
-            "models": models_dir,
-            "user": user_directory,
-            "output": output_directory
-        })
         self.prompt_server = prompt_server
         self.terminal_service = TerminalService(prompt_server)
 
     def setup_routes(self):
-        @self.routes.get('/files')
-        async def list_files(request):
-            directory_key = request.query.get('directory', '')
-            try:
-                file_list = self.file_service.list_files(directory_key)
-                return web.json_response({"files": file_list})
-            except ValueError as e:
-                return web.json_response({"error": str(e)}, status=400)
-            except Exception as e:
-                return web.json_response({"error": str(e)}, status=500)
-
         @self.routes.get('/logs')
         async def get_logs(request):
-            return web.json_response("".join([(l["t"] + " - " + l["m"]) for l in logger.get_logs()]))
+            return web.json_response("".join([(l["t"] + " - " + l["m"]) for l in app.logger.get_logs()]))
 
         @self.routes.get('/logs/raw')
-        async def get_logs_raw(request):
+        async def get_raw_logs(request):
             self.terminal_service.update_size()
             return web.json_response({
-                "entries": list(logger.get_logs()),
+                "entries": list(app.logger.get_logs()),
                 "size": {"cols": self.terminal_service.cols, "rows": self.terminal_service.rows}
             })
 
@@ -63,6 +43,7 @@ class InternalRoutes:
 
             return web.Response(status=200)
 
+
         @self.routes.get('/folder_paths')
         async def get_folder_paths(request):
             response = {}
@@ -70,6 +51,20 @@ class InternalRoutes:
                 response[key] = folder_names_and_paths[key][0]
             return web.json_response(response)
 
+        @self.routes.get('/files/{directory_type}')
+        async def get_files(request: web.Request) -> web.Response:
+            directory_type = request.match_info['directory_type']
+            if directory_type not in ("output", "input", "temp"):
+                return web.json_response({"error": "Invalid directory type"}, status=400)
+
+            directory = get_directory_by_type(directory_type)
+            sorted_files = sorted(
+                (entry for entry in os.scandir(directory) if entry.is_file()),
+                key=lambda entry: -entry.stat().st_mtime
+            )
+            return web.json_response([entry.name for entry in sorted_files], status=200)
+
+
     def get_app(self):
         if self._app is None:
             self._app = web.Application()

comfy/api_server/services/file_service.py

@@ -1,15 +0,0 @@
-from typing import Dict, List, Optional
-
-from ..utils.file_operations import FileSystemOperations, FileSystemItem
-
-
-class FileService:
-    def __init__(self, allowed_directories: Dict[str, str], file_system_ops: Optional[FileSystemOperations] = None):
-        self.allowed_directories: Dict[str, str] = allowed_directories
-        self.file_system_ops: FileSystemOperations = file_system_ops or FileSystemOperations()
-
-    def list_files(self, directory_key: str) -> List[FileSystemItem]:
-        if directory_key not in self.allowed_directories:
-            raise ValueError("Invalid directory key")
-        directory_path: str = self.allowed_directories[directory_key]
-        return self.file_system_ops.walk_directory(directory_path)

comfy/cli_args.py

@@ -35,12 +35,13 @@ def _create_parser() -> EnhancedConfigArgParser:
     parser.add_argument("--enable-cors-header", type=str, default=None, metavar="ORIGIN", nargs="?", const="*",
                         help="Enable CORS (Cross-Origin Resource Sharing) with optional origin or allow all with default '*'.")
     parser.add_argument("--max-upload-size", type=float, default=100, help="Set the maximum upload size in MB.")
+    parser.add_argument("--base-directory", type=str, default=None, help="Set the ComfyUI base directory for models, custom_nodes, input, output, temp, and user directories.")
     parser.add_argument("--extra-model-paths-config", type=str, default=None, metavar="PATH", nargs='+',
                         action='append', help="Load one or more extra_model_paths.yaml files.")
-    parser.add_argument("--output-directory", type=str, default=None, help="Set the ComfyUI output directory.")
+    parser.add_argument("--output-directory", type=str, default=None, help="Set the ComfyUI output directory. Overrides --base-directory.")
     parser.add_argument("--temp-directory", type=str, default=None,
-                        help="Set the ComfyUI temp directory (default is in the ComfyUI directory).")
-    parser.add_argument("--input-directory", type=str, default=None, help="Set the ComfyUI input directory.")
+                        help="Set the ComfyUI temp directory (default is in the ComfyUI directory). Overrides --base-directory.")
+    parser.add_argument("--input-directory", type=str, default=None, help="Set the ComfyUI input directory. Overrides --base-directory.")
     parser.add_argument("--auto-launch", action="store_true",
                         help="Automatically launch ComfyUI in the default browser.")
     parser.add_argument("--disable-auto-launch", action="store_true", help="Disable auto launching the browser.")
@@ -250,7 +251,9 @@ def _create_parser() -> EnhancedConfigArgParser:
         env_var="ANTHROPIC_API_KEY"
     )
 
-    parser.add_argument("--user-directory", type=is_valid_directory, default=None, help="Set the ComfyUI user directory with an absolute path.")
+    parser.add_argument("--user-directory", type=is_valid_directory, default=None, help="Set the ComfyUI user directory with an absolute path. Overrides --base-directory.")
+
+    parser.add_argument("--enable-compress-response-body", action="store_true", help="Enable compressing response body.")
 
     # now give plugins a chance to add configuration
     for entry_point in entry_points().select(group='comfyui.custom_config'):
@@ -284,6 +287,9 @@ def _parse_args(parser: Optional[argparse.ArgumentParser] = None, args_parsing:
     if args.disable_auto_launch:
         args.auto_launch = False
 
+    if args.force_fp16:
+        args.fp16_unet = True
+
     configuration_obj = Configuration(**vars(args))
     configuration_obj.config_files = config_files
     assert all(isinstance(config_file, str) for config_file in config_files)

comfy/cli_args_types.py

@@ -40,6 +40,7 @@ class Configuration(dict):
         config_files (Optional[List[str]]): Path to the configuration file(s) that were set in the arguments.
         cwd (Optional[str]): Working directory. Defaults to the current directory. This is always treated as a base path for model files, and it will be the place where model files are downloaded.
         base_paths (Optional[list[str]]): Additional base paths for custom nodes, models and inputs.
+        base_directory (Optional[str]): Set the ComfyUI base directory for models, custom_nodes, input, output, temp, and user directories.
         listen (str): IP address to listen on. Defaults to "127.0.0.1".
         port (int): Port number for the server to listen on. Defaults to 8188.
         enable_cors_header (Optional[str]): Enables CORS with the specified origin.
@@ -123,6 +124,7 @@ class Configuration(dict):
         user_directory (Optional[str]): Set the ComfyUI user directory with an absolute path.
         log_stdout (bool): Send normal process output to stdout instead of stderr (default)
         panic_when (list[str]): List of fully qualified exception class names to panic (sys.exit(1)) when a workflow raises it.
+        enable_compress_response_body (bool): Enable compressing response body.
     """
 
     def __init__(self, **kwargs):
@@ -131,9 +133,11 @@ class Configuration(dict):
         self.config_files = []
         self.cwd: Optional[str] = None
         self.base_paths: list[str] = []
+        self.base_directory = Optional[str] = None
         self.listen: str = "127.0.0.1"
         self.port: int = 8188
         self.enable_cors_header: Optional[str] = None
+        self.enable_compress_response_body: bool = False
         self.max_upload_size: float = 100.0
         self.extra_model_paths_config: Optional[List[str]] = []
         self.output_directory: Optional[str] = None

comfy/clip_model.py

@@ -103,9 +103,10 @@ class CLIPTextModel_(torch.nn.Module):
         mask = None
         if attention_mask is not None:
             mask = 1.0 - attention_mask.to(x.dtype).reshape((attention_mask.shape[0], 1, -1, attention_mask.shape[-1])).expand(attention_mask.shape[0], 1, attention_mask.shape[-1], attention_mask.shape[-1])
-            mask = mask.masked_fill(mask.to(torch.bool), float("-inf"))
+            mask = mask.masked_fill(mask.to(torch.bool), -torch.finfo(x.dtype).max)
+
+        causal_mask = torch.full((x.shape[1], x.shape[1]), -torch.finfo(x.dtype).max, dtype=x.dtype, device=x.device).triu_(1)
 
-        causal_mask = torch.empty(x.shape[1], x.shape[1], dtype=x.dtype, device=x.device).fill_(float("-inf")).triu_(1)
         if mask is not None:
             mask += causal_mask
         else:

comfy/cmd/folder_paths.py

@@ -65,7 +65,7 @@ def init_default_paths(folder_names_and_paths: FolderNames, configuration: Optio
     from ..cmd.main_pre import args
     configuration = configuration or args
 
-    base_paths = [Path(configuration.cwd) if configuration.cwd is not None else None] + configuration.base_paths
+    base_paths = [Path(configuration.cwd) if configuration.cwd is not None else None] + [Path(configuration.base_directory) if configuration.base_directory is not None else None] + configuration.base_paths
     base_paths = [Path(path) for path in base_paths if path is not None]
     if len(base_paths) == 0:
         base_paths = [Path(os.getcwd())]

comfy/cmd/server.py

@@ -80,6 +80,19 @@ async def cache_control(request: web.Request, handler):
     return response
 
 
+@web.middleware
+async def compress_body(request: web.Request, handler):
+    accept_encoding = request.headers.get("Accept-Encoding", "")
+    response: web.Response = await handler(request)
+    if not isinstance(response, web.Response):
+        return response
+    if response.content_type not in ["application/json", "text/plain"]:
+        return response
+    if response.body and "gzip" in accept_encoding:
+        response.enable_compression()
+    return response
+
+
 def create_cors_middleware(allowed_origin: str):
     @web.middleware
     async def cors_middleware(request: web.Request, handler):
@@ -169,7 +182,8 @@ class PromptServer(ExecutorToClientProgress):
         PromptServer.instance = self
 
         mimetypes.init()
-        mimetypes.types_map['.js'] = 'application/javascript; charset=utf-8'
+        mimetypes.add_type('application/javascript; charset=utf-8', '.js')
+        mimetypes.add_type('image/webp', '.webp')
 
         self.address: str = "0.0.0.0"
         self.user_manager = UserManager()
@@ -188,6 +202,9 @@ class PromptServer(ExecutorToClientProgress):
         self.background_tasks: dict[str, Task] = dict()
 
         middlewares = [cache_control]
+        if args.enable_compress_response_body:
+            middlewares.append(compress_body)
+
         if args.enable_cors_header:
             middlewares.append(create_cors_middleware(args.enable_cors_header))
         else:

comfy/comfy_types/node_typing.py

@@ -66,13 +66,26 @@ class IO(StrEnum):
         b = frozenset(value.split(","))
         return not (b.issubset(a) or a.issubset(b))
 
+class RemoteInputOptions(TypedDict):
+    route: str
+    """The route to the remote source."""
+    refresh_button: bool
+    """Specifies whether to show a refresh button in the UI below the widget."""
+    control_after_refresh: Literal["first", "last"]
+    """Specifies the control after the refresh button is clicked. If "first", the first item will be automatically selected, and so on."""
+    timeout: int
+    """The maximum amount of time to wait for a response from the remote source in milliseconds."""
+    max_retries: int
+    """The maximum number of retries before aborting the request."""
+    refresh: int
+    """The TTL of the remote input's value in milliseconds. Specifies the interval at which the remote input's value is refreshed."""
 
 class InputTypeOptions(TypedDict):
     """Provides type hinting for the return type of the INPUT_TYPES node function.
 
     Due to IDE limitations with unions, for now all options are available for all types (e.g. `label_on` is hinted even when the type is not `IO.BOOLEAN`).
 
-    Comfy Docs: https://docs.comfy.org/essentials/custom_node_datatypes
+    Comfy Docs: https://docs.comfy.org/custom-nodes/backend/datatypes
     """
 
     default: bool | str | float | int | list | tuple
@@ -113,6 +126,14 @@ class InputTypeOptions(TypedDict):
     # defaultVal: str
     dynamicPrompts: bool
     """Causes the front-end to evaluate dynamic prompts (``STRING``)"""
+    # class InputTypeCombo(InputTypeOptions):
+    image_upload: bool
+    """Specifies whether the input should have an image upload button and image preview attached to it. Requires that the input's name is `image`."""
+    image_folder: Literal["input", "output", "temp"]
+    """Specifies which folder to get preview images from if the input has the ``image_upload`` flag.
+    """
+    remote: RemoteInputOptions
+    """Specifies the configuration for a remote input."""
 
 
 class HiddenInputTypeDict(TypedDict):
@@ -133,7 +154,7 @@ class HiddenInputTypeDict(TypedDict):
 class InputTypeDict(TypedDict):
     """Provides type hinting for node INPUT_TYPES.
 
-    Comfy Docs: https://docs.comfy.org/essentials/custom_node_more_on_inputs
+    Comfy Docs: https://docs.comfy.org/custom-nodes/backend/more_on_inputs
     """
 
     required: dict[str, tuple[IO, InputTypeOptions]]
@@ -143,14 +164,14 @@ class InputTypeDict(TypedDict):
     hidden: HiddenInputTypeDict
     """Offers advanced functionality and server-client communication.
 
-    Comfy Docs: https://docs.comfy.org/essentials/custom_node_more_on_inputs#hidden-inputs
+    Comfy Docs: https://docs.comfy.org/custom-nodes/backend/more_on_inputs#hidden-inputs
     """
 
 
 class ComfyNodeABC(ABC):
     """Abstract base class for Comfy nodes.  Includes the names and expected types of attributes.
 
-    Comfy Docs: https://docs.comfy.org/essentials/custom_node_server_overview
+    Comfy Docs: https://docs.comfy.org/custom-nodes/backend/server_overview
     """
 
     DESCRIPTION: str
@@ -167,7 +188,7 @@ class ComfyNodeABC(ABC):
     CATEGORY: str
     """The category of the node, as per the "Add Node" menu.
 
-    Comfy Docs: https://docs.comfy.org/essentials/custom_node_server_overview#category
+    Comfy Docs: https://docs.comfy.org/custom-nodes/backend/server_overview#category
     """
     EXPERIMENTAL: bool
     """Flags a node as experimental, informing users that it may change or not work as expected."""
@@ -181,9 +202,9 @@ class ComfyNodeABC(ABC):
 
         * Must include the ``required`` key, which describes all inputs that must be connected for the node to execute.
         * The ``optional`` key can be added to describe inputs which do not need to be connected.
-        * The ``hidden`` key offers some advanced functionality.  More info at: https://docs.comfy.org/essentials/custom_node_more_on_inputs#hidden-inputs
+        * The ``hidden`` key offers some advanced functionality.  More info at: https://docs.comfy.org/custom-nodes/backend/more_on_inputs#hidden-inputs
 
-        Comfy Docs: https://docs.comfy.org/essentials/custom_node_server_overview#input-types
+        Comfy Docs: https://docs.comfy.org/custom-nodes/backend/server_overview#input-types
         """
         return {"required": {}}
 
@@ -198,7 +219,7 @@ class ComfyNodeABC(ABC):
 
     By default, a node is not considered an output. Set ``OUTPUT_NODE = True`` to specify that it is.
 
-    Comfy Docs: https://docs.comfy.org/essentials/custom_node_server_overview#output-node
+    Comfy Docs: https://docs.comfy.org/custom-nodes/backend/server_overview#output-node
     """
     INPUT_IS_LIST: bool
     """A flag indicating if this node implements the additional code necessary to deal with OUTPUT_IS_LIST nodes.
@@ -209,7 +230,7 @@ class ComfyNodeABC(ABC):
 
     A node can also override the default input behaviour and receive the whole list in a single call. This is done by setting a class attribute `INPUT_IS_LIST` to ``True``.
 
-    Comfy Docs: https://docs.comfy.org/essentials/custom_node_lists#list-processing
+    Comfy Docs: https://docs.comfy.org/custom-nodes/backend/lists#list-processing
     """
     OUTPUT_IS_LIST: tuple[bool]
     """A tuple indicating which node outputs are lists, but will be connected to nodes that expect individual items.
@@ -227,7 +248,7 @@ class ComfyNodeABC(ABC):
     the node should provide a class attribute `OUTPUT_IS_LIST`, which is a ``tuple[bool]``, of the same length as `RETURN_TYPES`,
     specifying which outputs which should be so treated.
 
-    Comfy Docs: https://docs.comfy.org/essentials/custom_node_lists#list-processing
+    Comfy Docs: https://docs.comfy.org/custom-nodes/backend/lists#list-processing
     """
 
     RETURN_TYPES: tuple[IO]
@@ -237,19 +258,19 @@ class ComfyNodeABC(ABC):
 
         RETURN_TYPES = (IO.INT, "INT", "CUSTOM_TYPE")
 
-    Comfy Docs: https://docs.comfy.org/essentials/custom_node_server_overview#return-types
+    Comfy Docs: https://docs.comfy.org/custom-nodes/backend/server_overview#return-types
     """
     RETURN_NAMES: tuple[str]
     """The output slot names for each item in `RETURN_TYPES`, e.g. ``RETURN_NAMES = ("count", "filter_string")``
 
-    Comfy Docs: https://docs.comfy.org/essentials/custom_node_server_overview#return-names
+    Comfy Docs: https://docs.comfy.org/custom-nodes/backend/server_overview#return-names
     """
     OUTPUT_TOOLTIPS: tuple[str]
     """A tuple of strings to use as tooltips for node outputs, one for each item in `RETURN_TYPES`."""
     FUNCTION: str
     """The name of the function to execute as a literal string, e.g. `FUNCTION = "execute"`
 
-    Comfy Docs: https://docs.comfy.org/essentials/custom_node_server_overview#function
+    Comfy Docs: https://docs.comfy.org/custom-nodes/backend/server_overview#function
     """
 
 
@@ -267,7 +288,7 @@ class CheckLazyMixin:
         Params should match the nodes execution ``FUNCTION`` (self, and all inputs by name).
         Will be executed repeatedly until it returns an empty list, or all requested items were already evaluated (and sent as params).
 
-        Comfy Docs: https://docs.comfy.org/essentials/custom_node_lazy_evaluation#defining-check-lazy-status
+        Comfy Docs: https://docs.comfy.org/custom-nodes/backend/lazy_evaluation#defining-check-lazy-status
         """
 
         need = [name for name in kwargs if kwargs[name] is None]

comfy/component_model/folder_path_types.py

@@ -13,7 +13,7 @@ from typing import Any, NamedTuple, Optional, Iterable
 
 from .platform_path import construct_path
 
-supported_pt_extensions = frozenset(['.ckpt', '.pt', '.bin', '.pth', '.safetensors', '.pkl', '.sft', ".index.json"])
+supported_pt_extensions = frozenset(['.ckpt', '.pt', '.pt2', '.bin', '.pth', '.safetensors', '.pkl', '.sft' ".index.json"])
 extension_mimetypes_cache = {
     "webp": "image",
 }

comfy/extra_config.py

@@ -7,7 +7,7 @@ import yaml
 def load_extra_path_config(yaml_path):
     from .cmd import folder_paths
 
-    with open(yaml_path, 'r') as stream:
+    with open(yaml_path, 'r', encoding='utf-8') as stream:
         config = yaml.safe_load(stream)
     yaml_dir = os.path.dirname(os.path.abspath(yaml_path))
     for c in config:
@@ -32,5 +32,6 @@ def load_extra_path_config(yaml_path):
                     full_path = os.path.join(base_path, full_path)
                 elif not os.path.isabs(full_path):
                     full_path = os.path.abspath(os.path.join(yaml_dir, y))
-                logging.info("Adding extra search path {} {}".format(x, full_path))
-                folder_paths.add_model_folder_path(x, full_path, is_default=is_default)
+                normalized_path = os.path.normpath(full_path)
+                logging.info("Adding extra search path {} {}".format(x, normalized_path))
+                folder_paths.add_model_folder_path(x, normalized_path, is_default=is_default)

comfy/k_diffusion/sampling.py

@@ -1311,7 +1311,7 @@ def sample_dpmpp_2m_cfg_pp(model, x, sigmas, extra_args=None, callback=None, dis
     return x
 
 @torch.no_grad()
-def res_multistep(model, x, sigmas, extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1., noise_sampler=None, cfg_pp=False):
+def res_multistep(model, x, sigmas, extra_args=None, callback=None, disable=None, s_noise=1., noise_sampler=None, eta=1., cfg_pp=False):
     extra_args = {} if extra_args is None else extra_args
     seed = extra_args.get("seed", None)
     noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler
@@ -1333,53 +1333,60 @@ def res_multistep(model, x, sigmas, extra_args=None, callback=None, disable=None
         extra_args["model_options"] = model_patcher.set_model_options_post_cfg_function(model_options, post_cfg_function, disable_cfg1_optimization=True)
 
     for i in trange(len(sigmas) - 1, disable=disable):
-        if s_churn > 0:
-            gamma = min(s_churn / (len(sigmas) - 1), 2**0.5 - 1) if s_tmin <= sigmas[i] <= s_tmax else 0.0
-            sigma_hat = sigmas[i] * (gamma + 1)
-        else:
-            gamma = 0
-            sigma_hat = sigmas[i]
-
-        if gamma > 0:
-            eps = torch.randn_like(x) * s_noise
-            x = x + eps * (sigma_hat**2 - sigmas[i] ** 2) ** 0.5
-        denoised = model(x, sigma_hat * s_in, **extra_args)
+        denoised = model(x, sigmas[i] * s_in, **extra_args)
+        sigma_down, sigma_up = get_ancestral_step(sigmas[i], sigmas[i + 1], eta=eta)
         if callback is not None:
-            callback({"x": x, "i": i, "sigma": sigmas[i], "sigma_hat": sigma_hat, "denoised": denoised})
-        if sigmas[i + 1] == 0 or old_denoised is None:
+            callback({"x": x, "i": i, "sigma": sigmas[i], "sigma_hat": sigmas[i], "denoised": denoised})
+        if sigma_down == 0 or old_denoised is None:
             # Euler method
             if cfg_pp:
-                d = to_d(x, sigma_hat, uncond_denoised)
-                x = denoised + d * sigmas[i + 1]
+                d = to_d(x, sigmas[i], uncond_denoised)
+                x = denoised + d * sigma_down
             else:
-                d = to_d(x, sigma_hat, denoised)
-                dt = sigmas[i + 1] - sigma_hat
+                d = to_d(x, sigmas[i], denoised)
+                dt = sigma_down - sigmas[i]
                 x = x + d * dt
         else:
             # Second order multistep method in https://arxiv.org/pdf/2308.02157
-            t, t_next, t_prev = t_fn(sigmas[i]), t_fn(sigmas[i + 1]), t_fn(sigmas[i - 1])
+            t, t_next, t_prev = t_fn(sigmas[i]), t_fn(sigma_down), t_fn(sigmas[i - 1])
             h = t_next - t
             c2 = (t_prev - t) / h
 
             phi1_val, phi2_val = phi1_fn(-h), phi2_fn(-h)
-            b1 = torch.nan_to_num(phi1_val - 1.0 / c2 * phi2_val, nan=0.0)
-            b2 = torch.nan_to_num(1.0 / c2 * phi2_val, nan=0.0)
+            b1 = torch.nan_to_num(phi1_val - phi2_val / c2, nan=0.0)
+            b2 = torch.nan_to_num(phi2_val / c2, nan=0.0)
 
             if cfg_pp:
                 x = x + (denoised - uncond_denoised)
+                x = sigma_fn(h) * x + h * (b1 * uncond_denoised + b2 * old_denoised)
+            else:
+                x = sigma_fn(h) * x + h * (b1 * denoised + b2 * old_denoised)
 
-            x = (sigma_fn(t_next) / sigma_fn(t)) * x + h * (b1 * denoised + b2 * old_denoised)
+        # Noise addition
+        if sigmas[i + 1] > 0:
+            x = x + noise_sampler(sigmas[i], sigmas[i + 1]) * s_noise * sigma_up
 
-        old_denoised = denoised
+        if cfg_pp:
+            old_denoised = uncond_denoised
+        else:
+            old_denoised = denoised
     return x
 
 @torch.no_grad()
-def sample_res_multistep(model, x, sigmas, extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1., noise_sampler=None):
-    return res_multistep(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, s_churn=s_churn, s_tmin=s_tmin, s_tmax=s_tmax, s_noise=s_noise, noise_sampler=noise_sampler, cfg_pp=False)
+def sample_res_multistep(model, x, sigmas, extra_args=None, callback=None, disable=None, s_noise=1., noise_sampler=None):
+    return res_multistep(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, s_noise=s_noise, noise_sampler=noise_sampler, eta=0., cfg_pp=False)
 
 @torch.no_grad()
-def sample_res_multistep_cfg_pp(model, x, sigmas, extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1., noise_sampler=None):
-    return res_multistep(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, s_churn=s_churn, s_tmin=s_tmin, s_tmax=s_tmax, s_noise=s_noise, noise_sampler=noise_sampler, cfg_pp=True)
+def sample_res_multistep_cfg_pp(model, x, sigmas, extra_args=None, callback=None, disable=None, s_noise=1., noise_sampler=None):
+    return res_multistep(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, s_noise=s_noise, noise_sampler=noise_sampler, eta=0., cfg_pp=True)
+
+@torch.no_grad()
+def sample_res_multistep_ancestral(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
+    return res_multistep(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, s_noise=s_noise, noise_sampler=noise_sampler, eta=eta, cfg_pp=False)
+
+@torch.no_grad()
+def sample_res_multistep_ancestral_cfg_pp(model, x, sigmas, extra_args=None, callback=None, disable=None, eta=1., s_noise=1., noise_sampler=None):
+    return res_multistep(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, s_noise=s_noise, noise_sampler=noise_sampler, eta=eta, cfg_pp=True)
 
 @torch.no_grad()
 def sample_gradient_estimation(model, x, sigmas, extra_args=None, callback=None, disable=None, ge_gamma=2.):

comfy/ldm/flux/math.py

@@ -22,7 +22,7 @@ def attention(q: Tensor, k: Tensor, v: Tensor, pe: Tensor, mask=None) -> Tensor:
 
 def rope(pos: Tensor, dim: int, theta: int) -> Tensor:
     assert dim % 2 == 0
-    if model_management.is_device_mps(pos.device) or model_management.is_intel_xpu():
+    if model_management.is_device_mps(pos.device) or model_management.is_intel_xpu() or model_management.is_directml_enabled():
         device = torch.device("cpu")
     else:
         device = pos.device

comfy/ldm/hunyuan_video/model.py

@@ -301,7 +301,7 @@ class HunyuanVideo(nn.Module):
             shape[i] = shape[i] // self.patch_size[i]
         img = img.reshape([img.shape[0]] + shape + [self.out_channels] + self.patch_size)
         img = img.permute(0, 4, 1, 5, 2, 6, 3, 7)
-        img = img.reshape(initial_shape)
+        img = img.reshape(initial_shape[0], self.out_channels, initial_shape[2], initial_shape[3], initial_shape[4])
         return img
 
     def forward(self, x, timestep, context, y, guidance=None, attention_mask=None, control=None, transformer_options={}, **kwargs):

comfy/ldm/lumina/model.py

@@ -0,0 +1,622 @@
+# Code from: https://github.com/Alpha-VLLM/Lumina-Image-2.0/blob/main/models/model.py
+from __future__ import annotations
+
+from typing import List, Optional, Tuple
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import comfy.ldm.common_dit
+
+from comfy.ldm.modules.diffusionmodules.mmdit import TimestepEmbedder, RMSNorm
+from comfy.ldm.modules.attention import optimized_attention_masked
+from comfy.ldm.flux.layers import EmbedND
+
+
+def modulate(x, scale):
+    return x * (1 + scale.unsqueeze(1))
+
+#############################################################################
+#                               Core NextDiT Model                              #
+#############################################################################
+
+
+class JointAttention(nn.Module):
+    """Multi-head attention module."""
+
+    def __init__(
+        self,
+        dim: int,
+        n_heads: int,
+        n_kv_heads: Optional[int],
+        qk_norm: bool,
+        operation_settings={},
+    ):
+        """
+        Initialize the Attention module.
+
+        Args:
+            dim (int): Number of input dimensions.
+            n_heads (int): Number of heads.
+            n_kv_heads (Optional[int]): Number of kv heads, if using GQA.
+
+        """
+        super().__init__()
+        self.n_kv_heads = n_heads if n_kv_heads is None else n_kv_heads
+        self.n_local_heads = n_heads
+        self.n_local_kv_heads = self.n_kv_heads
+        self.n_rep = self.n_local_heads // self.n_local_kv_heads
+        self.head_dim = dim // n_heads
+
+        self.qkv = operation_settings.get("operations").Linear(
+            dim,
+            (n_heads + self.n_kv_heads + self.n_kv_heads) * self.head_dim,
+            bias=False,
+            device=operation_settings.get("device"),
+            dtype=operation_settings.get("dtype"),
+        )
+        self.out = operation_settings.get("operations").Linear(
+            n_heads * self.head_dim,
+            dim,
+            bias=False,
+            device=operation_settings.get("device"),
+            dtype=operation_settings.get("dtype"),
+        )
+
+        if qk_norm:
+            self.q_norm = RMSNorm(self.head_dim, elementwise_affine=True, **operation_settings)
+            self.k_norm = RMSNorm(self.head_dim, elementwise_affine=True, **operation_settings)
+        else:
+            self.q_norm = self.k_norm = nn.Identity()
+
+    @staticmethod
+    def apply_rotary_emb(
+        x_in: torch.Tensor,
+        freqs_cis: torch.Tensor,
+    ) -> torch.Tensor:
+        """
+        Apply rotary embeddings to input tensors using the given frequency
+        tensor.
+
+        This function applies rotary embeddings to the given query 'xq' and
+        key 'xk' tensors using the provided frequency tensor 'freqs_cis'. The
+        input tensors are reshaped as complex numbers, and the frequency tensor
+        is reshaped for broadcasting compatibility. The resulting tensors
+        contain rotary embeddings and are returned as real tensors.
+
+        Args:
+            x_in (torch.Tensor): Query or Key tensor to apply rotary embeddings.
+            freqs_cis (torch.Tensor): Precomputed frequency tensor for complex
+                exponentials.
+
+        Returns:
+            Tuple[torch.Tensor, torch.Tensor]: Tuple of modified query tensor
+                and key tensor with rotary embeddings.
+        """
+
+        t_ = x_in.reshape(*x_in.shape[:-1], -1, 1, 2)
+        t_out = freqs_cis[..., 0] * t_[..., 0] + freqs_cis[..., 1] * t_[..., 1]
+        return t_out.reshape(*x_in.shape)
+
+    def forward(
+        self,
+        x: torch.Tensor,
+        x_mask: torch.Tensor,
+        freqs_cis: torch.Tensor,
+    ) -> torch.Tensor:
+        """
+
+        Args:
+            x:
+            x_mask:
+            freqs_cis:
+
+        Returns:
+
+        """
+        bsz, seqlen, _ = x.shape
+
+        xq, xk, xv = torch.split(
+            self.qkv(x),
+            [
+                self.n_local_heads * self.head_dim,
+                self.n_local_kv_heads * self.head_dim,
+                self.n_local_kv_heads * self.head_dim,
+            ],
+            dim=-1,
+        )
+        xq = xq.view(bsz, seqlen, self.n_local_heads, self.head_dim)
+        xk = xk.view(bsz, seqlen, self.n_local_kv_heads, self.head_dim)
+        xv = xv.view(bsz, seqlen, self.n_local_kv_heads, self.head_dim)
+
+        xq = self.q_norm(xq)
+        xk = self.k_norm(xk)
+
+        xq = JointAttention.apply_rotary_emb(xq, freqs_cis=freqs_cis)
+        xk = JointAttention.apply_rotary_emb(xk, freqs_cis=freqs_cis)
+
+        n_rep = self.n_local_heads // self.n_local_kv_heads
+        if n_rep >= 1:
+            xk = xk.unsqueeze(3).repeat(1, 1, 1, n_rep, 1).flatten(2, 3)
+            xv = xv.unsqueeze(3).repeat(1, 1, 1, n_rep, 1).flatten(2, 3)
+        output = optimized_attention_masked(xq.movedim(1, 2), xk.movedim(1, 2), xv.movedim(1, 2), self.n_local_heads, x_mask, skip_reshape=True)
+
+        return self.out(output)
+
+
+class FeedForward(nn.Module):
+    def __init__(
+        self,
+        dim: int,
+        hidden_dim: int,
+        multiple_of: int,
+        ffn_dim_multiplier: Optional[float],
+        operation_settings={},
+    ):
+        """
+        Initialize the FeedForward module.
+
+        Args:
+            dim (int): Input dimension.
+            hidden_dim (int): Hidden dimension of the feedforward layer.
+            multiple_of (int): Value to ensure hidden dimension is a multiple
+                of this value.
+            ffn_dim_multiplier (float, optional): Custom multiplier for hidden
+                dimension. Defaults to None.
+
+        """
+        super().__init__()
+        # custom dim factor multiplier
+        if ffn_dim_multiplier is not None:
+            hidden_dim = int(ffn_dim_multiplier * hidden_dim)
+        hidden_dim = multiple_of * ((hidden_dim + multiple_of - 1) // multiple_of)
+
+        self.w1 = operation_settings.get("operations").Linear(
+            dim,
+            hidden_dim,
+            bias=False,
+            device=operation_settings.get("device"),
+            dtype=operation_settings.get("dtype"),
+        )
+        self.w2 = operation_settings.get("operations").Linear(
+            hidden_dim,
+            dim,
+            bias=False,
+            device=operation_settings.get("device"),
+            dtype=operation_settings.get("dtype"),
+        )
+        self.w3 = operation_settings.get("operations").Linear(
+            dim,
+            hidden_dim,
+            bias=False,
+            device=operation_settings.get("device"),
+            dtype=operation_settings.get("dtype"),
+        )
+
+    # @torch.compile
+    def _forward_silu_gating(self, x1, x3):
+        return F.silu(x1) * x3
+
+    def forward(self, x):
+        return self.w2(self._forward_silu_gating(self.w1(x), self.w3(x)))
+
+
+class JointTransformerBlock(nn.Module):
+    def __init__(
+        self,
+        layer_id: int,
+        dim: int,
+        n_heads: int,
+        n_kv_heads: int,
+        multiple_of: int,
+        ffn_dim_multiplier: float,
+        norm_eps: float,
+        qk_norm: bool,
+        modulation=True,
+        operation_settings={},
+    ) -> None:
+        """
+        Initialize a TransformerBlock.
+
+        Args:
+            layer_id (int): Identifier for the layer.
+            dim (int): Embedding dimension of the input features.
+            n_heads (int): Number of attention heads.
+            n_kv_heads (Optional[int]): Number of attention heads in key and
+                value features (if using GQA), or set to None for the same as
+                query.
+            multiple_of (int):
+            ffn_dim_multiplier (float):
+            norm_eps (float):
+
+        """
+        super().__init__()
+        self.dim = dim
+        self.head_dim = dim // n_heads
+        self.attention = JointAttention(dim, n_heads, n_kv_heads, qk_norm, operation_settings=operation_settings)
+        self.feed_forward = FeedForward(
+            dim=dim,
+            hidden_dim=4 * dim,
+            multiple_of=multiple_of,
+            ffn_dim_multiplier=ffn_dim_multiplier,
+            operation_settings=operation_settings,
+        )
+        self.layer_id = layer_id
+        self.attention_norm1 = RMSNorm(dim, eps=norm_eps, elementwise_affine=True, **operation_settings)
+        self.ffn_norm1 = RMSNorm(dim, eps=norm_eps, elementwise_affine=True, **operation_settings)
+
+        self.attention_norm2 = RMSNorm(dim, eps=norm_eps, elementwise_affine=True, **operation_settings)
+        self.ffn_norm2 = RMSNorm(dim, eps=norm_eps, elementwise_affine=True, **operation_settings)
+
+        self.modulation = modulation
+        if modulation:
+            self.adaLN_modulation = nn.Sequential(
+                nn.SiLU(),
+                operation_settings.get("operations").Linear(
+                    min(dim, 1024),
+                    4 * dim,
+                    bias=True,
+                    device=operation_settings.get("device"),
+                    dtype=operation_settings.get("dtype"),
+                ),
+            )
+
+    def forward(
+        self,
+        x: torch.Tensor,
+        x_mask: torch.Tensor,
+        freqs_cis: torch.Tensor,
+        adaln_input: Optional[torch.Tensor]=None,
+    ):
+        """
+        Perform a forward pass through the TransformerBlock.
+
+        Args:
+            x (torch.Tensor): Input tensor.
+            freqs_cis (torch.Tensor): Precomputed cosine and sine frequencies.
+
+        Returns:
+            torch.Tensor: Output tensor after applying attention and
+                feedforward layers.
+
+        """
+        if self.modulation:
+            assert adaln_input is not None
+            scale_msa, gate_msa, scale_mlp, gate_mlp = self.adaLN_modulation(adaln_input).chunk(4, dim=1)
+
+            x = x + gate_msa.unsqueeze(1).tanh() * self.attention_norm2(
+                self.attention(
+                    modulate(self.attention_norm1(x), scale_msa),
+                    x_mask,
+                    freqs_cis,
+                )
+            )
+            x = x + gate_mlp.unsqueeze(1).tanh() * self.ffn_norm2(
+                self.feed_forward(
+                    modulate(self.ffn_norm1(x), scale_mlp),
+                )
+            )
+        else:
+            assert adaln_input is None
+            x = x + self.attention_norm2(
+                self.attention(
+                    self.attention_norm1(x),
+                    x_mask,
+                    freqs_cis,
+                )
+            )
+            x = x + self.ffn_norm2(
+                self.feed_forward(
+                    self.ffn_norm1(x),
+                )
+            )
+        return x
+
+
+class FinalLayer(nn.Module):
+    """
+    The final layer of NextDiT.
+    """
+
+    def __init__(self, hidden_size, patch_size, out_channels, operation_settings={}):
+        super().__init__()
+        self.norm_final = operation_settings.get("operations").LayerNorm(
+            hidden_size,
+            elementwise_affine=False,
+            eps=1e-6,
+            device=operation_settings.get("device"),
+            dtype=operation_settings.get("dtype"),
+        )
+        self.linear = operation_settings.get("operations").Linear(
+            hidden_size,
+            patch_size * patch_size * out_channels,
+            bias=True,
+            device=operation_settings.get("device"),
+            dtype=operation_settings.get("dtype"),
+        )
+
+        self.adaLN_modulation = nn.Sequential(
+            nn.SiLU(),
+            operation_settings.get("operations").Linear(
+                min(hidden_size, 1024),
+                hidden_size,
+                bias=True,
+                device=operation_settings.get("device"),
+                dtype=operation_settings.get("dtype"),
+            ),
+        )
+
+    def forward(self, x, c):
+        scale = self.adaLN_modulation(c)
+        x = modulate(self.norm_final(x), scale)
+        x = self.linear(x)
+        return x
+
+
+class NextDiT(nn.Module):
+    """
+    Diffusion model with a Transformer backbone.
+    """
+
+    def __init__(
+        self,
+        patch_size: int = 2,
+        in_channels: int = 4,
+        dim: int = 4096,
+        n_layers: int = 32,
+        n_refiner_layers: int = 2,
+        n_heads: int = 32,
+        n_kv_heads: Optional[int] = None,
+        multiple_of: int = 256,
+        ffn_dim_multiplier: Optional[float] = None,
+        norm_eps: float = 1e-5,
+        qk_norm: bool = False,
+        cap_feat_dim: int = 5120,
+        axes_dims: List[int] = (16, 56, 56),
+        axes_lens: List[int] = (1, 512, 512),
+        image_model=None,
+        device=None,
+        dtype=None,
+        operations=None,
+    ) -> None:
+        super().__init__()
+        self.dtype = dtype
+        operation_settings = {"operations": operations, "device": device, "dtype": dtype}
+        self.in_channels = in_channels
+        self.out_channels = in_channels
+        self.patch_size = patch_size
+
+        self.x_embedder = operation_settings.get("operations").Linear(
+            in_features=patch_size * patch_size * in_channels,
+            out_features=dim,
+            bias=True,
+            device=operation_settings.get("device"),
+            dtype=operation_settings.get("dtype"),
+        )
+
+        self.noise_refiner = nn.ModuleList(
+            [
+                JointTransformerBlock(
+                    layer_id,
+                    dim,
+                    n_heads,
+                    n_kv_heads,
+                    multiple_of,
+                    ffn_dim_multiplier,
+                    norm_eps,
+                    qk_norm,
+                    modulation=True,
+                    operation_settings=operation_settings,
+                )
+                for layer_id in range(n_refiner_layers)
+            ]
+        )
+        self.context_refiner = nn.ModuleList(
+            [
+                JointTransformerBlock(
+                    layer_id,
+                    dim,
+                    n_heads,
+                    n_kv_heads,
+                    multiple_of,
+                    ffn_dim_multiplier,
+                    norm_eps,
+                    qk_norm,
+                    modulation=False,
+                    operation_settings=operation_settings,
+                )
+                for layer_id in range(n_refiner_layers)
+            ]
+        )
+
+        self.t_embedder = TimestepEmbedder(min(dim, 1024), **operation_settings)
+        self.cap_embedder = nn.Sequential(
+            RMSNorm(cap_feat_dim, eps=norm_eps, elementwise_affine=True, **operation_settings),
+            operation_settings.get("operations").Linear(
+                cap_feat_dim,
+                dim,
+                bias=True,
+                device=operation_settings.get("device"),
+                dtype=operation_settings.get("dtype"),
+            ),
+        )
+
+        self.layers = nn.ModuleList(
+            [
+                JointTransformerBlock(
+                    layer_id,
+                    dim,
+                    n_heads,
+                    n_kv_heads,
+                    multiple_of,
+                    ffn_dim_multiplier,
+                    norm_eps,
+                    qk_norm,
+                    operation_settings=operation_settings,
+                )
+                for layer_id in range(n_layers)
+            ]
+        )
+        self.norm_final = RMSNorm(dim, eps=norm_eps, elementwise_affine=True, **operation_settings)
+        self.final_layer = FinalLayer(dim, patch_size, self.out_channels, operation_settings=operation_settings)
+
+        assert (dim // n_heads) == sum(axes_dims)
+        self.axes_dims = axes_dims
+        self.axes_lens = axes_lens
+        self.rope_embedder = EmbedND(dim=dim // n_heads, theta=10000.0, axes_dim=axes_dims)
+        self.dim = dim
+        self.n_heads = n_heads
+
+    def unpatchify(
+        self, x: torch.Tensor, img_size: List[Tuple[int, int]], cap_size: List[int], return_tensor=False
+    ) -> List[torch.Tensor]:
+        """
+        x: (N, T, patch_size**2 * C)
+        imgs: (N, H, W, C)
+        """
+        pH = pW = self.patch_size
+        imgs = []
+        for i in range(x.size(0)):
+            H, W = img_size[i]
+            begin = cap_size[i]
+            end = begin + (H // pH) * (W // pW)
+            imgs.append(
+                x[i][begin:end]
+                .view(H // pH, W // pW, pH, pW, self.out_channels)
+                .permute(4, 0, 2, 1, 3)
+                .flatten(3, 4)
+                .flatten(1, 2)
+            )
+
+        if return_tensor:
+            imgs = torch.stack(imgs, dim=0)
+        return imgs
+
+    def patchify_and_embed(
+        self, x: List[torch.Tensor] | torch.Tensor, cap_feats: torch.Tensor, cap_mask: torch.Tensor, t: torch.Tensor, num_tokens
+    ) -> Tuple[torch.Tensor, torch.Tensor, List[Tuple[int, int]], List[int], torch.Tensor]:
+        bsz = len(x)
+        pH = pW = self.patch_size
+        device = x[0].device
+        dtype = x[0].dtype
+
+        if cap_mask is not None:
+            l_effective_cap_len = cap_mask.sum(dim=1).tolist()
+        else:
+            l_effective_cap_len = [num_tokens] * bsz
+
+        if cap_mask is not None and not torch.is_floating_point(cap_mask):
+            cap_mask = (cap_mask - 1).to(dtype) * torch.finfo(dtype).max
+
+        img_sizes = [(img.size(1), img.size(2)) for img in x]
+        l_effective_img_len = [(H // pH) * (W // pW) for (H, W) in img_sizes]
+
+        max_seq_len = max(
+            (cap_len+img_len for cap_len, img_len in zip(l_effective_cap_len, l_effective_img_len))
+        )
+        max_cap_len = max(l_effective_cap_len)
+        max_img_len = max(l_effective_img_len)
+
+        position_ids = torch.zeros(bsz, max_seq_len, 3, dtype=torch.int32, device=device)
+
+        for i in range(bsz):
+            cap_len = l_effective_cap_len[i]
+            img_len = l_effective_img_len[i]
+            H, W = img_sizes[i]
+            H_tokens, W_tokens = H // pH, W // pW
+            assert H_tokens * W_tokens == img_len
+
+            position_ids[i, :cap_len, 0] = torch.arange(cap_len, dtype=torch.int32, device=device)
+            position_ids[i, cap_len:cap_len+img_len, 0] = cap_len
+            row_ids = torch.arange(H_tokens, dtype=torch.int32, device=device).view(-1, 1).repeat(1, W_tokens).flatten()
+            col_ids = torch.arange(W_tokens, dtype=torch.int32, device=device).view(1, -1).repeat(H_tokens, 1).flatten()
+            position_ids[i, cap_len:cap_len+img_len, 1] = row_ids
+            position_ids[i, cap_len:cap_len+img_len, 2] = col_ids
+
+        freqs_cis = self.rope_embedder(position_ids).movedim(1, 2).to(dtype)
+
+        # build freqs_cis for cap and image individually
+        cap_freqs_cis_shape = list(freqs_cis.shape)
+        # cap_freqs_cis_shape[1] = max_cap_len
+        cap_freqs_cis_shape[1] = cap_feats.shape[1]
+        cap_freqs_cis = torch.zeros(*cap_freqs_cis_shape, device=device, dtype=freqs_cis.dtype)
+
+        img_freqs_cis_shape = list(freqs_cis.shape)
+        img_freqs_cis_shape[1] = max_img_len
+        img_freqs_cis = torch.zeros(*img_freqs_cis_shape, device=device, dtype=freqs_cis.dtype)
+
+        for i in range(bsz):
+            cap_len = l_effective_cap_len[i]
+            img_len = l_effective_img_len[i]
+            cap_freqs_cis[i, :cap_len] = freqs_cis[i, :cap_len]
+            img_freqs_cis[i, :img_len] = freqs_cis[i, cap_len:cap_len+img_len]
+
+        # refine context
+        for layer in self.context_refiner:
+            cap_feats = layer(cap_feats, cap_mask, cap_freqs_cis)
+
+        # refine image
+        flat_x = []
+        for i in range(bsz):
+            img = x[i]
+            C, H, W = img.size()
+            img = img.view(C, H // pH, pH, W // pW, pW).permute(1, 3, 2, 4, 0).flatten(2).flatten(0, 1)
+            flat_x.append(img)
+        x = flat_x
+        padded_img_embed = torch.zeros(bsz, max_img_len, x[0].shape[-1], device=device, dtype=x[0].dtype)
+        padded_img_mask = torch.zeros(bsz, max_img_len, dtype=dtype, device=device)
+        for i in range(bsz):
+            padded_img_embed[i, :l_effective_img_len[i]] = x[i]
+            padded_img_mask[i, l_effective_img_len[i]:] = -torch.finfo(dtype).max
+
+        padded_img_embed = self.x_embedder(padded_img_embed)
+        padded_img_mask = padded_img_mask.unsqueeze(1)
+        for layer in self.noise_refiner:
+            padded_img_embed = layer(padded_img_embed, padded_img_mask, img_freqs_cis, t)
+
+        if cap_mask is not None:
+            mask = torch.zeros(bsz, max_seq_len, dtype=dtype, device=device)
+            mask[:, :max_cap_len] = cap_mask[:, :max_cap_len]
+        else:
+            mask = None
+
+        padded_full_embed = torch.zeros(bsz, max_seq_len, self.dim, device=device, dtype=x[0].dtype)
+        for i in range(bsz):
+            cap_len = l_effective_cap_len[i]
+            img_len = l_effective_img_len[i]
+
+            padded_full_embed[i, :cap_len] = cap_feats[i, :cap_len]
+            padded_full_embed[i, cap_len:cap_len+img_len] = padded_img_embed[i, :img_len]
+
+        return padded_full_embed, mask, img_sizes, l_effective_cap_len, freqs_cis
+
+    # def forward(self, x, t, cap_feats, cap_mask):
+    def forward(self, x, timesteps, context, num_tokens, attention_mask=None, **kwargs):
+        t = 1.0 - timesteps
+        cap_feats = context
+        cap_mask = attention_mask
+        bs, c, h, w = x.shape
+        x = comfy.ldm.common_dit.pad_to_patch_size(x, (self.patch_size, self.patch_size))
+        """
+        Forward pass of NextDiT.
+        t: (N,) tensor of diffusion timesteps
+        y: (N,) tensor of text tokens/features
+        """
+
+        t = self.t_embedder(t, dtype=x.dtype)  # (N, D)
+        adaln_input = t
+
+        cap_feats = self.cap_embedder(cap_feats)  # (N, L, D)  # todo check if able to batchify w.o. redundant compute
+
+        x_is_tensor = isinstance(x, torch.Tensor)
+        x, mask, img_size, cap_size, freqs_cis = self.patchify_and_embed(x, cap_feats, cap_mask, t, num_tokens)
+        freqs_cis = freqs_cis.to(x.device)
+
+        for layer in self.layers:
+            x = layer(x, mask, freqs_cis, adaln_input)
+
+        x = self.final_layer(x, adaln_input)
+        x = self.unpatchify(x, img_size, cap_size, return_tensor=x_is_tensor)[:,:,:h,:w]
+
+        return -x
+

comfy/ldm/modules/attention.py

@@ -30,11 +30,12 @@ FORCE_UPCAST_ATTENTION_DTYPE = model_management.force_upcast_attention_dtype()
 logger = logging.getLogger(__name__)
 
 
-def get_attn_precision(attn_precision):
+def get_attn_precision(attn_precision, current_dtype):
     if args.dont_upcast_attention:
         return None
-    if FORCE_UPCAST_ATTENTION_DTYPE is not None:
-        return FORCE_UPCAST_ATTENTION_DTYPE
+
+    if FORCE_UPCAST_ATTENTION_DTYPE is not None and current_dtype in FORCE_UPCAST_ATTENTION_DTYPE:
+        return FORCE_UPCAST_ATTENTION_DTYPE[current_dtype]
     return attn_precision
 
 
@@ -52,17 +53,6 @@ def default(val, d):
     return d
 
 
-def max_neg_value(t):
-    return -torch.finfo(t.dtype).max
-
-
-def init_(tensor):
-    dim = tensor.shape[-1]
-    std = 1 / math.sqrt(dim)
-    tensor.uniform_(-std, std)
-    return tensor
-
-
 # feedforward
 class GEGLU(nn.Module):
     def __init__(self, dim_in, dim_out, dtype=None, device=None, operations=ops):
@@ -99,7 +89,7 @@ def Normalize(in_channels, dtype=None, device=None):
 
 
 def attention_basic(q, k, v, heads, mask=None, attn_precision=None, skip_reshape=False, skip_output_reshape=False):
-    attn_precision = get_attn_precision(attn_precision)
+    attn_precision = get_attn_precision(attn_precision, q.dtype)
 
     if skip_reshape:
         b, _, _, dim_head = q.shape
@@ -168,7 +158,7 @@ def attention_basic(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
 
 
 def attention_sub_quad(query, key, value, heads, mask=None, attn_precision=None, skip_reshape=False, skip_output_reshape=False):
-    attn_precision = get_attn_precision(attn_precision)
+    attn_precision = get_attn_precision(attn_precision, query.dtype)
 
     if skip_reshape:
         b, _, _, dim_head = query.shape
@@ -238,7 +228,7 @@ def attention_sub_quad(query, key, value, heads, mask=None, attn_precision=None,
 
 
 def attention_split(q, k, v, heads, mask=None, attn_precision=None, skip_reshape=False, skip_output_reshape=False):
-    attn_precision = get_attn_precision(attn_precision)
+    attn_precision = get_attn_precision(attn_precision, q.dtype)
 
     if skip_reshape:
         b, _, _, dim_head = q.shape
@@ -430,6 +420,7 @@ def pytorch_style_decl(func):
     :param func:
     :return:
     """
+
     @wraps(func)
     def wrapper(q, k, v, heads, mask=None, attn_precision=None, skip_reshape=False):
         if skip_reshape:
@@ -487,12 +478,12 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha
             m = mask
             if mask is not None:
                 if mask.shape[0] > 1:
-                    m = mask[i : i + SDP_BATCH_LIMIT]
+                    m = mask[i: i + SDP_BATCH_LIMIT]
 
-            out[i : i + SDP_BATCH_LIMIT] = torch.nn.functional.scaled_dot_product_attention(
-                q[i : i + SDP_BATCH_LIMIT],
-                k[i : i + SDP_BATCH_LIMIT],
-                v[i : i + SDP_BATCH_LIMIT],
+            out[i: i + SDP_BATCH_LIMIT] = torch.nn.functional.scaled_dot_product_attention(
+                q[i: i + SDP_BATCH_LIMIT],
+                k[i: i + SDP_BATCH_LIMIT],
+                v[i: i + SDP_BATCH_LIMIT],
                 attn_mask=m,
                 dropout_p=0.0, is_causal=False
             ).transpose(1, 2).reshape(-1, q.shape[2], heads * dim_head)
@@ -502,7 +493,7 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha
 def attention_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape=False, skip_output_reshape=False):
     if skip_reshape:
         b, _, _, dim_head = q.shape
-        tensor_layout="HND"
+        tensor_layout = "HND"
     else:
         b, _, dim_head = q.shape
         dim_head //= heads
@@ -510,7 +501,7 @@ def attention_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape=
             lambda t: t.view(b, -1, heads, dim_head),
             (q, k, v),
         )
-        tensor_layout="NHD"
+        tensor_layout = "NHD"
 
     if mask is not None:
         # add a batch dimension if there isn't already one

comfy/ldm/modules/diffusionmodules/mmdit.py

@@ -323,7 +323,7 @@ class SelfAttention(nn.Module):
 
 class RMSNorm(torch.nn.Module):
     def __init__(
-        self, dim: int, elementwise_affine: bool = False, eps: float = 1e-6, device=None, dtype=None
+        self, dim: int, elementwise_affine: bool = False, eps: float = 1e-6, device=None, dtype=None, **kwargs
     ):
         """
         Initialize the RMSNorm normalization layer.

comfy/ldm/modules/diffusionmodules/model.py

@@ -305,7 +305,7 @@ def vae_attention():
     if model_management.xformers_enabled_vae():
         logging.debug("Using xformers attention in VAE")
         return xformers_attention
-    elif model_management.pytorch_attention_enabled():
+    elif model_management.pytorch_attention_enabled_vae():
         logging.debug("Using pytorch attention in VAE")
         return pytorch_attention
     else:

comfy/lora.py

@@ -309,7 +309,7 @@ def model_lora_keys_unet(model, key_map=None):
             if k.endswith(".weight"):
                 key_lora = k[len("diffusion_model."):-len(".weight")].replace(".", "_")
                 key_map["lora_unet_{}".format(key_lora)] = k
-                key_map["lora_prior_unet_{}".format(key_lora)] = k  # cascade lora: TODO put lora key prefix in the model config
+
                 key_map["{}".format(k[:-len(".weight")])] = k  # generic lora format without any weird key names
             else:
                 key_map["{}".format(k)] = k  # generic lora format for not .weight without any weird key names
@@ -329,6 +329,13 @@ def model_lora_keys_unet(model, key_map=None):
                     diffusers_lora_key = diffusers_lora_key[:-2]
                 key_map[diffusers_lora_key] = unet_key
 
+    if isinstance(model, model_base.StableCascade_C):
+        for k in sdk:
+            if k.startswith("diffusion_model."):
+                if k.endswith(".weight"):
+                    key_lora = k[len("diffusion_model."):-len(".weight")].replace(".", "_")
+                    key_map["lora_prior_unet_{}".format(key_lora)] = k
+
     if isinstance(model, model_base.SD3):  # Diffusers lora SD3
         diffusers_keys = utils.mmdit_to_diffusers(model.model_config.unet_config, output_prefix="diffusion_model.")
         for k in diffusers_keys:

comfy/model_base.py

@@ -39,6 +39,7 @@ from .ldm.genmo.joint_model.asymm_models_joint import AsymmDiTJoint
 from .ldm.hunyuan_video.model import HunyuanVideo as HunyuanVideoModel
 from .ldm.hydit.models import HunYuanDiT
 from .ldm.lightricks.model import LTXVModel
+from .ldm.lumina.model import NextDiT
 from .ldm.modules.diffusionmodules.mmdit import OpenAISignatureMMDITWrapper
 from .ldm.modules.diffusionmodules.openaimodel import UNetModel, Timestep
 from .ldm.modules.diffusionmodules.upscaling import ImageConcatWithNoiseAugmentation
@@ -181,9 +182,6 @@ class BaseModel(torch.nn.Module):
     def get_dtype(self):
         return self.diffusion_model.dtype
 
-    def is_adm(self):
-        return self.adm_channels > 0
-
     def encode_adm(self, **kwargs):
         return None
 
@@ -908,6 +906,15 @@ class HunyuanVideo(BaseModel):
         if cross_attn is not None:
             out['c_crossattn'] = conds.CONDRegular(cross_attn)
 
+        image = kwargs.get("concat_latent_image", None)
+        noise = kwargs.get("noise", None)
+
+        if image is not None:
+            padding_shape = (noise.shape[0], 16, noise.shape[2] - 1, noise.shape[3], noise.shape[4])
+            latent_padding = torch.zeros(padding_shape, device=noise.device, dtype=noise.dtype)
+            image_latents = torch.cat([image.to(noise), latent_padding], dim=2)
+            out['c_concat'] = conds.CONDNoiseShape(self.process_latent_in(image_latents))
+
         guidance = kwargs.get("guidance", 6.0)
         if guidance is not None:
             out['guidance'] = conds.CONDRegular(torch.FloatTensor([guidance]))
@@ -940,3 +947,19 @@ class CosmosVideo(BaseModel):
             latent_image = latent_image + noise
         latent_image = self.model_sampling.calculate_input(torch.tensor([sigma_noise_augmentation], device=latent_image.device, dtype=latent_image.dtype), latent_image)
         return latent_image * ((sigma ** 2 + self.model_sampling.sigma_data ** 2) ** 0.5)
+
+class Lumina2(BaseModel):
+    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
+        super().__init__(model_config, model_type, device=device, unet_model=NextDiT)
+
+    def extra_conds(self, **kwargs):
+        out = super().extra_conds(**kwargs)
+        attention_mask = kwargs.get("attention_mask", None)
+        if attention_mask is not None:
+            if torch.numel(attention_mask) != attention_mask.sum():
+                out['attention_mask'] = conds.CONDRegular(attention_mask)
+            out['num_tokens'] = conds.CONDConstant(max(1, torch.sum(attention_mask).item()))
+        cross_attn = kwargs.get("cross_attn", None)
+        if cross_attn is not None:
+            out['c_crossattn'] = conds.CONDRegular(cross_attn)
+        return out

comfy/model_detection.py

@@ -140,7 +140,7 @@ def detect_unet_config(state_dict, key_prefix):
     if '{}txt_in.individual_token_refiner.blocks.0.norm1.weight'.format(key_prefix) in state_dict_keys: #Hunyuan Video
         dit_config = {}
         dit_config["image_model"] = "hunyuan_video"
-        dit_config["in_channels"] = 16
+        dit_config["in_channels"] = state_dict['{}img_in.proj.weight'.format(key_prefix)].shape[1] #SkyReels img2video has 32 input channels
         dit_config["patch_size"] = [1, 2, 2]
         dit_config["out_channels"] = 16
         dit_config["vec_in_dim"] = 768
@@ -243,7 +243,7 @@ def detect_unet_config(state_dict, key_prefix):
             dit_config["micro_condition"] = False
         return dit_config
 
-    if '{}blocks.block0.blocks.0.block.attn.to_q.0.weight'.format(key_prefix) in state_dict_keys:
+    if '{}blocks.block0.blocks.0.block.attn.to_q.0.weight'.format(key_prefix) in state_dict_keys:  # Cosmos
         dit_config = {}
         dit_config["image_model"] = "cosmos"
         dit_config["max_img_h"] = 240
@@ -288,6 +288,21 @@ def detect_unet_config(state_dict, key_prefix):
             dit_config["extra_per_block_abs_pos_emb_type"] = "learnable"
         return dit_config
 
+    if '{}cap_embedder.1.weight'.format(key_prefix) in state_dict_keys:  # Lumina 2
+        dit_config = {}
+        dit_config["image_model"] = "lumina2"
+        dit_config["patch_size"] = 2
+        dit_config["in_channels"] = 16
+        dit_config["dim"] = 2304
+        dit_config["cap_feat_dim"] = 2304
+        dit_config["n_layers"] = 26
+        dit_config["n_heads"] = 24
+        dit_config["n_kv_heads"] = 8
+        dit_config["qk_norm"] = True
+        dit_config["axes_dims"] = [32, 32, 32]
+        dit_config["axes_lens"] = [300, 512, 512]
+        return dit_config
+
     if '{}input_blocks.0.0.weight'.format(key_prefix) not in state_dict_keys:
         return None
 

comfy/model_management.py

@@ -72,7 +72,9 @@ xpu_available = False
 torch_version = ""
 try:
     torch_version = torch.version.__version__
-    xpu_available = (int(torch_version[0]) < 2 or (int(torch_version[0]) == 2 and int(torch_version[2]) <= 4)) and torch.xpu.is_available()
+    temp = torch_version.split(".")
+    torch_version_numeric = (int(temp[0]), int(temp[1]))
+    xpu_available = (torch_version_numeric[0] < 2 or (torch_version_numeric[0] == 2 and torch_version_numeric[1] <= 4)) and torch.xpu.is_available()
 except:
     pass
 
@@ -259,7 +261,7 @@ def is_amd():
 
 MIN_WEIGHT_MEMORY_RATIO = 0.4
 if is_nvidia():
-    MIN_WEIGHT_MEMORY_RATIO = 0.1
+    MIN_WEIGHT_MEMORY_RATIO = 0.0
 
 ENABLE_PYTORCH_ATTENTION = False
 if args.use_pytorch_cross_attention:
@@ -268,7 +270,7 @@ if args.use_pytorch_cross_attention:
 
 try:
     if is_nvidia() or is_amd():
-        if int(torch_version[0]) >= 2:
+        if torch_version_numeric[0] >= 2:
             if ENABLE_PYTORCH_ATTENTION == False and args.use_split_cross_attention == False and args.use_quad_cross_attention == False:
                 ENABLE_PYTORCH_ATTENTION = True
     if is_intel_xpu() or is_ascend_npu():
@@ -277,13 +279,32 @@ try:
 except:
     pass
 
+try:
+    if is_amd():
+        arch = torch.cuda.get_device_properties(get_torch_device()).gcnArchName
+        logging.info("AMD arch: {}".format(arch))
+        if args.use_split_cross_attention == False and args.use_quad_cross_attention == False:
+            if torch_version_numeric[0] >= 2 and torch_version_numeric[1] >= 7:  # works on 2.6 but doesn't actually seem to improve much
+                if any((a in arch) for a in ["gfx1100", "gfx1101"]):  # TODO: more arches
+                    ENABLE_PYTORCH_ATTENTION = True
+except:
+    pass
+
 if ENABLE_PYTORCH_ATTENTION:
     torch.backends.cuda.enable_math_sdp(True)
     torch.backends.cuda.enable_flash_sdp(True)
     torch.backends.cuda.enable_mem_efficient_sdp(True)
 
+PRIORITIZE_FP16 = False  # TODO: remove and replace with something that shows exactly which dtype is faster than the other
 try:
-    if int(torch_version[0]) == 2 and int(torch_version[2]) >= 5:
+    if is_nvidia() and args.fast:
+        torch.backends.cuda.matmul.allow_fp16_accumulation = True
+        PRIORITIZE_FP16 = True  # TODO: limit to cards where it actually boosts performance
+except:
+    pass
+
+try:
+    if torch_version_numeric[0] == 2 and torch_version_numeric[1] >= 5:
         torch.backends.cuda.allow_fp16_bf16_reduction_math_sdp(True)  # pylint: disable=no-member
 except:
     logging.warning("Warning, could not set allow_fp16_bf16_reduction_math_sdp")
@@ -629,7 +650,6 @@ def _load_models_gpu(models: Sequence[ModelManageable], memory_required: int = 0
         current_loaded_models.insert(0, loaded_model)
         logger.debug(f"Loaded {loaded_model}")
 
-
     span = get_current_span()
     span.set_attribute("models_to_load", list(map(str, models_to_load)))
     span.set_attribute("models_freed", list(map(str, models_freed)))
@@ -759,6 +779,10 @@ def unet_dtype(device=None, model_params=0, supported_dtypes=(torch.float16, tor
         if model_params * 2 > free_model_memory:
             return fp8_dtype
 
+    if PRIORITIZE_FP16:
+        if torch.float16 in supported_dtypes and should_use_fp16(device=device, model_params=model_params):
+            return torch.float16
+
     for dt in supported_dtypes:
         if dt == torch.float16 and should_use_fp16(device=device, model_params=model_params):
             if torch.float16 in supported_dtypes:
@@ -1037,6 +1061,12 @@ def pytorch_attention_enabled():
     return ENABLE_PYTORCH_ATTENTION
 
 
+def pytorch_attention_enabled_vae():
+    if is_amd():
+        return False  # enabling pytorch attention on AMD currently causes crash when doing high res
+    return pytorch_attention_enabled()
+
+
 def pytorch_attention_flash_attention():
     global ENABLE_PYTORCH_ATTENTION
     if ENABLE_PYTORCH_ATTENTION:
@@ -1047,6 +1077,8 @@ def pytorch_attention_flash_attention():
             return True
         if is_ascend_npu():
             return True
+        if is_amd():
+            return True  # if you have pytorch attention enabled on AMD it probably supports at least mem efficient attention
     return False
 
 
@@ -1061,11 +1093,11 @@ def force_upcast_attention_dtype():
     upcast = args.force_upcast_attention
 
     macos_version = mac_version()
-    if macos_version is not None and ((14, 5) <= macos_version <= (15, 2)):  # black image bug on recent versions of macOS
+    if macos_version is not None and ((14, 5) <= macos_version < (16,)):  # black image bug on recent versions of macOS
         upcast = True
 
     if upcast:
-        return torch.float32
+        return {torch.float16: torch.float32}
     else:
         return None
 
@@ -1139,21 +1171,27 @@ def is_device_cuda(device):
     return is_device_type(device, 'cuda')
 
 
-def should_use_fp16(device=None, model_params=0, prioritize_performance=True, manual_cast=False):
-    global directml_device
+def is_directml_enabled():
+    global directml_enabled
+    if directml_enabled:
+        return True
 
+    return False
+
+
+def should_use_fp16(device=None, model_params=0, prioritize_performance=True, manual_cast=False):
     if device is not None:
         if is_device_cpu(device):
             return False
 
-    if FORCE_FP16:
+    if args.force_fp16:
         return True
 
     if FORCE_FP32:
         return False
 
-    if directml_device:
-        return False
+    if is_directml_enabled():
+        return True
 
     if (device is not None and is_device_mps(device)) or mps_mode():
         return True
@@ -1234,6 +1272,16 @@ def should_use_bf16(device=None, model_params=0, prioritize_performance=True, ma
     if is_intel_xpu():
         return True
 
+    if is_ascend_npu():
+        return True
+
+    if is_amd():
+        arch = torch.cuda.get_device_properties(device).gcnArchName
+        if any((a in arch) for a in ["gfx1030", "gfx1031", "gfx1010", "gfx1011", "gfx1012", "gfx906", "gfx900", "gfx803"]):  # RDNA2 and older don't support bf16
+            if manual_cast:
+                return True
+            return False
+
     try:
         props_major = min(torch.cuda.get_device_properties(torch.device(f"cuda:{i}")).major for i in range(torch.cuda.device_count()))
         if props_major >= 8:
@@ -1247,7 +1295,7 @@ def should_use_bf16(device=None, model_params=0, prioritize_performance=True, ma
 
     bf16_works = torch.cuda.is_bf16_supported()
 
-    if bf16_works or manual_cast:
+    if bf16_works and manual_cast:
         free_model_memory = maximum_vram_for_weights(device)
         if (not prioritize_performance) or model_params * 4 > free_model_memory:
             return True
@@ -1271,11 +1319,11 @@ def supports_fp8_compute(device=None):
     if props.minor < 9:
         return False
 
-    if int(torch_version[0]) < 2 or (int(torch_version[0]) == 2 and int(torch_version[2]) < 3):
+    if torch_version_numeric[0] < 2 or (torch_version_numeric[0] == 2 and torch_version_numeric[1] < 3):
         return False
 
     if WINDOWS:
-        if (int(torch_version[0]) == 2 and int(torch_version[2]) < 4):
+        if (torch_version_numeric[0] == 2 and torch_version_numeric[1] < 4):
             return False
 
     return True

comfy/model_management_types.py

@@ -126,6 +126,16 @@ class ModelManageable(Protocol):
             self.unpatch_model(self.offload_device, unpatch_weights=unpatch_all)
         return self.model
 
+    def set_model_compute_dtype(self, dtype: torch.dtype):
+        pass
+
+    def add_weight_wrapper(self, name, function):
+        pass
+
+    @property
+    def force_cast_weights(self) -> bool:
+        return False
+
 
 @dataclasses.dataclass
 class MemoryMeasurements:

comfy/model_patcher.py

@@ -112,8 +112,28 @@ def wipe_lowvram_weight(m):
     if hasattr(m, "prev_comfy_cast_weights"):
         m.comfy_cast_weights = m.prev_comfy_cast_weights
         del m.prev_comfy_cast_weights
-    m.weight_function = None
-    m.bias_function = None
+
+    if hasattr(m, "weight_function"):
+        m.weight_function = []
+
+    if hasattr(m, "bias_function"):
+        m.bias_function = []
+
+def move_weight_functions(m, device):
+    if device is None:
+        return 0
+
+    memory = 0
+    if hasattr(m, "weight_function"):
+        for f in m.weight_function:
+            if hasattr(f, "move_to"):
+                memory += f.move_to(device=device)
+
+    if hasattr(m, "bias_function"):
+        for f in m.bias_function:
+            if hasattr(f, "move_to"):
+                memory += f.move_to(device=device)
+    return memory
 
 
 class LowVramPatch:
@@ -207,11 +227,13 @@ class ModelPatcher(ModelManageable):
         self.backup = {}
         self.object_patches = {}
         self.object_patches_backup = {}
+        self.weight_wrapper_patches = {}
         self._model_options: ModelOptions = {"transformer_options": {}}
         self.model_size()
         self.load_device = load_device
         self.offload_device = offload_device
         self.weight_inplace_update = weight_inplace_update
+        self._force_cast_weights = False
         self._parent: ModelManageable | None = None
         self.patches_uuid: uuid.UUID = uuid.uuid4()
         self.ckpt_name = ckpt_name
@@ -262,6 +284,14 @@ class ModelPatcher(ModelManageable):
     def parent(self) -> Optional["ModelPatcher"]:
         return self._parent
 
+    @property
+    def force_cast_weights(self) -> bool:
+        return self._force_cast_weights
+
+    @force_cast_weights.setter
+    def force_cast_weights(self, value:bool) -> None:
+        self._force_cast_weights = value
+
     def lowvram_patch_counter(self):
         return self._memory_measurements.lowvram_patch_counter
 
@@ -284,11 +314,14 @@ class ModelPatcher(ModelManageable):
         n.patches_uuid = self.patches_uuid
 
         n.object_patches = self.object_patches.copy()
+        n.weight_wrapper_patches = self.weight_wrapper_patches.copy()
         n._model_options = copy.deepcopy(self.model_options)
         n.backup = self.backup
         n.object_patches_backup = self.object_patches_backup
         n._parent = self
 
+        n.force_cast_weights = self.force_cast_weights
+
         # attachments
         n.attachments = {}
         for k in self.attachments:
@@ -435,6 +468,16 @@ class ModelPatcher(ModelManageable):
     def add_object_patch(self, name, obj):
         self.object_patches[name] = obj
 
+    def set_model_compute_dtype(self, dtype):
+        self.add_object_patch("manual_cast_dtype", dtype)
+        if dtype is not None:
+            self.force_cast_weights = True
+        self.patches_uuid = uuid.uuid4() #TODO: optimize by preventing a full model reload for this
+
+    def add_weight_wrapper(self, name, function):
+        self.weight_wrapper_patches[name] = self.weight_wrapper_patches.get(name, []) + [function]
+        self.patches_uuid = uuid.uuid4()
+
     def get_model_object(self, name: str) -> torch.nn.Module | typing.Any:
         """Retrieves a nested attribute from an object using dot notation considering
         object patches.
@@ -617,6 +660,9 @@ class ModelPatcher(ModelManageable):
 
                 lowvram_weight = False
 
+                weight_key = "{}.weight".format(n)
+                bias_key = "{}.bias".format(n)
+
                 if not full_load and hasattr(m, "comfy_cast_weights"):
                     if mem_counter + module_mem >= lowvram_model_memory:
                         lowvram_weight = True
@@ -624,34 +670,46 @@ class ModelPatcher(ModelManageable):
                         if hasattr(m, "prev_comfy_cast_weights"):  # Already lowvramed
                             continue
 
-                weight_key = "{}.weight".format(n)
-                bias_key = "{}.bias".format(n)
-
+                cast_weight = self.force_cast_weights
                 if lowvram_weight:
+                    if hasattr(m, "comfy_cast_weights"):
+                        m.weight_function = []
+                        m.bias_function = []
+
                     if weight_key in self.patches:
                         if force_patch_weights:
                             self.patch_weight_to_device(weight_key)
                         else:
-                            m.weight_function = LowVramPatch(weight_key, self.patches)
+                            m.weight_function = [LowVramPatch(weight_key, self.patches)]
                             patch_counter += 1
                     if bias_key in self.patches:
                         if force_patch_weights:
                             self.patch_weight_to_device(bias_key)
                         else:
-                            m.bias_function = LowVramPatch(bias_key, self.patches)
+                            m.bias_function = [LowVramPatch(bias_key, self.patches)]
                             patch_counter += 1
 
-                    m.prev_comfy_cast_weights = m.comfy_cast_weights
-                    m.comfy_cast_weights = True
+                    cast_weight = True
                 else:
                     if hasattr(m, "comfy_cast_weights"):
-                        if m.comfy_cast_weights:
-                            wipe_lowvram_weight(m)
+                        wipe_lowvram_weight(m)
 
                     if full_load or mem_counter + module_mem < lowvram_model_memory:
                         mem_counter += module_mem
                         load_completely.append((module_mem, n, m, params))
 
+                if cast_weight:
+                    m.prev_comfy_cast_weights = m.comfy_cast_weights
+                    m.comfy_cast_weights = True
+
+                if weight_key in self.weight_wrapper_patches:
+                    m.weight_function.extend(self.weight_wrapper_patches[weight_key])
+
+                if bias_key in self.weight_wrapper_patches:
+                    m.bias_function.extend(self.weight_wrapper_patches[bias_key])
+
+                mem_counter += move_weight_functions(m, device_to)
+
             load_completely.sort(reverse=True)
             for x in load_completely:
                 n = x[1]
@@ -714,6 +772,7 @@ class ModelPatcher(ModelManageable):
             self.unpatch_hooks()
             if self._memory_measurements.model_lowvram:
                 for m in self.model.modules():
+                    move_weight_functions(m, device_to)
                     wipe_lowvram_weight(m)
 
                 self._memory_measurements.model_lowvram = False
@@ -780,15 +839,19 @@ class ModelPatcher(ModelManageable):
                     weight_key = "{}.weight".format(n)
                     bias_key = "{}.bias".format(n)
                     if move_weight:
+                        cast_weight = self.force_cast_weights
                         m.to(device_to)
+                        module_mem += move_weight_functions(m, device_to)
                         if lowvram_possible:
                             if weight_key in self.patches:
-                                m.weight_function = LowVramPatch(weight_key, self.patches)
+                                m.weight_function.append(LowVramPatch(weight_key, self.patches))
                                 patch_counter += 1
                             if bias_key in self.patches:
-                                m.bias_function = LowVramPatch(bias_key, self.patches)
+                                m.bias_function.append(LowVramPatch(bias_key, self.patches))
                                 patch_counter += 1
+                            cast_weight = True
 
+                        if cast_weight:
                             m.prev_comfy_cast_weights = m.comfy_cast_weights
                             m.comfy_cast_weights = True
                         m.comfy_patched_weights = False

comfy/model_sampling.py

@@ -46,6 +46,7 @@ class EPS(ModelSampling):
         return model_input - model_output * sigma
 
     def noise_scaling(self, sigma, noise, latent_image, max_denoise=False):
+        sigma = sigma.view(sigma.shape[:1] + (1,) * (noise.ndim - 1))
         if max_denoise:
             noise = noise * torch.sqrt(1.0 + sigma ** 2.0)
         else:
@@ -79,9 +80,11 @@ class CONST(ModelSampling):
         return model_input - model_output * sigma
 
     def noise_scaling(self, sigma, noise, latent_image, max_denoise=False):
+        sigma = sigma.view(sigma.shape[:1] + (1,) * (noise.ndim - 1))
         return sigma * noise + (1.0 - sigma) * latent_image
 
     def inverse_noise_scaling(self, sigma, latent):
+        sigma = sigma.view(sigma.shape[:1] + (1,) * (latent.ndim - 1))
         return latent / (1.0 - sigma)
 
 

comfy/node_helpers.py

@@ -3,7 +3,6 @@ import hashlib
 from PIL import ImageFile, UnidentifiedImageError
 
 from .cli_args import args
-from .nodes.package_typing import CustomNode
 
 
 def conditioning_set_values(conditioning, values: dict = None):
@@ -50,7 +49,6 @@ def export_custom_nodes():
     Must be called from within the module where the CustomNode classes are defined.
     """
     import inspect
-    from abc import ABC
     from .nodes.package_typing import CustomNode
 
     # Get the calling module
@@ -76,3 +74,13 @@ def export_custom_nodes():
         del frame
 
     return custom_nodes
+
+
+def string_to_torch_dtype(string):
+    import torch
+    if string == "fp32":
+        return torch.float32
+    if string == "fp16":
+        return torch.float16
+    if string == "bf16":
+        return torch.bfloat16

comfy/nodes/base_nodes.py

@@ -946,7 +946,7 @@ class CLIPLoader:
     @classmethod
     def INPUT_TYPES(s):
         return {"required": { "clip_name": (get_filename_list_with_downloadable("text_encoders", KNOWN_CLIP_MODELS),),
-                              "type": (["stable_diffusion", "stable_cascade", "sd3", "stable_audio", "mochi", "ltxv", "pixart", "cosmos"], ),
+                              "type": (["stable_diffusion", "stable_cascade", "sd3", "stable_audio", "mochi", "ltxv", "pixart", "cosmos", "lumina2"], ),
                               },
                 "optional": {
                               "device": (["default", "cpu"], {"advanced": True}),
@@ -956,7 +956,7 @@ class CLIPLoader:
 
     CATEGORY = "advanced/loaders"
 
-    DESCRIPTION = "[Recipes]\n\nstable_diffusion: clip-l\nstable_cascade: clip-g\nsd3: t5 / clip-g / clip-l\nstable_audio: t5\nmochi: t5\ncosmos: old t5 xxl"
+    DESCRIPTION = "[Recipes]\n\nstable_diffusion: clip-l\nstable_cascade: clip-g\nsd3: t5 / clip-g / clip-l\nstable_audio: t5\nmochi: t5\ncosmos: old t5 xxl\nlumina2: gemma 2 2B"
 
     def load_clip(self, clip_name, type="stable_diffusion", device="default"):
         clip_type = sd.CLIPType.STABLE_DIFFUSION
@@ -974,6 +974,8 @@ class CLIPLoader:
             clip_type = sd.CLIPType.PIXART
         elif type == "cosmos":
             clip_type = sd.CLIPType.COSMOS
+        elif type == "lumina2":
+            clip_type = comfy.sd.CLIPType.LUMINA2
         else:
             logging.warning(f"Unknown clip type argument passed: {type} for model {clip_name}")
 
@@ -1101,10 +1103,11 @@ class StyleModelApply:
         for t in conditioning:
             (txt, keys) = t
             keys = keys.copy()
-            if strength_type == "attn_bias" and strength != 1.0:
+            # even if the strength is 1.0 (i.e, no change), if there's already a mask, we have to add to it
+            if "attention_mask" in keys or (strength_type == "attn_bias" and strength != 1.0):
                 # math.log raises an error if the argument is zero
                 # torch.log returns -inf, which is what we want
-                attn_bias = torch.log(torch.Tensor([strength]))
+                attn_bias = torch.log(torch.Tensor([strength if strength_type == "attn_bias" else 1.0]))
                 # get the size of the mask image
                 mask_ref_size = keys.get("attention_mask_img_shape", (1, 1))
                 n_ref = mask_ref_size[0] * mask_ref_size[1]
@@ -1793,6 +1796,36 @@ class LoadImageMask:
 
         return True
 
+
+class LoadImageOutput(LoadImage):
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "image": ("COMBO", {
+                    "image_upload": True,
+                    "image_folder": "output",
+                    "remote": {
+                        "route": "/internal/files/output",
+                        "refresh_button": True,
+                        "control_after_refresh": "first",
+                    },
+                }),
+            }
+        }
+
+    DESCRIPTION = "Load an image from the output folder. When the refresh button is clicked, the node will update the image list and automatically select the first image, allowing for easy iteration."
+    EXPERIMENTAL = True
+    FUNCTION = "load_image_output"
+
+    def load_image_output(self, image):
+        return self.load_image(f"{image} [output]")
+
+    @classmethod
+    def VALIDATE_INPUTS(s, image):
+        return True
+
+
 class ImageScale:
     upscale_methods = ["nearest-exact", "bilinear", "area", "bicubic", "lanczos"]
     crop_methods = ["disabled", "center"]
@@ -1979,6 +2012,7 @@ NODE_CLASS_MAPPINGS = {
     "PreviewImage": PreviewImage,
     "LoadImage": LoadImage,
     "LoadImageMask": LoadImageMask,
+    "LoadImageOutput": LoadImageOutput,
     "ImageScale": ImageScale,
     "ImageScaleBy": ImageScaleBy,
     "ImageInvert": ImageInvert,
@@ -2081,6 +2115,7 @@ NODE_DISPLAY_NAME_MAPPINGS = {
     "PreviewImage": "Preview Image",
     "LoadImage": "Load Image",
     "LoadImageMask": "Load Image (as Mask)",
+    "LoadImageOutput": "Load Image (from Outputs)",
     "ImageScale": "Upscale Image",
     "ImageScaleBy": "Upscale Image By",
     "ImageUpscaleWithModel": "Upscale Image (using Model)",

comfy/ops.py

@@ -44,15 +44,17 @@ def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None):
     bias = None
     non_blocking = True if torch.jit.is_tracing() or torch.jit.is_scripting() else model_management.device_supports_non_blocking(device)
     if s.bias is not None:
-        has_function = s.bias_function is not None
+        has_function = len(s.bias_function) > 0
         bias = model_management.cast_to(s.bias, bias_dtype, device, non_blocking=non_blocking, copy=has_function)
         if has_function:
-            bias = s.bias_function(bias)
+            for f in s.bias_function:
+                bias = f(bias)
 
-    has_function = s.weight_function is not None
+    has_function = len(s.weight_function) > 0
     weight = model_management.cast_to(s.weight, dtype, device, non_blocking=non_blocking, copy=has_function)
     if has_function:
-        weight = s.weight_function(weight)
+        for f in s.weight_function:
+            weight = f(weight)
     return weight, bias
 
 
@@ -63,8 +65,8 @@ class SkipInit:
 
 class CastWeightBiasOp:
     comfy_cast_weights = False
-    weight_function = None
-    bias_function = None
+    weight_function = []
+    bias_function = []
 
 
 class skip_init:
@@ -118,7 +120,7 @@ class disable_weight_init:
             return torch.nn.functional.linear(input, weight, bias)
 
         def forward(self, *args, **kwargs):
-            if self.comfy_cast_weights:
+            if self.comfy_cast_weights or len(self.weight_function) > 0 or len(self.bias_function) > 0:
                 return self.forward_comfy_cast_weights(*args, **kwargs)
             else:
                 return super().forward(*args, **kwargs)
@@ -132,7 +134,7 @@ class disable_weight_init:
             return self._conv_forward(input, weight, bias)
 
         def forward(self, *args, **kwargs):
-            if self.comfy_cast_weights:
+            if self.comfy_cast_weights or len(self.weight_function) > 0 or len(self.bias_function) > 0:
                 return self.forward_comfy_cast_weights(*args, **kwargs)
             else:
                 return super().forward(*args, **kwargs)
@@ -146,7 +148,7 @@ class disable_weight_init:
             return self._conv_forward(input, weight, bias)
 
         def forward(self, *args, **kwargs):
-            if self.comfy_cast_weights:
+            if self.comfy_cast_weights or len(self.weight_function) > 0 or len(self.bias_function) > 0:
                 return self.forward_comfy_cast_weights(*args, **kwargs)
             else:
                 return super().forward(*args, **kwargs)
@@ -160,7 +162,7 @@ class disable_weight_init:
             return self._conv_forward(input, weight, bias)
 
         def forward(self, *args, **kwargs):
-            if self.comfy_cast_weights:
+            if self.comfy_cast_weights or len(self.weight_function) > 0 or len(self.bias_function) > 0:
                 return self.forward_comfy_cast_weights(*args, **kwargs)
             else:
                 return super().forward(*args, **kwargs)
@@ -174,7 +176,7 @@ class disable_weight_init:
             return torch.nn.functional.group_norm(input, self.num_groups, weight, bias, self.eps)
 
         def forward(self, *args, **kwargs):
-            if self.comfy_cast_weights:
+            if self.comfy_cast_weights or len(self.weight_function) > 0 or len(self.bias_function) > 0:
                 return self.forward_comfy_cast_weights(*args, **kwargs)
             else:
                 return super().forward(*args, **kwargs)
@@ -192,7 +194,7 @@ class disable_weight_init:
             return torch.nn.functional.layer_norm(input, self.normalized_shape, weight, bias, self.eps)
 
         def forward(self, *args, **kwargs):
-            if self.comfy_cast_weights:
+            if self.comfy_cast_weights or len(self.weight_function) > 0 or len(self.bias_function) > 0:
                 return self.forward_comfy_cast_weights(*args, **kwargs)
             else:
                 return super().forward(*args, **kwargs)
@@ -213,7 +215,7 @@ class disable_weight_init:
                 output_padding, self.groups, self.dilation)
 
         def forward(self, *args, **kwargs):
-            if self.comfy_cast_weights:
+            if self.comfy_cast_weights or len(self.weight_function) > 0 or len(self.bias_function) > 0:
                 return self.forward_comfy_cast_weights(*args, **kwargs)
             else:
                 return super().forward(*args, **kwargs)
@@ -234,7 +236,7 @@ class disable_weight_init:
                 output_padding, self.groups, self.dilation)
 
         def forward(self, *args, **kwargs):
-            if self.comfy_cast_weights:
+            if self.comfy_cast_weights or len(self.weight_function) > 0 or len(self.bias_function) > 0:
                 return self.forward_comfy_cast_weights(*args, **kwargs)
             else:
                 return super().forward(*args, **kwargs)
@@ -252,7 +254,7 @@ class disable_weight_init:
             return torch.nn.functional.embedding(input, weight, self.padding_idx, self.max_norm, self.norm_type, self.scale_grad_by_freq, self.sparse).to(dtype=output_dtype)
 
         def forward(self, *args, **kwargs):
-            if self.comfy_cast_weights:
+            if self.comfy_cast_weights or len(self.weight_function) > 0 or len(self.bias_function) > 0:
                 return self.forward_comfy_cast_weights(*args, **kwargs)
             else:
                 if "out_dtype" in kwargs:

comfy/samplers.py

@@ -724,7 +724,8 @@ class Sampler:
 KSAMPLER_NAMES = ["euler", "euler_cfg_pp", "euler_ancestral", "euler_ancestral_cfg_pp", "heun", "heunpp2", "dpm_2", "dpm_2_ancestral",
                   "lms", "dpm_fast", "dpm_adaptive", "dpmpp_2s_ancestral", "dpmpp_2s_ancestral_cfg_pp", "dpmpp_sde", "dpmpp_sde_gpu",
                   "dpmpp_2m", "dpmpp_2m_cfg_pp", "dpmpp_2m_sde", "dpmpp_2m_sde_gpu", "dpmpp_3m_sde", "dpmpp_3m_sde_gpu", "ddpm", "lcm",
-                  "ipndm", "ipndm_v", "deis", "res_multistep", "res_multistep_cfg_pp", "gradient_estimation"]
+                  "ipndm", "ipndm_v", "deis", "res_multistep", "res_multistep_cfg_pp", "res_multistep_ancestral", "res_multistep_ancestral_cfg_pp",
+                  "gradient_estimation"]
 
 
 class KSAMPLER(Sampler):

comfy/sd.py

@@ -42,6 +42,7 @@ from .text_encoders import hunyuan_video
 from .text_encoders import hydit
 from .text_encoders import long_clipl
 from .text_encoders import lt
+from .text_encoders import lumina2
 from .text_encoders import pixart_t5
 from .text_encoders import sa_t5
 from .text_encoders import sd2_clip
@@ -676,6 +677,7 @@ class CLIPType(Enum):
     HUNYUAN_VIDEO = 9
     PIXART = 10
     COSMOS = 11
+    LUMINA2 = 12
 
 
 @dataclasses.dataclass
@@ -704,6 +706,7 @@ class TEModel(Enum):
     T5_BASE = 6
     LLAMA3_8 = 7
     T5_XXL_OLD = 8
+    GEMMA_2_2B = 9
 
 
 def detect_te_model(sd):
@@ -723,6 +726,8 @@ def detect_te_model(sd):
         return TEModel.T5_XXL_OLD
     if "encoder.block.0.layer.0.SelfAttention.k.weight" in sd:
         return TEModel.T5_BASE
+    if 'model.layers.0.post_feedforward_layernorm.weight' in sd:
+        return TEModel.GEMMA_2_2B
     if "model.layers.0.post_attention_layernorm.weight" in sd:
         return TEModel.LLAMA3_8
     return None
@@ -762,6 +767,7 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
             if "text_projection" in clip_data[i]:
                 clip_data[i]["text_projection.weight"] = clip_data[i]["text_projection"].transpose(0, 1)  # old models saved with the CLIPSave node
 
+    tokenizer_data = {}
     clip_target = CLIPTarget()
     clip_target.params = {}
     if len(clip_data) == 1:
@@ -801,6 +807,10 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
         elif te_model == TEModel.T5_BASE:
             clip_target.clip = sa_t5.SAT5Model
             clip_target.tokenizer = sa_t5.SAT5Tokenizer
+        elif te_model == TEModel.GEMMA_2_2B:
+            clip_target.clip = lumina2.te(**llama_detect(clip_data))
+            clip_target.tokenizer = lumina2.LuminaTokenizer
+            tokenizer_data["spiece_model"] = clip_data[0].get("spiece_model", None)
         else:
             if clip_type == CLIPType.SD3:
                 clip_target.clip = sd3_clip.sd3_clip(clip_l=True, clip_g=False, t5=False)
@@ -830,7 +840,6 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
         clip_target.tokenizer = sd3_clip.SD3Tokenizer
 
     parameters = 0
-    tokenizer_data = {}
     for c in clip_data:
         parameters += utils.calculate_parameters(c)
         tokenizer_data, model_options = long_clipl.model_options_long_clip(c, tokenizer_data, model_options)

comfy/sd1_clip.py

@@ -500,9 +500,11 @@ SDTokenizerT = TypeVar('SDTokenizerT', bound='SDTokenizer')
 
 
 class SDTokenizer:
-    def __init__(self, tokenizer_path: torch.Tensor | bytes | bytearray | memoryview | str | Path | Traversable = None, max_length=77, pad_with_end=True, embedding_directory=None, embedding_size=768, embedding_key='clip_l', tokenizer_class=CLIPTokenizer, has_start_token=True, has_end_token=True, pad_to_max_length=True, min_length=None, pad_token=None, end_token=None, tokenizer_data=None):
+    def __init__(self, tokenizer_path: torch.Tensor | bytes | bytearray | memoryview | str | Path | Traversable = None, max_length=77, pad_with_end=True, embedding_directory=None, embedding_size=768, embedding_key='clip_l', tokenizer_class=CLIPTokenizer, has_start_token=True, has_end_token=True, pad_to_max_length=True, min_length=None, pad_token=None, end_token=None, tokenizer_data=None, tokenizer_args=None):
         if tokenizer_data is None:
             tokenizer_data = dict()
+        if tokenizer_args is None:
+            tokenizer_args = dict()
         if tokenizer_path is None:
             tokenizer_path = files.get_package_as_path("comfy.sd1_tokenizer")
         if isinstance(tokenizer_path, Path):
@@ -515,7 +517,7 @@ class SDTokenizer:
             tokenizer_path = get_package_as_path('comfy.sd1_tokenizer')
         self.tokenizer_class = tokenizer_class
         self.tokenizer_path = tokenizer_path
-        self.tokenizer: PreTrainedTokenizerBase | SPieceTokenizer = tokenizer_class.from_pretrained(tokenizer_path)
+        self.tokenizer: PreTrainedTokenizerBase | SPieceTokenizer = tokenizer_class.from_pretrained(tokenizer_path, **tokenizer_args)
         self.max_length = max_length
         self.min_length = min_length
         self.end_token = None
@@ -699,11 +701,15 @@ SD1TokenizerT = TypeVar("SD1TokenizerT", bound="SD1Tokenizer")
 
 
 class SD1Tokenizer:
-    def __init__(self, embedding_directory=None, tokenizer_data=None, clip_name="l", tokenizer=SDTokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data: dict=None, clip_name="l", tokenizer=SDTokenizer, name=None):
         if tokenizer_data is None:
             tokenizer_data = {}
-        self.clip_name = clip_name
-        self.clip = "clip_{}".format(self.clip_name)
+        if name is not None:
+            self.clip_name = name
+            self.clip = "{}".format(self.clip_name)
+        else:
+            self.clip_name = clip_name
+            self.clip = "clip_{}".format(self.clip_name)
         tokenizer = tokenizer_data.get("{}_tokenizer_class".format(self.clip), tokenizer)
         self.sd_tokenizer = tokenizer(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data)
 
@@ -729,7 +735,7 @@ class SD1Tokenizer:
         return sd1_tokenizer
 
     def state_dict(self):
-        return {}
+        return getattr(self, self.clip).state_dict()
 
 class SD1CheckpointClipModel(SDClipModel):
     def __init__(self, device="cpu", dtype=None, model_options=None, textmodel_json_config=None):

comfy/supported_models.py

@@ -14,6 +14,7 @@ from .text_encoders import genmo
 from .text_encoders import hunyuan_video
 from .text_encoders import hydit
 from .text_encoders import lt
+from .text_encoders import lumina2
 from .text_encoders import pixart_t5
 from .text_encoders import sa_t5
 from .text_encoders import sd2_clip
@@ -63,7 +64,9 @@ class SD15(supported_models_base.BASE):
         replace_prefix = {"clip_l.": "cond_stage_model."}
         return utils.state_dict_prefix_replace(state_dict, replace_prefix)
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         return supported_models_base.ClipTarget(sd1_clip.SD1Tokenizer, sd1_clip.SD1ClipModel)
 
 
@@ -108,7 +111,9 @@ class SD20(supported_models_base.BASE):
         state_dict = diffusers_convert.convert_text_enc_state_dict_v20(state_dict)
         return state_dict
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         return supported_models_base.ClipTarget(sd2_clip.SD2Tokenizer, sd2_clip.SD2ClipModel)
 
 
@@ -173,7 +178,9 @@ class SDXLRefiner(supported_models_base.BASE):
         state_dict_g = utils.state_dict_prefix_replace(state_dict_g, replace_prefix)
         return state_dict_g
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         return supported_models_base.ClipTarget(sdxl_clip.SDXLTokenizer, sdxl_clip.SDXLRefinerClipModel)
 
 
@@ -246,7 +253,9 @@ class SDXL(supported_models_base.BASE):
         state_dict_g = utils.state_dict_prefix_replace(state_dict_g, replace_prefix)
         return state_dict_g
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         return supported_models_base.ClipTarget(sdxl_clip.SDXLTokenizer, sdxl_clip.SDXLClipModel)
 
 
@@ -322,7 +331,9 @@ class SVD_img2vid(supported_models_base.BASE):
         out = model_base.SVD_img2vid(self, device=device)
         return out
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         return None
 
 
@@ -390,7 +401,9 @@ class Stable_Zero123(supported_models_base.BASE):
         out = model_base.Stable_Zero123(self, device=device, cc_projection_weight=state_dict["cc_projection.weight"], cc_projection_bias=state_dict["cc_projection.bias"])
         return out
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         return None
 
 
@@ -466,7 +479,9 @@ class Stable_Cascade_C(supported_models_base.BASE):
         out = model_base.StableCascade_C(self, device=device)
         return out
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         return supported_models_base.ClipTarget(sdxl_clip.StableCascadeTokenizer, sdxl_clip.StableCascadeClipModel)
 
 
@@ -541,7 +556,9 @@ class SD3(supported_models_base.BASE):
         out = model_base.SD3(self, device=device)
         return out
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         clip_l = False
         clip_g = False
         t5 = False
@@ -585,7 +602,9 @@ class StableAudio(supported_models_base.BASE):
         replace_prefix = {"": "model.model."}
         return utils.state_dict_prefix_replace(state_dict, replace_prefix)
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         return supported_models_base.ClipTarget(sa_t5.SAT5Tokenizer, sa_t5.SAT5Model)
 
 
@@ -609,7 +628,9 @@ class AuraFlow(supported_models_base.BASE):
         out = model_base.AuraFlow(self, device=device)
         return out
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         return supported_models_base.ClipTarget(aura_t5.AuraT5Tokenizer, aura_t5.AuraT5Model)
 
 
@@ -675,7 +696,9 @@ class HunyuanDiT(supported_models_base.BASE):
         out = model_base.HunyuanDiT(self, device=device)
         return out
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         return supported_models_base.ClipTarget(hydit.HyditTokenizer, hydit.HyditModel)
 
 
@@ -715,7 +738,9 @@ class Flux(supported_models_base.BASE):
         out = model_base.Flux(self, device=device)
         return out
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         pref = self.text_encoder_key_prefix[0]
         t5_detect = sd3_clip.t5_xxl_detect(state_dict, "{}t5xxl.transformer.".format(pref))
         return supported_models_base.ClipTarget(flux.FluxTokenizer, flux.flux_clip(**t5_detect))
@@ -802,7 +827,9 @@ class LTXV(supported_models_base.BASE):
         out = model_base.LTXV(self, device=device)
         return out
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         pref = self.text_encoder_key_prefix[0]
         t5_detect = sd3_clip.t5_xxl_detect(state_dict, "{}t5xxl.transformer.".format(pref))
         return supported_models_base.ClipTarget(lt.LTXVT5Tokenizer, lt.ltxv_te(**t5_detect))
@@ -885,7 +912,9 @@ class CosmosT2V(supported_models_base.BASE):
         out = model_base.CosmosVideo(self, device=device)
         return out
 
-    def clip_target(self, state_dict={}):
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
         pref = self.text_encoder_key_prefix[0]
         t5_detect = sd3_clip.t5_xxl_detect(state_dict, "{}t5xxl.transformer.".format(pref))
         return supported_models_base.ClipTarget(cosmos.CosmosT5Tokenizer, cosmos.te(**t5_detect))
@@ -902,6 +931,38 @@ class CosmosI2V(CosmosT2V):
         return out
 
 
-models = [Stable_Zero123, SD15_instructpix2pix, SD15, SD20, SD21UnclipL, SD21UnclipH, SDXL_instructpix2pix, SDXLRefiner, SDXL, SSD1B, KOALA_700M, KOALA_1B, Segmind_Vega, SD_X4Upscaler, Stable_Cascade_C, Stable_Cascade_B, SV3D_u, SV3D_p, SD3, StableAudio, AuraFlow, PixArtAlpha, PixArtSigma, HunyuanDiT, HunyuanDiT1, FluxInpaint, Flux, FluxSchnell, GenmoMochi, LTXV, HunyuanVideo, CosmosT2V, CosmosI2V]
+class Lumina2(supported_models_base.BASE):
+    unet_config = {
+        "image_model": "lumina2",
+    }
+
+    sampling_settings = {
+        "multiplier": 1.0,
+        "shift": 6.0,
+    }
+
+    memory_usage_factor = 1.2
+
+    unet_extra_config = {}
+    latent_format = latent_formats.Flux
+
+    supported_inference_dtypes = [torch.bfloat16, torch.float32]
+
+    vae_key_prefix = ["vae."]
+    text_encoder_key_prefix = ["text_encoders."]
+
+    def get_model(self, state_dict, prefix="", device=None):
+        out = model_base.Lumina2(self, device=device)
+        return out
+
+    def clip_target(self, state_dict=None):
+        if state_dict is None:
+            state_dict = {}
+        pref = self.text_encoder_key_prefix[0]
+        hunyuan_detect = hunyuan_video.llama_detect(state_dict, "{}gemma2_2b.transformer.".format(pref))
+        return supported_models_base.ClipTarget(lumina2.LuminaTokenizer, lumina2.te(**hunyuan_detect))
+
+
+models = [Stable_Zero123, SD15_instructpix2pix, SD15, SD20, SD21UnclipL, SD21UnclipH, SDXL_instructpix2pix, SDXLRefiner, SDXL, SSD1B, KOALA_700M, KOALA_1B, Segmind_Vega, SD_X4Upscaler, Stable_Cascade_C, Stable_Cascade_B, SV3D_u, SV3D_p, SD3, StableAudio, AuraFlow, PixArtAlpha, PixArtSigma, HunyuanDiT, HunyuanDiT1, FluxInpaint, Flux, FluxSchnell, GenmoMochi, LTXV, HunyuanVideo, CosmosT2V, CosmosI2V, Lumina2]
 
 models += [SVD_img2vid]

comfy/text_encoders/bert.py

@@ -120,7 +120,7 @@ class BertModel_(torch.nn.Module):
         mask = None
         if attention_mask is not None:
             mask = 1.0 - attention_mask.to(x.dtype).reshape((attention_mask.shape[0], 1, -1, attention_mask.shape[-1])).expand(attention_mask.shape[0], 1, attention_mask.shape[-1], attention_mask.shape[-1])
-            mask = mask.masked_fill(mask.to(torch.bool), float("-inf"))
+            mask = mask.masked_fill(mask.to(torch.bool), -torch.finfo(x.dtype).max)
 
         x, i = self.encoder(x, mask, intermediate_output)
         return x, i

comfy/text_encoders/llama.py

@@ -1,9 +1,9 @@
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
 from dataclasses import dataclass
 from typing import Optional, Any
 
+import torch
+import torch.nn as nn
+
 from ..ldm.common_dit import rms_norm
 from ..ldm.modules.attention import optimized_attention_for_device
 
@@ -19,21 +19,48 @@ class Llama2Config:
     max_position_embeddings: int = 8192
     rms_norm_eps: float = 1e-5
     rope_theta: float = 500000.0
+    transformer_type: str = "llama"
+    head_dim = 128
+    rms_norm_add = False
+    mlp_activation = "silu"
+
+
+@dataclass
+class Gemma2_2B_Config:
+    vocab_size: int = 256000
+    hidden_size: int = 2304
+    intermediate_size: int = 9216
+    num_hidden_layers: int = 26
+    num_attention_heads: int = 8
+    num_key_value_heads: int = 4
+    max_position_embeddings: int = 8192
+    rms_norm_eps: float = 1e-6
+    rope_theta: float = 10000.0
+    transformer_type: str = "gemma2"
+    head_dim = 256
+    rms_norm_add = True
+    mlp_activation = "gelu_pytorch_tanh"
+
 
 class RMSNorm(nn.Module):
-    def __init__(self, dim: int, eps: float = 1e-5, device=None, dtype=None):
+    def __init__(self, dim: int, eps: float = 1e-5, add=False, device=None, dtype=None):
         super().__init__()
         self.eps = eps
         self.weight = nn.Parameter(torch.empty(dim, device=device, dtype=dtype))
+        self.add = add
 
     def forward(self, x: torch.Tensor):
-        return rms_norm(x, self.weight, self.eps)
+        w = self.weight
+        if self.add:
+            w = w + 1.0
+
+        return rms_norm(x, w, self.eps)
 
 
 def rotate_half(x):
     """Rotates half the hidden dims of the input."""
     x1 = x[..., : x.shape[-1] // 2]
-    x2 = x[..., x.shape[-1] // 2 :]
+    x2 = x[..., x.shape[-1] // 2:]
     return torch.cat((-x2, x1), dim=-1)
 
 
@@ -66,23 +93,24 @@ class Attention(nn.Module):
         self.num_heads = config.num_attention_heads
         self.num_kv_heads = config.num_key_value_heads
         self.hidden_size = config.hidden_size
-        self.head_dim = self.hidden_size // self.num_heads
+
+        self.head_dim = config.head_dim
+        self.inner_size = self.num_heads * self.head_dim
 
         ops = ops or nn
-        self.q_proj = ops.Linear(config.hidden_size, config.hidden_size, bias=False, device=device, dtype=dtype)
+        self.q_proj = ops.Linear(config.hidden_size, self.inner_size, bias=False, device=device, dtype=dtype)
         self.k_proj = ops.Linear(config.hidden_size, self.num_kv_heads * self.head_dim, bias=False, device=device, dtype=dtype)
         self.v_proj = ops.Linear(config.hidden_size, self.num_kv_heads * self.head_dim, bias=False, device=device, dtype=dtype)
-        self.o_proj = ops.Linear(config.hidden_size, config.hidden_size, bias=False, device=device, dtype=dtype)
+        self.o_proj = ops.Linear(self.inner_size, config.hidden_size, bias=False, device=device, dtype=dtype)
 
     def forward(
-        self,
-        hidden_states: torch.Tensor,
-        attention_mask: Optional[torch.Tensor] = None,
-        freqs_cis: Optional[torch.Tensor] = None,
-        optimized_attention=None,
+            self,
+            hidden_states: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            freqs_cis: Optional[torch.Tensor] = None,
+            optimized_attention=None,
     ):
         batch_size, seq_length, _ = hidden_states.shape
-
         xq = self.q_proj(hidden_states)
         xk = self.k_proj(hidden_states)
         xv = self.v_proj(hidden_states)
@@ -99,6 +127,7 @@ class Attention(nn.Module):
         output = optimized_attention(xq, xk, xv, self.num_heads, mask=attention_mask, skip_reshape=True)
         return self.o_proj(output)
 
+
 class MLP(nn.Module):
     def __init__(self, config: Llama2Config, device=None, dtype=None, ops: Any = None):
         super().__init__()
@@ -106,9 +135,14 @@ class MLP(nn.Module):
         self.gate_proj = ops.Linear(config.hidden_size, config.intermediate_size, bias=False, device=device, dtype=dtype)
         self.up_proj = ops.Linear(config.hidden_size, config.intermediate_size, bias=False, device=device, dtype=dtype)
         self.down_proj = ops.Linear(config.intermediate_size, config.hidden_size, bias=False, device=device, dtype=dtype)
+        if config.mlp_activation == "silu":
+            self.activation = torch.nn.functional.silu
+        elif config.mlp_activation == "gelu_pytorch_tanh":
+            self.activation = lambda a: torch.nn.functional.gelu(a, approximate="tanh")
 
     def forward(self, x):
-        return self.down_proj(F.silu(self.gate_proj(x)) * self.up_proj(x))
+        return self.down_proj(self.activation(self.gate_proj(x)) * self.up_proj(x))
+
 
 class TransformerBlock(nn.Module):
     def __init__(self, config: Llama2Config, device=None, dtype=None, ops: Any = None):
@@ -119,11 +153,11 @@ class TransformerBlock(nn.Module):
         self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, device=device, dtype=dtype)
 
     def forward(
-        self,
-        x: torch.Tensor,
-        attention_mask: Optional[torch.Tensor] = None,
-        freqs_cis: Optional[torch.Tensor] = None,
-        optimized_attention=None,
+            self,
+            x: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            freqs_cis: Optional[torch.Tensor] = None,
+            optimized_attention=None,
     ):
         # Self Attention
         residual = x
@@ -144,6 +178,47 @@ class TransformerBlock(nn.Module):
 
         return x
 
+
+class TransformerBlockGemma2(nn.Module):
+    def __init__(self, config: Llama2Config, device=None, dtype=None, ops: Any = None):
+        super().__init__()
+        self.self_attn = Attention(config, device=device, dtype=dtype, ops=ops)
+        self.mlp = MLP(config, device=device, dtype=dtype, ops=ops)
+        self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, add=config.rms_norm_add, device=device, dtype=dtype)
+        self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, add=config.rms_norm_add, device=device, dtype=dtype)
+        self.pre_feedforward_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, add=config.rms_norm_add, device=device, dtype=dtype)
+        self.post_feedforward_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, add=config.rms_norm_add, device=device, dtype=dtype)
+
+    def forward(
+            self,
+            x: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            freqs_cis: Optional[torch.Tensor] = None,
+            optimized_attention=None,
+    ):
+        # Self Attention
+        residual = x
+        x = self.input_layernorm(x)
+        x = self.self_attn(
+            hidden_states=x,
+            attention_mask=attention_mask,
+            freqs_cis=freqs_cis,
+            optimized_attention=optimized_attention,
+        )
+
+        x = self.post_attention_layernorm(x)
+        x = residual + x
+
+        # MLP
+        residual = x
+        x = self.pre_feedforward_layernorm(x)
+        x = self.mlp(x)
+        x = self.post_feedforward_layernorm(x)
+        x = residual + x
+
+        return x
+
+
 class Llama2_(nn.Module):
     def __init__(self, config, device=None, dtype=None, ops=None):
         super().__init__()
@@ -156,17 +231,27 @@ class Llama2_(nn.Module):
             device=device,
             dtype=dtype
         )
+        if self.config.transformer_type == "gemma2":
+            transformer = TransformerBlockGemma2
+            self.normalize_in = True
+        else:
+            transformer = TransformerBlock
+            self.normalize_in = False
+
         self.layers = nn.ModuleList([
-            TransformerBlock(config, device=device, dtype=dtype, ops=ops)
+            transformer(config, device=device, dtype=dtype, ops=ops)
             for _ in range(config.num_hidden_layers)
         ])
-        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, device=device, dtype=dtype)
+        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, add=config.rms_norm_add, device=device, dtype=dtype)
         # self.lm_head = ops.Linear(config.hidden_size, config.vocab_size, bias=False, device=device, dtype=dtype)
 
     def forward(self, x, attention_mask=None, intermediate_output=None, final_layer_norm_intermediate=True, dtype=None):
         x = self.embed_tokens(x, out_dtype=dtype)
 
-        freqs_cis = precompute_freqs_cis(self.config.hidden_size // self.config.num_attention_heads,
+        if self.normalize_in:
+            x *= self.config.hidden_size ** 0.5
+
+        freqs_cis = precompute_freqs_cis(self.config.head_dim,
                                          x.shape[1],
                                          self.config.rope_theta,
                                          device=x.device)
@@ -205,15 +290,7 @@ class Llama2_(nn.Module):
         return x, intermediate
 
 
-class Llama2(torch.nn.Module):
-    def __init__(self, config_dict, dtype, device, operations):
-        super().__init__()
-        config = Llama2Config(**config_dict)
-        self.num_layers = config.num_hidden_layers
-
-        self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
-        self.dtype = dtype
-
+class BaseLlama:
     def get_input_embeddings(self):
         return self.model.embed_tokens
 
@@ -222,3 +299,23 @@ class Llama2(torch.nn.Module):
 
     def forward(self, input_ids, *args, **kwargs):
         return self.model(input_ids, *args, **kwargs)
+
+
+class Llama2(BaseLlama, torch.nn.Module):
+    def __init__(self, config_dict, dtype, device, operations):
+        super().__init__()
+        config = Llama2Config(**config_dict)
+        self.num_layers = config.num_hidden_layers
+
+        self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
+        self.dtype = dtype
+
+
+class Gemma2_2B(BaseLlama, torch.nn.Module):
+    def __init__(self, config_dict, dtype, device, operations):
+        super().__init__()
+        config = Gemma2_2B_Config(**config_dict)
+        self.num_layers = config.num_hidden_layers
+
+        self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
+        self.dtype = dtype

comfy/text_encoders/lumina2.py

@@ -0,0 +1,44 @@
+from comfy import sd1_clip
+from .spiece_tokenizer import SPieceTokenizer
+import comfy.text_encoders.llama
+
+
+class Gemma2BTokenizer(sd1_clip.SDTokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        tokenizer = tokenizer_data.get("spiece_model", None)
+        super().__init__(tokenizer, pad_with_end=False, embedding_size=2304, embedding_key='gemma2_2b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, tokenizer_args={"add_bos": True, "add_eos": False})
+
+    def state_dict(self):
+        return {"spiece_model": self.tokenizer.serialize_model()}
+
+
+class LuminaTokenizer(sd1_clip.SD1Tokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, name="gemma2_2b", tokenizer=Gemma2BTokenizer)
+
+
+class Gemma2_2BModel(sd1_clip.SDClipModel):
+    def __init__(self, device="cpu", layer="hidden", layer_idx=-2, dtype=None, attention_mask=True, model_options={}):
+        llama_scaled_fp8 = model_options.get("llama_scaled_fp8", None)
+        if llama_scaled_fp8 is not None:
+            model_options = model_options.copy()
+            model_options["scaled_fp8"] = llama_scaled_fp8
+
+        super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config={}, dtype=dtype, special_tokens={"start": 2, "pad": 0}, layer_norm_hidden_state=False, model_class=comfy.text_encoders.llama.Gemma2_2B, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options)
+
+
+class LuminaModel(sd1_clip.SD1ClipModel):
+    def __init__(self, device="cpu", dtype=None, model_options={}):
+        super().__init__(device=device, dtype=dtype, name="gemma2_2b", clip_model=Gemma2_2BModel, model_options=model_options)
+
+
+def te(dtype_llama=None, llama_scaled_fp8=None):
+    class LuminaTEModel_(LuminaModel):
+        def __init__(self, device="cpu", dtype=None, model_options={}):
+            if llama_scaled_fp8 is not None and "llama_scaled_fp8" not in model_options:
+                model_options = model_options.copy()
+                model_options["llama_scaled_fp8"] = llama_scaled_fp8
+                if dtype_llama is not None:
+                    dtype = dtype_llama
+            super().__init__(device=device, dtype=dtype, model_options=model_options)
+    return LuminaTEModel_

comfy/text_encoders/spiece_tokenizer.py

@@ -1,26 +1,31 @@
 import copy
+from pathlib import Path
 
 import sentencepiece
 import torch
 
 
 class SPieceTokenizer:
-    add_eos = True
-
     @staticmethod
-    def from_pretrained(path):
-        return SPieceTokenizer(path)
+    def from_pretrained(path, **kwargs):
+        return SPieceTokenizer(path, **kwargs)
 
-    def __init__(self, tokenizer_path):
+    def __init__(self, tokenizer_path: bytes | str | Path, add_bos=False, add_eos=True):
+        self.add_bos = add_bos
+        self.add_eos = add_eos
         if torch.is_tensor(tokenizer_path):
             tokenizer_path = tokenizer_path.numpy().tobytes()
 
-        construction_args = {}
+        construction_args = {
+            'add_bos': self.add_bos,
+            'add_eos': self.add_eos
+        }
+
         if isinstance(tokenizer_path, bytes):
             construction_args["model_proto"] = tokenizer_path
         else:
             construction_args["model_file"] = tokenizer_path
-        self.tokenizer = sentencepiece.SentencePieceProcessor(add_eos=SPieceTokenizer.add_eos, **construction_args)  # pylint: disable=unexpected-keyword-arg
+        self.tokenizer = sentencepiece.SentencePieceProcessor(**construction_args)  # pylint: disable=unexpected-keyword-arg
 
         self.end = self.tokenizer.eos_id()
         self.eos_token_id = self.end
@@ -41,4 +46,3 @@ class SPieceTokenizer:
 
     def clone(self):
         return copy.copy(self)
-

comfy/text_encoders/t5.py

@@ -214,7 +214,7 @@ class T5Stack(torch.nn.Module):
         mask = None
         if attention_mask is not None:
             mask = 1.0 - attention_mask.to(x.dtype).reshape((attention_mask.shape[0], 1, -1, attention_mask.shape[-1])).expand(attention_mask.shape[0], 1, attention_mask.shape[-1], attention_mask.shape[-1])
-            mask = mask.masked_fill(mask.to(torch.bool), float("-inf"))
+            mask = mask.masked_fill(mask.to(torch.bool), -torch.finfo(x.dtype).max)
 
         intermediate = None
         optimized_attention = optimized_attention_for_device(x.device, mask=attention_mask is not None, small_input=True)

comfy/utils.py

@@ -97,7 +97,7 @@ def load_torch_file(ckpt: str, safe_load=False, device=None):
                 if "HeaderTooLarge" in message:
                     raise ValueError("{}\n\nFile path: {}\n\nThe safetensors file is corrupt or invalid. Make sure this is actually a safetensors file and not a ckpt or pt or other filetype.".format(message, ckpt))
                 if "MetadataIncompleteBuffer" in message:
-                    raise ValueError("{}\n\nFile path: {}\n\nThe safetensors file is incomplete. Check the file size and make sure you have copied/downloaded it correctly.".format(message, ckpt))
+                    raise ValueError("{}\n\nFile path: {}\n\nThe safetensors file is corrupt/incomplete. Check the file size and make sure you have copied/downloaded it correctly.".format(message, ckpt))
             raise e
     elif ckpt.lower().endswith("index.json"):
         # from accelerate

comfy/web/assets/BaseViewTemplate-BTbuZf5t.js

@@ -1,4 +1,8 @@
+<<<<<<<< HEAD:comfy/web/assets/BaseViewTemplate-DDUNNAbV.js
 import { d as defineComponent, U as ref, p as onMounted, b4 as isElectron, W as nextTick, b5 as electronAPI, o as openBlock, f as createElementBlock, i as withDirectives, v as vShow, j as unref, b6 as isNativeWindow, m as createBaseVNode, A as renderSlot, ai as normalizeClass } from "./index-BsGgXmrT.js";
+========
+import { d as defineComponent, T as ref, p as onMounted, b8 as isElectron, V as nextTick, b9 as electronAPI, o as openBlock, f as createElementBlock, i as withDirectives, v as vShow, j as unref, ba as isNativeWindow, m as createBaseVNode, A as renderSlot, aj as normalizeClass } from "./index-Bv0b06LE.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/BaseViewTemplate-BTbuZf5t.js
 const _hoisted_1 = { class: "flex-grow w-full flex items-center justify-center overflow-auto" };
 const _sfc_main = /* @__PURE__ */ defineComponent({
   __name: "BaseViewTemplate",
@@ -27,7 +31,7 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
     });
     return (_ctx, _cache) => {
       return openBlock(), createElementBlock("div", {
-        class: normalizeClass(["font-sans w-screen h-screen flex flex-col pointer-events-auto", [
+        class: normalizeClass(["font-sans w-screen h-screen flex flex-col", [
           props.dark ? "text-neutral-300 bg-neutral-900 dark-theme" : "text-neutral-900 bg-neutral-300"
         ]])
       }, [
@@ -48,4 +52,8 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
 export {
   _sfc_main as _
 };
+<<<<<<<< HEAD:comfy/web/assets/BaseViewTemplate-DDUNNAbV.js
 //# sourceMappingURL=BaseViewTemplate-DDUNNAbV.js.map
+========
+//# sourceMappingURL=BaseViewTemplate-BTbuZf5t.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/BaseViewTemplate-BTbuZf5t.js

comfy/web/assets/DesktopStartView-D9r53Bue.js

@@ -0,0 +1,19 @@
+import { d as defineComponent, o as openBlock, y as createBlock, z as withCtx, k as createVNode, j as unref, bE as script } from "./index-Bv0b06LE.js";
+import { _ as _sfc_main$1 } from "./BaseViewTemplate-BTbuZf5t.js";
+const _sfc_main = /* @__PURE__ */ defineComponent({
+  __name: "DesktopStartView",
+  setup(__props) {
+    return (_ctx, _cache) => {
+      return openBlock(), createBlock(_sfc_main$1, { dark: "" }, {
+        default: withCtx(() => [
+          createVNode(unref(script), { class: "m-8 w-48 h-48" })
+        ]),
+        _: 1
+      });
+    };
+  }
+});
+export {
+  _sfc_main as default
+};
+//# sourceMappingURL=DesktopStartView-D9r53Bue.js.map

comfy/web/assets/DesktopStartView-elroCqfp.js

@@ -1,22 +0,0 @@
-import { d as defineComponent, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, k as createVNode, j as unref, bs as script } from "./index-BsGgXmrT.js";
-import { _ as _sfc_main$1 } from "./BaseViewTemplate-DDUNNAbV.js";
-const _hoisted_1 = { class: "max-w-screen-sm w-screen p-8" };
-const _sfc_main = /* @__PURE__ */ defineComponent({
-  __name: "DesktopStartView",
-  setup(__props) {
-    return (_ctx, _cache) => {
-      return openBlock(), createBlock(_sfc_main$1, { dark: "" }, {
-        default: withCtx(() => [
-          createBaseVNode("div", _hoisted_1, [
-            createVNode(unref(script), { mode: "indeterminate" })
-          ])
-        ]),
-        _: 1
-      });
-    };
-  }
-});
-export {
-  _sfc_main as default
-};
-//# sourceMappingURL=DesktopStartView-elroCqfp.js.map

comfy/web/assets/DesktopUpdateView-C-R0415K.js

@@ -0,0 +1,58 @@
+var __defProp = Object.defineProperty;
+var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+import { d as defineComponent, T as ref, d8 as onUnmounted, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, j as unref, bg as t, k as createVNode, bE as script, l as script$1, b9 as electronAPI, _ as _export_sfc } from "./index-Bv0b06LE.js";
+import { s as script$2 } from "./index-A_bXPJCN.js";
+import { _ as _sfc_main$1 } from "./TerminalOutputDrawer-CKr7Br7O.js";
+import { _ as _sfc_main$2 } from "./BaseViewTemplate-BTbuZf5t.js";
+const _hoisted_1 = { class: "h-screen w-screen grid items-center justify-around overflow-y-auto" };
+const _hoisted_2 = { class: "relative m-8 text-center" };
+const _hoisted_3 = { class: "download-bg pi-download text-4xl font-bold" };
+const _hoisted_4 = { class: "m-8" };
+const _sfc_main = /* @__PURE__ */ defineComponent({
+  __name: "DesktopUpdateView",
+  setup(__props) {
+    const electron = electronAPI();
+    const terminalVisible = ref(false);
+    const toggleConsoleDrawer = /* @__PURE__ */ __name(() => {
+      terminalVisible.value = !terminalVisible.value;
+    }, "toggleConsoleDrawer");
+    onUnmounted(() => electron.Validation.dispose());
+    return (_ctx, _cache) => {
+      return openBlock(), createBlock(_sfc_main$2, { dark: "" }, {
+        default: withCtx(() => [
+          createBaseVNode("div", _hoisted_1, [
+            createBaseVNode("div", _hoisted_2, [
+              createBaseVNode("h1", _hoisted_3, toDisplayString(unref(t)("desktopUpdate.title")), 1),
+              createBaseVNode("div", _hoisted_4, [
+                createBaseVNode("span", null, toDisplayString(unref(t)("desktopUpdate.description")), 1)
+              ]),
+              createVNode(unref(script), { class: "m-8 w-48 h-48" }),
+              createVNode(unref(script$1), {
+                style: { "transform": "translateX(-50%)" },
+                class: "fixed bottom-0 left-1/2 my-8",
+                label: unref(t)("maintenance.consoleLogs"),
+                icon: "pi pi-desktop",
+                "icon-pos": "left",
+                severity: "secondary",
+                onClick: toggleConsoleDrawer
+              }, null, 8, ["label"]),
+              createVNode(_sfc_main$1, {
+                modelValue: terminalVisible.value,
+                "onUpdate:modelValue": _cache[0] || (_cache[0] = ($event) => terminalVisible.value = $event),
+                header: unref(t)("g.terminal"),
+                "default-message": unref(t)("desktopUpdate.terminalDefaultMessage")
+              }, null, 8, ["modelValue", "header", "default-message"])
+            ])
+          ]),
+          createVNode(unref(script$2))
+        ]),
+        _: 1
+      });
+    };
+  }
+});
+const DesktopUpdateView = /* @__PURE__ */ _export_sfc(_sfc_main, [["__scopeId", "data-v-8d77828d"]]);
+export {
+  DesktopUpdateView as default
+};
+//# sourceMappingURL=DesktopUpdateView-C-R0415K.js.map

comfy/web/assets/DesktopUpdateView-CxchaIvw.css

@@ -0,0 +1,20 @@
+
+.download-bg[data-v-8d77828d]::before {
+  position: absolute;
+  margin: 0px;
+  color: var(--p-text-muted-color);
+  font-family: 'primeicons';
+  top: -2rem;
+  right: 2rem;
+  speak: none;
+  font-style: normal;
+  font-weight: normal;
+  font-variant: normal;
+  text-transform: none;
+  line-height: 1;
+  display: inline-block;
+  -webkit-font-smoothing: antialiased;
+  opacity: 0.02;
+  font-size: min(14rem, 90vw);
+  z-index: 0
+}

comfy/web/assets/DownloadGitView-PWqK5ke4.js

@@ -1,7 +1,12 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/DownloadGitView-BFcFCk37.js
 import { d as defineComponent, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, k as createVNode, j as unref, l as script, be as useRouter } from "./index-BsGgXmrT.js";
 import { _ as _sfc_main$1 } from "./BaseViewTemplate-DDUNNAbV.js";
+========
+import { d as defineComponent, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, k as createVNode, j as unref, l as script, bi as useRouter } from "./index-Bv0b06LE.js";
+import { _ as _sfc_main$1 } from "./BaseViewTemplate-BTbuZf5t.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/DownloadGitView-PWqK5ke4.js
 const _hoisted_1 = { class: "max-w-screen-sm flex flex-col gap-8 p-8 bg-[url('/assets/images/Git-Logo-White.svg')] bg-no-repeat bg-right-top bg-origin-padding" };
 const _hoisted_2 = { class: "mt-24 text-4xl font-bold text-red-500" };
 const _hoisted_3 = { class: "space-y-4" };
@@ -55,4 +60,8 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
 export {
   _sfc_main as default
 };
+<<<<<<<< HEAD:comfy/web/assets/DownloadGitView-BFcFCk37.js
 //# sourceMappingURL=DownloadGitView-BFcFCk37.js.map
+========
+//# sourceMappingURL=DownloadGitView-PWqK5ke4.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/DownloadGitView-PWqK5ke4.js

comfy/web/assets/ExtensionPanel-Ba57xrmg.js

@@ -1,8 +1,14 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/ExtensionPanel-BPpLOa_B.js
 import { d as defineComponent, U as ref, df as FilterMatchMode, dk as useExtensionStore, a as useSettingStore, p as onMounted, c as computed, o as openBlock, y as createBlock, z as withCtx, k as createVNode, dg as SearchBox, j as unref, bj as script, m as createBaseVNode, f as createElementBlock, D as renderList, E as toDisplayString, a7 as createTextVNode, F as Fragment, l as script$1, B as createCommentVNode, a4 as script$3, ax as script$4, bn as script$5, dh as _sfc_main$1 } from "./index-BsGgXmrT.js";
 import { g as script$2, h as script$6 } from "./index-Br6dw1F6.js";
 import "./index-COyiXDAn.js";
+========
+import { d as defineComponent, T as ref, dx as FilterMatchMode, dC as useExtensionStore, a as useSettingStore, p as onMounted, c as computed, o as openBlock, y as createBlock, z as withCtx, k as createVNode, dy as SearchBox, j as unref, bn as script, m as createBaseVNode, f as createElementBlock, D as renderList, E as toDisplayString, a8 as createTextVNode, F as Fragment, l as script$1, B as createCommentVNode, a5 as script$3, ay as script$4, br as script$5, dz as _sfc_main$1 } from "./index-Bv0b06LE.js";
+import { g as script$2, h as script$6 } from "./index-CgMyWf7n.js";
+import "./index-Dzu9WL4p.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/ExtensionPanel-Ba57xrmg.js
 const _hoisted_1 = { class: "flex justify-end" };
 const _sfc_main = /* @__PURE__ */ defineComponent({
   __name: "ExtensionPanel",
@@ -179,4 +185,8 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
 export {
   _sfc_main as default
 };
+<<<<<<<< HEAD:comfy/web/assets/ExtensionPanel-BPpLOa_B.js
 //# sourceMappingURL=ExtensionPanel-BPpLOa_B.js.map
+========
+//# sourceMappingURL=ExtensionPanel-Ba57xrmg.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/ExtensionPanel-Ba57xrmg.js

comfy/web/assets/GraphView-Bo28XDd0.css

@@ -1,6 +1,5 @@
 
-.comfy-menu-hamburger[data-v-7ed57d1a] {
-    pointer-events: auto;
+.comfy-menu-hamburger[data-v-82120b51] {
     position: fixed;
     z-index: 9999;
     display: flex;
@@ -41,19 +40,19 @@
   z-index: 999;
 }
 
-.p-buttongroup-vertical[data-v-cb8f9a1a] {
+.p-buttongroup-vertical[data-v-27a9500c] {
   display: flex;
   flex-direction: column;
   border-radius: var(--p-button-border-radius);
   overflow: hidden;
   border: 1px solid var(--p-panel-border-color);
 }
-.p-buttongroup-vertical .p-button[data-v-cb8f9a1a] {
+.p-buttongroup-vertical .p-button[data-v-27a9500c] {
   margin: 0;
   border-radius: 0;
 }
 
-.node-tooltip[data-v-46859edf] {
+.node-tooltip[data-v-f03142eb] {
   background: var(--comfy-input-bg);
   border-radius: 5px;
   box-shadow: 0 0 5px rgba(0, 0, 0, 0.4);
@@ -133,13 +132,11 @@
   border-right: 4px solid var(--p-button-text-primary-color);
 }
 
-.side-tool-bar-container[data-v-33cac83a] {
+.side-tool-bar-container[data-v-04875455] {
   display: flex;
   flex-direction: column;
   align-items: center;
 
-  pointer-events: auto;
-
   width: var(--sidebar-width);
   height: 100%;
 
@@ -150,16 +147,16 @@
   --sidebar-width: 4rem;
   --sidebar-icon-size: 1.5rem;
 }
-.side-tool-bar-container.small-sidebar[data-v-33cac83a] {
+.side-tool-bar-container.small-sidebar[data-v-04875455] {
   --sidebar-width: 2.5rem;
   --sidebar-icon-size: 1rem;
 }
-.side-tool-bar-end[data-v-33cac83a] {
+.side-tool-bar-end[data-v-04875455] {
   align-self: flex-end;
   margin-top: auto;
 }
 
-.status-indicator[data-v-8d011a31] {
+.status-indicator[data-v-fd6ae3af] {
   position: absolute;
   font-weight: 700;
   font-size: 1.5rem;
@@ -221,7 +218,7 @@
   border-radius: 0px
 }
 
-[data-v-38831d8e] .workflow-tabs {
+[data-v-6ab68035] .workflow-tabs {
   background-color: var(--comfy-menu-bg);
 }
 
@@ -235,31 +232,36 @@
   border-bottom-right-radius: 0;
 }
 
-.actionbar[data-v-915e5456] {
+.actionbar[data-v-ebd56d51] {
   pointer-events: all;
   position: fixed;
   z-index: 1000;
 }
-.actionbar.is-docked[data-v-915e5456] {
+.actionbar.is-docked[data-v-ebd56d51] {
   position: static;
   border-style: none;
   background-color: transparent;
   padding: 0px;
 }
-.actionbar.is-dragging[data-v-915e5456] {
+.actionbar.is-dragging[data-v-ebd56d51] {
   -webkit-user-select: none;
      -moz-user-select: none;
           user-select: none;
 }
-[data-v-915e5456] .p-panel-content {
+[data-v-ebd56d51] .p-panel-content {
   padding: 0.25rem;
 }
-.is-docked[data-v-915e5456] .p-panel-content {
+.is-docked[data-v-ebd56d51] .p-panel-content {
   padding: 0px;
 }
-[data-v-915e5456] .p-panel-header {
+[data-v-ebd56d51] .p-panel-header {
   display: none;
 }
+.drag-handle[data-v-ebd56d51] {
+  height: -moz-max-content;
+  height: max-content;
+  width: 0.75rem;
+}
 
 .top-menubar[data-v-56df69d2] .p-menubar-item-link svg {
   display: none;
@@ -275,7 +277,11 @@
   border-style: solid;
 }
 
+<<<<<<<< HEAD:comfy/web/assets/GraphView-BL5xAPb-.css
 .comfyui-menu[data-v-929e7543] {
+========
+.comfyui-menu[data-v-68d3b5b9] {
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/GraphView-Bo28XDd0.css
   width: 100vw;
   height: var(--comfy-topbar-height);
   background: var(--comfy-menu-bg);
@@ -288,6 +294,7 @@
   order: 0;
   grid-column: 1/-1;
 }
+<<<<<<<< HEAD:comfy/web/assets/GraphView-BL5xAPb-.css
 .comfyui-menu.dropzone[data-v-929e7543] {
   background: var(--p-highlight-background);
 }
@@ -298,9 +305,96 @@
   line-height: revert;
 }
 .comfyui-logo[data-v-929e7543] {
+========
+.comfyui-menu.dropzone[data-v-68d3b5b9] {
+  background: var(--p-highlight-background);
+}
+.comfyui-menu.dropzone-active[data-v-68d3b5b9] {
+  background: var(--p-highlight-background-focus);
+}
+[data-v-68d3b5b9] .p-menubar-item-label {
+  line-height: revert;
+}
+.comfyui-logo[data-v-68d3b5b9] {
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/GraphView-Bo28XDd0.css
   font-size: 1.2em;
   -webkit-user-select: none;
      -moz-user-select: none;
           user-select: none;
   cursor: default;
 }
+
+.comfyui-body[data-v-e89d9273] {
+  grid-template-columns: auto 1fr auto;
+  grid-template-rows: auto 1fr auto;
+}
+
+/**
+  +------------------+------------------+------------------+
+  |                                                        |
+  |  .comfyui-body-                                        |
+  |       top                                              |
+  | (spans all cols)                                       |
+  |                                                        |
+  +------------------+------------------+------------------+
+  |                  |                  |                  |
+  | .comfyui-body-   |   #graph-canvas  | .comfyui-body-   |
+  |      left        |                  |      right       |
+  |                  |                  |                  |
+  |                  |                  |                  |
+  +------------------+------------------+------------------+
+  |                                                        |
+  |  .comfyui-body-                                        |
+  |      bottom                                            |
+  | (spans all cols)                                       |
+  |                                                        |
+  +------------------+------------------+------------------+
+*/
+.comfyui-body-top[data-v-e89d9273] {
+  order: -5;
+  /* Span across all columns */
+  grid-column: 1/-1;
+  /* Position at the first row */
+  grid-row: 1;
+  /* Top menu bar dropdown needs to be above of graph canvas splitter overlay which is z-index: 999 */
+  /* Top menu bar z-index needs to be higher than bottom menu bar z-index as by default
+  pysssss's image feed is located at body-bottom, and it can overlap with the queue button, which
+  is located in body-top. */
+  z-index: 1001;
+  display: flex;
+  flex-direction: column;
+}
+.comfyui-body-left[data-v-e89d9273] {
+  order: -4;
+  /* Position in the first column */
+  grid-column: 1;
+  /* Position below the top element */
+  grid-row: 2;
+  z-index: 10;
+  display: flex;
+}
+.graph-canvas-container[data-v-e89d9273] {
+  width: 100%;
+  height: 100%;
+  order: -3;
+  grid-column: 2;
+  grid-row: 2;
+  position: relative;
+  overflow: hidden;
+}
+.comfyui-body-right[data-v-e89d9273] {
+  order: -2;
+  z-index: 10;
+  grid-column: 3;
+  grid-row: 2;
+}
+.comfyui-body-bottom[data-v-e89d9273] {
+  order: 4;
+  /* Span across all columns */
+  grid-column: 1/-1;
+  grid-row: 3;
+  /* Bottom menu bar dropdown needs to be above of graph canvas splitter overlay which is z-index: 999 */
+  z-index: 1000;
+  display: flex;
+  flex-direction: column;
+}

comfy/web/assets/InstallView-DW9xwU_F.js

@@ -1,12 +1,13 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
-import { d as defineComponent, U as ref, bm as useModel, o as openBlock, f as createElementBlock, m as createBaseVNode, E as toDisplayString, k as createVNode, j as unref, bn as script, bh as script$1, ar as withModifiers, z as withCtx, ab as script$2, K as useI18n, c as computed, ai as normalizeClass, B as createCommentVNode, a4 as script$3, a7 as createTextVNode, b5 as electronAPI, _ as _export_sfc, p as onMounted, r as resolveDirective, bg as script$4, i as withDirectives, bo as script$5, bp as script$6, l as script$7, y as createBlock, bj as script$8, bq as MigrationItems, w as watchEffect, F as Fragment, D as renderList, br as script$9, be as useRouter, ag as toRaw } from "./index-BsGgXmrT.js";
-import { s as script$a, a as script$b, b as script$c, c as script$d, d as script$e } from "./index-DC_-jkme.js";
-import { _ as _sfc_main$5 } from "./BaseViewTemplate-DDUNNAbV.js";
-const _hoisted_1$4 = { class: "flex flex-col gap-6 w-[600px]" };
-const _hoisted_2$4 = { class: "flex flex-col gap-4" };
-const _hoisted_3$4 = { class: "text-2xl font-semibold text-neutral-100" };
-const _hoisted_4$4 = { class: "text-neutral-400 my-0" };
+import { d as defineComponent, T as ref, bq as useModel, o as openBlock, f as createElementBlock, m as createBaseVNode, E as toDisplayString, k as createVNode, j as unref, br as script, bl as script$1, as as withModifiers, z as withCtx, ac as script$2, I as useI18n, c as computed, aj as normalizeClass, B as createCommentVNode, a5 as script$3, a8 as createTextVNode, b9 as electronAPI, _ as _export_sfc, p as onMounted, r as resolveDirective, bk as script$4, i as withDirectives, bs as script$5, bt as script$6, l as script$7, y as createBlock, bn as script$8, bu as MigrationItems, w as watchEffect, F as Fragment, D as renderList, bv as script$9, bw as mergeModels, bx as ValidationState, X as normalizeI18nKey, N as watch, by as checkMirrorReachable, bz as _sfc_main$7, bA as isInChina, bB as mergeValidationStates, bg as t, b3 as script$a, bC as CUDA_TORCH_URL, bD as NIGHTLY_CPU_TORCH_URL, bi as useRouter, ah as toRaw } from "./index-Bv0b06LE.js";
+import { s as script$b, a as script$c, b as script$d, c as script$e, d as script$f } from "./index-SeIZOWJp.js";
+import { P as PYTHON_MIRROR, a as PYPI_MIRROR } from "./uvMirrors-B-HKMf6X.js";
+import { _ as _sfc_main$8 } from "./BaseViewTemplate-BTbuZf5t.js";
+const _hoisted_1$5 = { class: "flex flex-col gap-6 w-[600px]" };
+const _hoisted_2$5 = { class: "flex flex-col gap-4" };
+const _hoisted_3$5 = { class: "text-2xl font-semibold text-neutral-100" };
+const _hoisted_4$5 = { class: "text-neutral-400 my-0" };
 const _hoisted_5$3 = { class: "flex flex-col bg-neutral-800 p-4 rounded-lg" };
 const _hoisted_6$3 = { class: "flex items-center gap-4" };
 const _hoisted_7$3 = { class: "flex-1" };
@@ -27,7 +28,7 @@ const _hoisted_20 = {
   target: "_blank",
   class: "text-blue-400 hover:text-blue-300 underline"
 };
-const _sfc_main$4 = /* @__PURE__ */ defineComponent({
+const _sfc_main$6 = /* @__PURE__ */ defineComponent({
   __name: "DesktopSettingsConfiguration",
   props: {
     "autoUpdate": { type: Boolean, ...{ required: true } },
@@ -44,10 +45,10 @@ const _sfc_main$4 = /* @__PURE__ */ defineComponent({
       showDialog.value = true;
     }, "showMetricsInfo");
     return (_ctx, _cache) => {
-      return openBlock(), createElementBlock("div", _hoisted_1$4, [
-        createBaseVNode("div", _hoisted_2$4, [
-          createBaseVNode("h2", _hoisted_3$4, toDisplayString(_ctx.$t("install.desktopAppSettings")), 1),
-          createBaseVNode("p", _hoisted_4$4, toDisplayString(_ctx.$t("install.desktopAppSettingsDescription")), 1)
+      return openBlock(), createElementBlock("div", _hoisted_1$5, [
+        createBaseVNode("div", _hoisted_2$5, [
+          createBaseVNode("h2", _hoisted_3$5, toDisplayString(_ctx.$t("install.desktopAppSettings")), 1),
+          createBaseVNode("p", _hoisted_4$5, toDisplayString(_ctx.$t("install.desktopAppSettingsDescription")), 1)
         ]),
         createBaseVNode("div", _hoisted_5$3, [
           createBaseVNode("div", _hoisted_6$3, [
@@ -122,10 +123,10 @@ const _sfc_main$4 = /* @__PURE__ */ defineComponent({
 const _imports_0 = "" + new URL("images/nvidia-logo.svg", import.meta.url).href;
 const _imports_1 = "" + new URL("images/apple-mps-logo.png", import.meta.url).href;
 const _imports_2 = "" + new URL("images/manual-configuration.svg", import.meta.url).href;
-const _hoisted_1$3 = { class: "flex flex-col gap-6 w-[600px] h-[30rem] select-none" };
-const _hoisted_2$3 = { class: "grow flex flex-col gap-4 text-neutral-300" };
-const _hoisted_3$3 = { class: "text-2xl font-semibold text-neutral-100" };
-const _hoisted_4$3 = { class: "m-1 text-neutral-400" };
+const _hoisted_1$4 = { class: "flex flex-col gap-6 w-[600px] h-[30rem] select-none" };
+const _hoisted_2$4 = { class: "grow flex flex-col gap-4 text-neutral-300" };
+const _hoisted_3$4 = { class: "text-2xl font-semibold text-neutral-100" };
+const _hoisted_4$4 = { class: "m-1 text-neutral-400" };
 const _hoisted_5$2 = {
   key: 0,
   class: "m-1"
@@ -146,7 +147,7 @@ const _hoisted_12$2 = {
   for: "cpu-mode",
   class: "select-none"
 };
-const _sfc_main$3 = /* @__PURE__ */ defineComponent({
+const _sfc_main$5 = /* @__PURE__ */ defineComponent({
   __name: "GpuPicker",
   props: {
     "device": {
@@ -156,7 +157,7 @@ const _sfc_main$3 = /* @__PURE__ */ defineComponent({
   },
   emits: ["update:device"],
   setup(__props) {
-    const { t } = useI18n();
+    const { t: t2 } = useI18n();
     const cpuMode = computed({
       get: /* @__PURE__ */ __name(() => selected.value === "cpu", "get"),
       set: /* @__PURE__ */ __name((value) => {
@@ -171,10 +172,10 @@ const _sfc_main$3 = /* @__PURE__ */ defineComponent({
       selected.value = newValue;
     }, "pickGpu");
     return (_ctx, _cache) => {
-      return openBlock(), createElementBlock("div", _hoisted_1$3, [
-        createBaseVNode("div", _hoisted_2$3, [
-          createBaseVNode("h2", _hoisted_3$3, toDisplayString(_ctx.$t("install.gpuSelection.selectGpu")), 1),
-          createBaseVNode("p", _hoisted_4$3, toDisplayString(_ctx.$t("install.gpuSelection.selectGpuDescription")) + ": ", 1),
+      return openBlock(), createElementBlock("div", _hoisted_1$4, [
+        createBaseVNode("div", _hoisted_2$4, [
+          createBaseVNode("h2", _hoisted_3$4, toDisplayString(_ctx.$t("install.gpuSelection.selectGpu")), 1),
+          createBaseVNode("p", _hoisted_4$4, toDisplayString(_ctx.$t("install.gpuSelection.selectGpuDescription")) + ": ", 1),
           createBaseVNode("div", {
             class: normalizeClass(["flex gap-2 text-center transition-opacity", { selected: selected.value }])
           }, [
@@ -240,7 +241,7 @@ const _sfc_main$3 = /* @__PURE__ */ defineComponent({
               createVNode(unref(script$3), {
                 icon: "pi pi-exclamation-triangle",
                 severity: "warn",
-                value: unref(t)("icon.exclamation-triangle")
+                value: unref(t2)("icon.exclamation-triangle")
               }, null, 8, ["value"]),
               createTextVNode(" " + toDisplayString(_ctx.$t("install.gpuSelection.customSkipsPython")), 1)
             ]),
@@ -258,7 +259,7 @@ const _sfc_main$3 = /* @__PURE__ */ defineComponent({
               createVNode(unref(script$3), {
                 icon: "pi pi-exclamation-triangle",
                 severity: "warn",
-                value: unref(t)("icon.exclamation-triangle")
+                value: unref(t2)("icon.exclamation-triangle")
               }, null, 8, ["value"]),
               createTextVNode(" " + toDisplayString(_ctx.$t("install.gpuSelection.cpuModeDescription")), 1)
             ]),
@@ -282,11 +283,11 @@ const _sfc_main$3 = /* @__PURE__ */ defineComponent({
     };
   }
 });
-const GpuPicker = /* @__PURE__ */ _export_sfc(_sfc_main$3, [["__scopeId", "data-v-79125ff6"]]);
-const _hoisted_1$2 = { class: "flex flex-col gap-6 w-[600px]" };
-const _hoisted_2$2 = { class: "flex flex-col gap-4" };
-const _hoisted_3$2 = { class: "text-2xl font-semibold text-neutral-100" };
-const _hoisted_4$2 = { class: "text-neutral-400 my-0" };
+const GpuPicker = /* @__PURE__ */ _export_sfc(_sfc_main$5, [["__scopeId", "data-v-79125ff6"]]);
+const _hoisted_1$3 = { class: "flex flex-col gap-6 w-[600px]" };
+const _hoisted_2$3 = { class: "flex flex-col gap-4" };
+const _hoisted_3$3 = { class: "text-2xl font-semibold text-neutral-100" };
+const _hoisted_4$3 = { class: "text-neutral-400 my-0" };
 const _hoisted_5$1 = { class: "flex gap-2" };
 const _hoisted_6$1 = { class: "bg-neutral-800 p-4 rounded-lg" };
 const _hoisted_7$1 = { class: "text-lg font-medium mt-0 mb-3 text-neutral-100" };
@@ -297,7 +298,7 @@ const _hoisted_11$1 = { class: "pi pi-info-circle" };
 const _hoisted_12$1 = { class: "flex items-center gap-2" };
 const _hoisted_13 = { class: "text-neutral-200" };
 const _hoisted_14 = { class: "pi pi-info-circle" };
-const _sfc_main$2 = /* @__PURE__ */ defineComponent({
+const _sfc_main$4 = /* @__PURE__ */ defineComponent({
   __name: "InstallLocationPicker",
   props: {
     "installPath": { required: true },
@@ -307,12 +308,13 @@ const _sfc_main$2 = /* @__PURE__ */ defineComponent({
   },
   emits: ["update:installPath", "update:pathError"],
   setup(__props) {
-    const { t } = useI18n();
+    const { t: t2 } = useI18n();
     const installPath = useModel(__props, "installPath");
     const pathError = useModel(__props, "pathError");
     const pathExists = ref(false);
     const appData = ref("");
     const appPath = ref("");
+    const inputTouched = ref(false);
     const electron = electronAPI();
     onMounted(async () => {
       const paths = await electron.getSystemPaths();
@@ -328,19 +330,19 @@ const _sfc_main$2 = /* @__PURE__ */ defineComponent({
         const validation = await electron.validateInstallPath(path);
         if (!validation.isValid) {
           const errors = [];
-          if (validation.cannotWrite) errors.push(t("install.cannotWrite"));
+          if (validation.cannotWrite) errors.push(t2("install.cannotWrite"));
           if (validation.freeSpace < validation.requiredSpace) {
             const requiredGB = validation.requiredSpace / 1024 / 1024 / 1024;
-            errors.push(`${t("install.insufficientFreeSpace")}: ${requiredGB} GB`);
+            errors.push(`${t2("install.insufficientFreeSpace")}: ${requiredGB} GB`);
           }
-          if (validation.parentMissing) errors.push(t("install.parentMissing"));
+          if (validation.parentMissing) errors.push(t2("install.parentMissing"));
           if (validation.error)
-            errors.push(`${t("install.unhandledError")}: ${validation.error}`);
+            errors.push(`${t2("install.unhandledError")}: ${validation.error}`);
           pathError.value = errors.join("\n");
         }
         if (validation.exists) pathExists.value = true;
       } catch (error) {
-        pathError.value = t("install.pathValidationFailed");
+        pathError.value = t2("install.pathValidationFailed");
       }
     }, "validatePath");
     const browsePath = /* @__PURE__ */ __name(async () => {
@@ -351,15 +353,22 @@ const _sfc_main$2 = /* @__PURE__ */ defineComponent({
           await validatePath(result);
         }
       } catch (error) {
-        pathError.value = t("install.failedToSelectDirectory");
+        pathError.value = t2("install.failedToSelectDirectory");
       }
     }, "browsePath");
+    const onFocus = /* @__PURE__ */ __name(() => {
+      if (!inputTouched.value) {
+        inputTouched.value = true;
+        return;
+      }
+      validatePath(installPath.value);
+    }, "onFocus");
     return (_ctx, _cache) => {
       const _directive_tooltip = resolveDirective("tooltip");
-      return openBlock(), createElementBlock("div", _hoisted_1$2, [
-        createBaseVNode("div", _hoisted_2$2, [
-          createBaseVNode("h2", _hoisted_3$2, toDisplayString(_ctx.$t("install.chooseInstallationLocation")), 1),
-          createBaseVNode("p", _hoisted_4$2, toDisplayString(_ctx.$t("install.installLocationDescription")), 1),
+      return openBlock(), createElementBlock("div", _hoisted_1$3, [
+        createBaseVNode("div", _hoisted_2$3, [
+          createBaseVNode("h2", _hoisted_3$3, toDisplayString(_ctx.$t("install.chooseInstallationLocation")), 1),
+          createBaseVNode("p", _hoisted_4$3, toDisplayString(_ctx.$t("install.installLocationDescription")), 1),
           createBaseVNode("div", _hoisted_5$1, [
             createVNode(unref(script$6), { class: "flex-1" }, {
               default: withCtx(() => [
@@ -369,10 +378,16 @@ const _sfc_main$2 = /* @__PURE__ */ defineComponent({
                     _cache[0] || (_cache[0] = ($event) => installPath.value = $event),
                     validatePath
                   ],
-                  class: normalizeClass(["w-full", { "p-invalid": pathError.value }])
+                  class: normalizeClass(["w-full", { "p-invalid": pathError.value }]),
+                  onFocus
                 }, null, 8, ["modelValue", "class"]),
                 withDirectives(createVNode(unref(script$5), { class: "pi pi-info-circle" }, null, 512), [
-                  [_directive_tooltip, _ctx.$t("install.installLocationTooltip")]
+                  [
+                    _directive_tooltip,
+                    _ctx.$t("install.installLocationTooltip"),
+                    void 0,
+                    { top: true }
+                  ]
                 ])
               ]),
               _: 1
@@ -428,10 +443,10 @@ const _sfc_main$2 = /* @__PURE__ */ defineComponent({
     };
   }
 });
-const _hoisted_1$1 = { class: "flex flex-col gap-6 w-[600px]" };
-const _hoisted_2$1 = { class: "flex flex-col gap-4" };
-const _hoisted_3$1 = { class: "text-2xl font-semibold text-neutral-100" };
-const _hoisted_4$1 = { class: "text-neutral-400 my-0" };
+const _hoisted_1$2 = { class: "flex flex-col gap-6 w-[600px]" };
+const _hoisted_2$2 = { class: "flex flex-col gap-4" };
+const _hoisted_3$2 = { class: "text-2xl font-semibold text-neutral-100" };
+const _hoisted_4$2 = { class: "text-neutral-400 my-0" };
 const _hoisted_5 = { class: "flex gap-2" };
 const _hoisted_6 = {
   key: 0,
@@ -446,7 +461,7 @@ const _hoisted_12 = {
   key: 1,
   class: "text-neutral-400 italic"
 };
-const _sfc_main$1 = /* @__PURE__ */ defineComponent({
+const _sfc_main$3 = /* @__PURE__ */ defineComponent({
   __name: "MigrationPicker",
   props: {
     "sourcePath": { required: false },
@@ -458,7 +473,7 @@ const _sfc_main$1 = /* @__PURE__ */ defineComponent({
   },
   emits: ["update:sourcePath", "update:migrationItemIds"],
   setup(__props) {
-    const { t } = useI18n();
+    const { t: t2 } = useI18n();
     const electron = electronAPI();
     const sourcePath = useModel(__props, "sourcePath");
     const migrationItemIds = useModel(__props, "migrationItemIds");
@@ -483,7 +498,7 @@ const _sfc_main$1 = /* @__PURE__ */ defineComponent({
         if (!validation.isValid) pathError.value = validation.error;
       } catch (error) {
         console.error(error);
-        pathError.value = t("install.pathValidationFailed");
+        pathError.value = t2("install.pathValidationFailed");
       }
     }, "validateSource");
     const browsePath = /* @__PURE__ */ __name(async () => {
@@ -495,17 +510,17 @@ const _sfc_main$1 = /* @__PURE__ */ defineComponent({
         }
       } catch (error) {
         console.error(error);
-        pathError.value = t("install.failedToSelectDirectory");
+        pathError.value = t2("install.failedToSelectDirectory");
       }
     }, "browsePath");
     watchEffect(() => {
       migrationItemIds.value = migrationItems.value.filter((item) => item.selected).map((item) => item.id);
     });
     return (_ctx, _cache) => {
-      return openBlock(), createElementBlock("div", _hoisted_1$1, [
-        createBaseVNode("div", _hoisted_2$1, [
-          createBaseVNode("h2", _hoisted_3$1, toDisplayString(_ctx.$t("install.migrateFromExistingInstallation")), 1),
-          createBaseVNode("p", _hoisted_4$1, toDisplayString(_ctx.$t("install.migrationSourcePathDescription")), 1),
+      return openBlock(), createElementBlock("div", _hoisted_1$2, [
+        createBaseVNode("div", _hoisted_2$2, [
+          createBaseVNode("h2", _hoisted_3$2, toDisplayString(_ctx.$t("install.migrateFromExistingInstallation")), 1),
+          createBaseVNode("p", _hoisted_4$2, toDisplayString(_ctx.$t("install.migrationSourcePathDescription")), 1),
           createBaseVNode("div", _hoisted_5, [
             createVNode(unref(script$4), {
               modelValue: sourcePath.value,
@@ -564,10 +579,170 @@ const _sfc_main$1 = /* @__PURE__ */ defineComponent({
     };
   }
 });
+const _hoisted_1$1 = { class: "flex flex-col items-center gap-4" };
+const _hoisted_2$1 = { class: "w-full" };
+const _hoisted_3$1 = { class: "text-lg font-medium text-neutral-100" };
+const _hoisted_4$1 = { class: "text-sm text-neutral-400 mt-1" };
+const _sfc_main$2 = /* @__PURE__ */ defineComponent({
+  __name: "MirrorItem",
+  props: /* @__PURE__ */ mergeModels({
+    item: {}
+  }, {
+    "modelValue": { required: true },
+    "modelModifiers": {}
+  }),
+  emits: /* @__PURE__ */ mergeModels(["state-change"], ["update:modelValue"]),
+  setup(__props, { emit: __emit }) {
+    const emit = __emit;
+    const modelValue = useModel(__props, "modelValue");
+    const validationState = ref(ValidationState.IDLE);
+    const normalizedSettingId = computed(() => {
+      return normalizeI18nKey(__props.item.settingId);
+    });
+    onMounted(() => {
+      modelValue.value = __props.item.mirror;
+    });
+    watch(validationState, (newState) => {
+      emit("state-change", newState);
+      if (newState === ValidationState.INVALID && modelValue.value === __props.item.mirror) {
+        modelValue.value = __props.item.fallbackMirror;
+      }
+    });
+    return (_ctx, _cache) => {
+      return openBlock(), createElementBlock("div", _hoisted_1$1, [
+        createBaseVNode("div", _hoisted_2$1, [
+          createBaseVNode("h3", _hoisted_3$1, toDisplayString(_ctx.$t(`settings.${normalizedSettingId.value}.name`)), 1),
+          createBaseVNode("p", _hoisted_4$1, toDisplayString(_ctx.$t(`settings.${normalizedSettingId.value}.tooltip`)), 1)
+        ]),
+        createVNode(_sfc_main$7, {
+          modelValue: modelValue.value,
+          "onUpdate:modelValue": _cache[0] || (_cache[0] = ($event) => modelValue.value = $event),
+          "validate-url-fn": /* @__PURE__ */ __name((mirror) => unref(checkMirrorReachable)(mirror + (_ctx.item.validationPathSuffix ?? "")), "validate-url-fn"),
+          onStateChange: _cache[1] || (_cache[1] = ($event) => validationState.value = $event)
+        }, null, 8, ["modelValue", "validate-url-fn"])
+      ]);
+    };
+  }
+});
+const _sfc_main$1 = /* @__PURE__ */ defineComponent({
+  __name: "MirrorsConfiguration",
+  props: /* @__PURE__ */ mergeModels({
+    device: {}
+  }, {
+    "pythonMirror": { required: true },
+    "pythonMirrorModifiers": {},
+    "pypiMirror": { required: true },
+    "pypiMirrorModifiers": {},
+    "torchMirror": { required: true },
+    "torchMirrorModifiers": {}
+  }),
+  emits: ["update:pythonMirror", "update:pypiMirror", "update:torchMirror"],
+  setup(__props) {
+    const showMirrorInputs = ref(false);
+    const pythonMirror = useModel(__props, "pythonMirror");
+    const pypiMirror = useModel(__props, "pypiMirror");
+    const torchMirror = useModel(__props, "torchMirror");
+    const getTorchMirrorItem = /* @__PURE__ */ __name((device) => {
+      const settingId = "Comfy-Desktop.UV.TorchInstallMirror";
+      switch (device) {
+        case "mps":
+          return {
+            settingId,
+            mirror: NIGHTLY_CPU_TORCH_URL,
+            fallbackMirror: NIGHTLY_CPU_TORCH_URL
+          };
+        case "nvidia":
+          return {
+            settingId,
+            mirror: CUDA_TORCH_URL,
+            fallbackMirror: CUDA_TORCH_URL
+          };
+        case "cpu":
+        default:
+          return {
+            settingId,
+            mirror: PYPI_MIRROR.mirror,
+            fallbackMirror: PYPI_MIRROR.fallbackMirror
+          };
+      }
+    }, "getTorchMirrorItem");
+    const userIsInChina = ref(false);
+    onMounted(async () => {
+      userIsInChina.value = await isInChina();
+    });
+    const useFallbackMirror = /* @__PURE__ */ __name((mirror) => ({
+      ...mirror,
+      mirror: mirror.fallbackMirror
+    }), "useFallbackMirror");
+    const mirrors = computed(
+      () => [
+        [PYTHON_MIRROR, pythonMirror],
+        [PYPI_MIRROR, pypiMirror],
+        [getTorchMirrorItem(__props.device), torchMirror]
+      ].map(([item, modelValue]) => [
+        userIsInChina.value ? useFallbackMirror(item) : item,
+        modelValue
+      ])
+    );
+    const validationStates = ref(
+      mirrors.value.map(() => ValidationState.IDLE)
+    );
+    const validationState = computed(() => {
+      return mergeValidationStates(validationStates.value);
+    });
+    const validationStateTooltip = computed(() => {
+      switch (validationState.value) {
+        case ValidationState.INVALID:
+          return t("install.settings.mirrorsUnreachable");
+        case ValidationState.VALID:
+          return t("install.settings.mirrorsReachable");
+        default:
+          return t("install.settings.checkingMirrors");
+      }
+    });
+    return (_ctx, _cache) => {
+      const _directive_tooltip = resolveDirective("tooltip");
+      return openBlock(), createBlock(unref(script$a), {
+        header: _ctx.$t("install.settings.mirrorSettings"),
+        toggleable: "",
+        collapsed: !showMirrorInputs.value,
+        "pt:root": "bg-neutral-800 border-none w-[600px]"
+      }, {
+        icons: withCtx(() => [
+          withDirectives(createBaseVNode("i", {
+            class: normalizeClass({
+              "pi pi-spin pi-spinner text-neutral-400": validationState.value === unref(ValidationState).LOADING,
+              "pi pi-check text-green-500": validationState.value === unref(ValidationState).VALID,
+              "pi pi-times text-red-500": validationState.value === unref(ValidationState).INVALID
+            })
+          }, null, 2), [
+            [_directive_tooltip, validationStateTooltip.value]
+          ])
+        ]),
+        default: withCtx(() => [
+          (openBlock(true), createElementBlock(Fragment, null, renderList(mirrors.value, ([item, modelValue], index) => {
+            return openBlock(), createElementBlock(Fragment, {
+              key: item.settingId + item.mirror
+            }, [
+              index > 0 ? (openBlock(), createBlock(unref(script$1), { key: 0 })) : createCommentVNode("", true),
+              createVNode(_sfc_main$2, {
+                item,
+                modelValue: modelValue.value,
+                "onUpdate:modelValue": /* @__PURE__ */ __name(($event) => modelValue.value = $event, "onUpdate:modelValue"),
+                onStateChange: /* @__PURE__ */ __name(($event) => validationStates.value[index] = $event, "onStateChange")
+              }, null, 8, ["item", "modelValue", "onUpdate:modelValue", "onStateChange"])
+            ], 64);
+          }), 128))
+        ]),
+        _: 1
+      }, 8, ["header", "collapsed"]);
+    };
+  }
+});
 const _hoisted_1 = { class: "flex pt-6 justify-end" };
 const _hoisted_2 = { class: "flex pt-6 justify-between" };
 const _hoisted_3 = { class: "flex pt-6 justify-between" };
-const _hoisted_4 = { class: "flex pt-6 justify-between" };
+const _hoisted_4 = { class: "flex mt-6 justify-between" };
 const _sfc_main = /* @__PURE__ */ defineComponent({
   __name: "InstallView",
   setup(__props) {
@@ -578,6 +753,9 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
     const migrationItemIds = ref([]);
     const autoUpdate = ref(true);
     const allowMetrics = ref(true);
+    const pythonMirror = ref("");
+    const pypiMirror = ref("");
+    const torchMirror = ref("");
     const highestStep = ref(0);
     const handleStepChange = /* @__PURE__ */ __name((value) => {
       setHighestStep(value);
@@ -600,6 +778,9 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
         allowMetrics: allowMetrics.value,
         migrationSourcePath: migrationSourcePath.value,
         migrationItemIds: toRaw(migrationItemIds.value),
+        pythonMirror: pythonMirror.value,
+        pypiMirror: pypiMirror.value,
+        torchMirror: torchMirror.value,
         device: device.value
       };
       electron.installComfyUI(options);
@@ -618,23 +799,23 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
       });
     });
     return (_ctx, _cache) => {
-      return openBlock(), createBlock(_sfc_main$5, { dark: "" }, {
+      return openBlock(), createBlock(_sfc_main$8, { dark: "" }, {
         default: withCtx(() => [
-          createVNode(unref(script$e), {
+          createVNode(unref(script$f), {
             class: "h-full p-8 2xl:p-16",
             value: "0",
             "onUpdate:value": handleStepChange
           }, {
             default: withCtx(() => [
-              createVNode(unref(script$a), { class: "select-none" }, {
+              createVNode(unref(script$b), { class: "select-none" }, {
                 default: withCtx(() => [
-                  createVNode(unref(script$b), { value: "0" }, {
+                  createVNode(unref(script$c), { value: "0" }, {
                     default: withCtx(() => [
                       createTextVNode(toDisplayString(_ctx.$t("install.gpu")), 1)
                     ]),
                     _: 1
                   }),
-                  createVNode(unref(script$b), {
+                  createVNode(unref(script$c), {
                     value: "1",
                     disabled: noGpu.value
                   }, {
@@ -643,7 +824,7 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
                     ]),
                     _: 1
                   }, 8, ["disabled"]),
-                  createVNode(unref(script$b), {
+                  createVNode(unref(script$c), {
                     value: "2",
                     disabled: noGpu.value || hasError.value || highestStep.value < 1
                   }, {
@@ -652,7 +833,7 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
                     ]),
                     _: 1
                   }, 8, ["disabled"]),
-                  createVNode(unref(script$b), {
+                  createVNode(unref(script$c), {
                     value: "3",
                     disabled: noGpu.value || hasError.value || highestStep.value < 2
                   }, {
@@ -664,9 +845,9 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
                 ]),
                 _: 1
               }),
-              createVNode(unref(script$c), null, {
+              createVNode(unref(script$d), null, {
                 default: withCtx(() => [
-                  createVNode(unref(script$d), { value: "0" }, {
+                  createVNode(unref(script$e), { value: "0" }, {
                     default: withCtx(({ activateCallback }) => [
                       createVNode(GpuPicker, {
                         device: device.value,
@@ -684,9 +865,9 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
                     ]),
                     _: 1
                   }),
-                  createVNode(unref(script$d), { value: "1" }, {
+                  createVNode(unref(script$e), { value: "1" }, {
                     default: withCtx(({ activateCallback }) => [
-                      createVNode(_sfc_main$2, {
+                      createVNode(_sfc_main$4, {
                         installPath: installPath.value,
                         "onUpdate:installPath": _cache[1] || (_cache[1] = ($event) => installPath.value = $event),
                         pathError: pathError.value,
@@ -710,9 +891,9 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
                     ]),
                     _: 1
                   }),
-                  createVNode(unref(script$d), { value: "2" }, {
+                  createVNode(unref(script$e), { value: "2" }, {
                     default: withCtx(({ activateCallback }) => [
-                      createVNode(_sfc_main$1, {
+                      createVNode(_sfc_main$3, {
                         sourcePath: migrationSourcePath.value,
                         "onUpdate:sourcePath": _cache[3] || (_cache[3] = ($event) => migrationSourcePath.value = $event),
                         migrationItemIds: migrationItemIds.value,
@@ -735,14 +916,24 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
                     ]),
                     _: 1
                   }),
-                  createVNode(unref(script$d), { value: "3" }, {
+                  createVNode(unref(script$e), { value: "3" }, {
                     default: withCtx(({ activateCallback }) => [
-                      createVNode(_sfc_main$4, {
+                      createVNode(_sfc_main$6, {
                         autoUpdate: autoUpdate.value,
                         "onUpdate:autoUpdate": _cache[5] || (_cache[5] = ($event) => autoUpdate.value = $event),
                         allowMetrics: allowMetrics.value,
                         "onUpdate:allowMetrics": _cache[6] || (_cache[6] = ($event) => allowMetrics.value = $event)
                       }, null, 8, ["autoUpdate", "allowMetrics"]),
+                      createVNode(_sfc_main$1, {
+                        device: device.value,
+                        pythonMirror: pythonMirror.value,
+                        "onUpdate:pythonMirror": _cache[7] || (_cache[7] = ($event) => pythonMirror.value = $event),
+                        pypiMirror: pypiMirror.value,
+                        "onUpdate:pypiMirror": _cache[8] || (_cache[8] = ($event) => pypiMirror.value = $event),
+                        torchMirror: torchMirror.value,
+                        "onUpdate:torchMirror": _cache[9] || (_cache[9] = ($event) => torchMirror.value = $event),
+                        class: "mt-6"
+                      }, null, 8, ["device", "pythonMirror", "pypiMirror", "torchMirror"]),
                       createBaseVNode("div", _hoisted_4, [
                         createVNode(unref(script$7), {
                           label: _ctx.$t("g.back"),
@@ -755,7 +946,7 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
                           icon: "pi pi-check",
                           iconPos: "right",
                           disabled: hasError.value,
-                          onClick: _cache[7] || (_cache[7] = ($event) => install())
+                          onClick: _cache[10] || (_cache[10] = ($event) => install())
                         }, null, 8, ["label", "disabled"])
                       ])
                     ]),
@@ -773,8 +964,8 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
     };
   }
 });
-const InstallView = /* @__PURE__ */ _export_sfc(_sfc_main, [["__scopeId", "data-v-0a97b0ae"]]);
+const InstallView = /* @__PURE__ */ _export_sfc(_sfc_main, [["__scopeId", "data-v-cd6731d2"]]);
 export {
   InstallView as default
 };
-//# sourceMappingURL=InstallView-C1fnMZKt.js.map
+//# sourceMappingURL=InstallView-DW9xwU_F.js.map

comfy/web/assets/InstallView-DbJ2cGfL.css

@@ -76,6 +76,6 @@ div.selected {
   text-align: center;
 }
 
-[data-v-0a97b0ae] .p-steppanel {
+[data-v-cd6731d2] .p-steppanel {
     background-color: transparent
 }

comfy/web/assets/KeybindingPanel-CDYVPYDp.css

@@ -0,0 +1,8 @@
+
+[data-v-8454e24f] .p-datatable-tbody > tr > td {
+  padding: 0.25rem;
+  min-height: 2rem
+}
+[data-v-8454e24f] .p-datatable-row-selected .actions,[data-v-8454e24f] .p-datatable-selectable-row:hover .actions {
+  visibility: visible
+}

comfy/web/assets/KeybindingPanel-Caa8sD2X.css

@@ -1,8 +0,0 @@
-
-[data-v-2554ab36] .p-datatable-tbody > tr > td {
-  padding: 0.25rem;
-  min-height: 2rem
-}
-[data-v-2554ab36] .p-datatable-row-selected .actions,[data-v-2554ab36] .p-datatable-selectable-row:hover .actions {
-  visibility: visible
-}

comfy/web/assets/KeybindingPanel-oavhFdkz.js

@@ -1,9 +1,16 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/KeybindingPanel-BRfso_Vt.js
 import { d as defineComponent, c as computed, o as openBlock, f as createElementBlock, F as Fragment, D as renderList, k as createVNode, z as withCtx, a7 as createTextVNode, E as toDisplayString, j as unref, a4 as script, B as createCommentVNode, U as ref, df as FilterMatchMode, an as useKeybindingStore, L as useCommandStore, K as useI18n, Y as normalizeI18nKey, w as watchEffect, aR as useToast, r as resolveDirective, y as createBlock, dg as SearchBox, m as createBaseVNode, l as script$2, bg as script$4, ar as withModifiers, bj as script$5, ab as script$6, i as withDirectives, dh as _sfc_main$2, di as KeyComboImpl, dj as KeybindingImpl, _ as _export_sfc } from "./index-BsGgXmrT.js";
 import { g as script$1, h as script$3 } from "./index-Br6dw1F6.js";
 import { u as useKeybindingService } from "./keybindingService-DoUb2RT6.js";
 import "./index-COyiXDAn.js";
+========
+import { d as defineComponent, c as computed, o as openBlock, f as createElementBlock, F as Fragment, D as renderList, k as createVNode, z as withCtx, a8 as createTextVNode, E as toDisplayString, j as unref, a5 as script, B as createCommentVNode, T as ref, dx as FilterMatchMode, ao as useKeybindingStore, J as useCommandStore, I as useI18n, X as normalizeI18nKey, w as watchEffect, aV as useToast, r as resolveDirective, y as createBlock, dy as SearchBox, m as createBaseVNode, l as script$2, bk as script$4, as as withModifiers, bn as script$5, ac as script$6, i as withDirectives, dz as _sfc_main$2, dA as KeyComboImpl, dB as KeybindingImpl, _ as _export_sfc } from "./index-Bv0b06LE.js";
+import { g as script$1, h as script$3 } from "./index-CgMyWf7n.js";
+import { u as useKeybindingService } from "./keybindingService-DyjX-nxF.js";
+import "./index-Dzu9WL4p.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/KeybindingPanel-oavhFdkz.js
 const _hoisted_1$1 = {
   key: 0,
   class: "px-2"
@@ -96,6 +103,16 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
     }
     __name(removeKeybinding, "removeKeybinding");
     function captureKeybinding(event) {
+      if (!event.shiftKey && !event.altKey && !event.ctrlKey && !event.metaKey) {
+        switch (event.key) {
+          case "Escape":
+            cancelEdit();
+            return;
+          case "Enter":
+            saveKeybinding();
+            return;
+        }
+      }
       const keyCombo = KeyComboImpl.fromEvent(event);
       newBindingKeyCombo.value = keyCombo;
     }
@@ -151,7 +168,7 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
             value: commandsData.value,
             selection: selectedCommandData.value,
             "onUpdate:selection": _cache[1] || (_cache[1] = ($event) => selectedCommandData.value = $event),
-            "global-filter-fields": ["id"],
+            "global-filter-fields": ["id", "label"],
             filters: filters.value,
             selectionMode: "single",
             stripedRows: "",
@@ -216,7 +233,7 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
             visible: editDialogVisible.value,
             "onUpdate:visible": _cache[2] || (_cache[2] = ($event) => editDialogVisible.value = $event),
             modal: "",
-            header: currentEditingCommand.value?.id,
+            header: currentEditingCommand.value?.label,
             onHide: cancelEdit
           }, {
             footer: withCtx(() => [
@@ -275,8 +292,12 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
     };
   }
 });
-const KeybindingPanel = /* @__PURE__ */ _export_sfc(_sfc_main, [["__scopeId", "data-v-2554ab36"]]);
+const KeybindingPanel = /* @__PURE__ */ _export_sfc(_sfc_main, [["__scopeId", "data-v-8454e24f"]]);
 export {
   KeybindingPanel as default
 };
+<<<<<<<< HEAD:comfy/web/assets/KeybindingPanel-BRfso_Vt.js
 //# sourceMappingURL=KeybindingPanel-BRfso_Vt.js.map
+========
+//# sourceMappingURL=KeybindingPanel-oavhFdkz.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/KeybindingPanel-oavhFdkz.js

comfy/web/assets/MaintenanceView-DEJCj8SR.css

@@ -0,0 +1,87 @@
+
+.task-card-ok[data-v-c3bd7658] {
+
+  position: absolute;
+
+  right: -1rem;
+
+  bottom: -1rem;
+
+  grid-column: 1 / -1;
+
+  grid-row: 1 / -1;
+
+  --tw-text-opacity: 1;
+
+  color: rgb(150 206 76 / var(--tw-text-opacity));
+
+  opacity: 1;
+
+  transition-property: opacity;
+
+  transition-timing-function: cubic-bezier(0.4, 0, 0.2, 1);
+
+  transition-duration: 150ms;
+
+  font-size: 4rem;
+  text-shadow: 0.25rem 0 0.5rem black;
+  z-index: 10;
+}
+.p-card {
+&[data-v-c3bd7658] {
+
+  transition-property: opacity;
+
+  transition-timing-function: cubic-bezier(0.4, 0, 0.2, 1);
+
+  transition-duration: 150ms;
+
+  --p-card-background: var(--p-button-secondary-background);
+  opacity: 0.9;
+  }
+&.opacity-65[data-v-c3bd7658] {
+    opacity: 0.4;
+}
+&[data-v-c3bd7658]:hover {
+    opacity: 1;
+}
+}
+[data-v-c3bd7658] .p-card-header {
+  z-index: 0;
+}
+[data-v-c3bd7658] .p-card-body {
+  z-index: 1;
+  flex-grow: 1;
+  justify-content: space-between;
+}
+.task-div {
+> i[data-v-c3bd7658] {
+    pointer-events: none;
+}
+&:hover > i[data-v-c3bd7658] {
+    opacity: 0.2;
+}
+}
+
+[data-v-dd50a7dd] .p-tag {
+  --p-tag-gap: 0.375rem;
+}
+.backspan[data-v-dd50a7dd]::before {
+  position: absolute;
+  margin: 0px;
+  color: var(--p-text-muted-color);
+  font-family: 'primeicons';
+  top: -2rem;
+  right: -2rem;
+  speak: none;
+  font-style: normal;
+  font-weight: normal;
+  font-variant: normal;
+  text-transform: none;
+  line-height: 1;
+  display: inline-block;
+  -webkit-font-smoothing: antialiased;
+  opacity: 0.02;
+  font-size: min(14rem, 90vw);
+  z-index: 0;
+}

comfy/web/assets/ManualConfigurationView-DTLyJ3VG.js

@@ -1,7 +1,12 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/ManualConfigurationView-DlH3kpjW.js
 import { d as defineComponent, K as useI18n, U as ref, p as onMounted, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, k as createVNode, j as unref, a4 as script, a$ as script$1, l as script$2, b5 as electronAPI, _ as _export_sfc } from "./index-BsGgXmrT.js";
 import { _ as _sfc_main$1 } from "./BaseViewTemplate-DDUNNAbV.js";
+========
+import { d as defineComponent, I as useI18n, T as ref, p as onMounted, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, k as createVNode, j as unref, a5 as script, b3 as script$1, l as script$2, b9 as electronAPI, _ as _export_sfc } from "./index-Bv0b06LE.js";
+import { _ as _sfc_main$1 } from "./BaseViewTemplate-BTbuZf5t.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/ManualConfigurationView-DTLyJ3VG.js
 const _hoisted_1 = { class: "comfy-installer grow flex flex-col gap-4 text-neutral-300 max-w-110" };
 const _hoisted_2 = { class: "text-2xl font-semibold text-neutral-100" };
 const _hoisted_3 = { class: "m-1 text-neutral-300" };
@@ -71,4 +76,8 @@ const ManualConfigurationView = /* @__PURE__ */ _export_sfc(_sfc_main, [["__scop
 export {
   ManualConfigurationView as default
 };
+<<<<<<<< HEAD:comfy/web/assets/ManualConfigurationView-DlH3kpjW.js
 //# sourceMappingURL=ManualConfigurationView-DlH3kpjW.js.map
+========
+//# sourceMappingURL=ManualConfigurationView-DTLyJ3VG.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/ManualConfigurationView-DTLyJ3VG.js

comfy/web/assets/MetricsConsentView-C80fk2cl.js

@@ -1,7 +1,12 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/MetricsConsentView-BgqqjOyd.js
 import { _ as _sfc_main$1 } from "./BaseViewTemplate-DDUNNAbV.js";
 import { d as defineComponent, aR as useToast, K as useI18n, U as ref, be as useRouter, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, a7 as createTextVNode, k as createVNode, j as unref, bn as script, l as script$1, b5 as electronAPI } from "./index-BsGgXmrT.js";
+========
+import { _ as _sfc_main$1 } from "./BaseViewTemplate-BTbuZf5t.js";
+import { d as defineComponent, aV as useToast, I as useI18n, T as ref, bi as useRouter, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, a8 as createTextVNode, k as createVNode, j as unref, br as script, l as script$1, b9 as electronAPI } from "./index-Bv0b06LE.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/MetricsConsentView-C80fk2cl.js
 const _hoisted_1 = { class: "h-full p-8 2xl:p-16 flex flex-col items-center justify-center" };
 const _hoisted_2 = { class: "bg-neutral-800 rounded-lg shadow-lg p-6 w-full max-w-[600px] flex flex-col gap-6" };
 const _hoisted_3 = { class: "text-3xl font-semibold text-neutral-100" };
@@ -83,4 +88,8 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
 export {
   _sfc_main as default
 };
+<<<<<<<< HEAD:comfy/web/assets/MetricsConsentView-BgqqjOyd.js
 //# sourceMappingURL=MetricsConsentView-BgqqjOyd.js.map
+========
+//# sourceMappingURL=MetricsConsentView-C80fk2cl.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/MetricsConsentView-C80fk2cl.js

comfy/web/assets/NotSupportedView-B78ZVR9Z.js

@@ -1,7 +1,12 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/NotSupportedView-IH8EV0bV.js
 import { d as defineComponent, be as useRouter, r as resolveDirective, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, k as createVNode, j as unref, l as script, i as withDirectives, _ as _export_sfc } from "./index-BsGgXmrT.js";
 import { _ as _sfc_main$1 } from "./BaseViewTemplate-DDUNNAbV.js";
+========
+import { d as defineComponent, bi as useRouter, r as resolveDirective, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, k as createVNode, j as unref, l as script, i as withDirectives, _ as _export_sfc } from "./index-Bv0b06LE.js";
+import { _ as _sfc_main$1 } from "./BaseViewTemplate-BTbuZf5t.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/NotSupportedView-B78ZVR9Z.js
 const _imports_0 = "" + new URL("images/sad_girl.png", import.meta.url).href;
 const _hoisted_1 = { class: "sad-container" };
 const _hoisted_2 = { class: "no-drag sad-text flex items-center" };
@@ -83,4 +88,8 @@ const NotSupportedView = /* @__PURE__ */ _export_sfc(_sfc_main, [["__scopeId", "
 export {
   NotSupportedView as default
 };
+<<<<<<<< HEAD:comfy/web/assets/NotSupportedView-IH8EV0bV.js
 //# sourceMappingURL=NotSupportedView-IH8EV0bV.js.map
+========
+//# sourceMappingURL=NotSupportedView-B78ZVR9Z.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/NotSupportedView-B78ZVR9Z.js

comfy/web/assets/ServerConfigPanel-BYrt6wyr.js

@@ -1,7 +1,12 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/ServerConfigPanel-u0ozNLZ4.js
 import { o as openBlock, f as createElementBlock, m as createBaseVNode, H as markRaw, d as defineComponent, a as useSettingStore, ae as storeToRefs, O as watch, ds as useCopyToClipboard, K as useI18n, y as createBlock, z as withCtx, j as unref, bj as script, E as toDisplayString, D as renderList, F as Fragment, k as createVNode, l as script$1, B as createCommentVNode, bh as script$2, dt as FormItem, dh as _sfc_main$1, b5 as electronAPI } from "./index-BsGgXmrT.js";
 import { u as useServerConfigStore } from "./serverConfigStore-B9riwnSX.js";
+========
+import { o as openBlock, f as createElementBlock, m as createBaseVNode, H as markRaw, d as defineComponent, a as useSettingStore, af as storeToRefs, N as watch, dJ as useCopyToClipboard, I as useI18n, y as createBlock, z as withCtx, j as unref, bn as script, E as toDisplayString, D as renderList, F as Fragment, k as createVNode, l as script$1, B as createCommentVNode, bl as script$2, dK as FormItem, dz as _sfc_main$1, b9 as electronAPI } from "./index-Bv0b06LE.js";
+import { u as useServerConfigStore } from "./serverConfigStore-D2Vr0L0h.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/ServerConfigPanel-BYrt6wyr.js
 const _hoisted_1$1 = {
   viewBox: "0 0 24 24",
   width: "1.2em",
@@ -153,4 +158,8 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
 export {
   _sfc_main as default
 };
+<<<<<<<< HEAD:comfy/web/assets/ServerConfigPanel-u0ozNLZ4.js
 //# sourceMappingURL=ServerConfigPanel-u0ozNLZ4.js.map
+========
+//# sourceMappingURL=ServerConfigPanel-BYrt6wyr.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/ServerConfigPanel-BYrt6wyr.js

comfy/web/assets/ServerStartView-B7TlHxYo.js

@@ -1,7 +1,12 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/ServerStartView-DgywG2so.js
 import { d as defineComponent, K as useI18n, U as ref, bk as ProgressStatus, p as onMounted, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, a7 as createTextVNode, E as toDisplayString, j as unref, f as createElementBlock, B as createCommentVNode, k as createVNode, l as script, i as withDirectives, v as vShow, bl as BaseTerminal, b5 as electronAPI, _ as _export_sfc } from "./index-BsGgXmrT.js";
 import { _ as _sfc_main$1 } from "./BaseViewTemplate-DDUNNAbV.js";
+========
+import { d as defineComponent, I as useI18n, T as ref, bo as ProgressStatus, p as onMounted, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, a8 as createTextVNode, E as toDisplayString, j as unref, f as createElementBlock, B as createCommentVNode, k as createVNode, l as script, i as withDirectives, v as vShow, bp as BaseTerminal, b9 as electronAPI, _ as _export_sfc } from "./index-Bv0b06LE.js";
+import { _ as _sfc_main$1 } from "./BaseViewTemplate-BTbuZf5t.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/ServerStartView-B7TlHxYo.js
 const _hoisted_1 = { class: "flex flex-col w-full h-full items-center" };
 const _hoisted_2 = { class: "text-2xl font-bold" };
 const _hoisted_3 = { key: 0 };
@@ -93,8 +98,12 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
     };
   }
 });
-const ServerStartView = /* @__PURE__ */ _export_sfc(_sfc_main, [["__scopeId", "data-v-4140d62b"]]);
+const ServerStartView = /* @__PURE__ */ _export_sfc(_sfc_main, [["__scopeId", "data-v-e6ba9633"]]);
 export {
   ServerStartView as default
 };
+<<<<<<<< HEAD:comfy/web/assets/ServerStartView-DgywG2so.js
 //# sourceMappingURL=ServerStartView-DgywG2so.js.map
+========
+//# sourceMappingURL=ServerStartView-B7TlHxYo.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/ServerStartView-B7TlHxYo.js

comfy/web/assets/ServerStartView-BZ7uhZHv.css

@@ -1,5 +1,5 @@
 
-[data-v-4140d62b] .xterm-helper-textarea {
+[data-v-e6ba9633] .xterm-helper-textarea {
   /* Hide this as it moves all over when uv is running */
   display: none;
 }

comfy/web/assets/UserSelectView-C703HOyO.js

@@ -1,7 +1,12 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/UserSelectView-DkeVSFwW.js
 import { d as defineComponent, aj as useUserStore, be as useRouter, U as ref, c as computed, p as onMounted, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, k as createVNode, bf as withKeys, j as unref, bg as script, bh as script$1, bi as script$2, bj as script$3, a7 as createTextVNode, B as createCommentVNode, l as script$4 } from "./index-BsGgXmrT.js";
 import { _ as _sfc_main$1 } from "./BaseViewTemplate-DDUNNAbV.js";
+========
+import { d as defineComponent, ak as useUserStore, bi as useRouter, T as ref, c as computed, p as onMounted, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, k as createVNode, bj as withKeys, j as unref, bk as script, bl as script$1, bm as script$2, bn as script$3, a8 as createTextVNode, B as createCommentVNode, l as script$4 } from "./index-Bv0b06LE.js";
+import { _ as _sfc_main$1 } from "./BaseViewTemplate-BTbuZf5t.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/UserSelectView-C703HOyO.js
 const _hoisted_1 = {
   id: "comfy-user-selection",
   class: "min-w-84 relative rounded-lg bg-[var(--comfy-menu-bg)] p-5 px-10 shadow-lg"
@@ -98,4 +103,8 @@ const _sfc_main = /* @__PURE__ */ defineComponent({
 export {
   _sfc_main as default
 };
+<<<<<<<< HEAD:comfy/web/assets/UserSelectView-DkeVSFwW.js
 //# sourceMappingURL=UserSelectView-DkeVSFwW.js.map
+========
+//# sourceMappingURL=UserSelectView-C703HOyO.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/UserSelectView-C703HOyO.js

comfy/web/assets/WelcomeView-DIFvbWc2.js

@@ -1,7 +1,12 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/WelcomeView-CXVMqRFA.js
 import { d as defineComponent, be as useRouter, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, k as createVNode, j as unref, l as script, _ as _export_sfc } from "./index-BsGgXmrT.js";
 import { _ as _sfc_main$1 } from "./BaseViewTemplate-DDUNNAbV.js";
+========
+import { d as defineComponent, bi as useRouter, o as openBlock, y as createBlock, z as withCtx, m as createBaseVNode, E as toDisplayString, k as createVNode, j as unref, l as script, _ as _export_sfc } from "./index-Bv0b06LE.js";
+import { _ as _sfc_main$1 } from "./BaseViewTemplate-BTbuZf5t.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/WelcomeView-DIFvbWc2.js
 const _hoisted_1 = { class: "flex flex-col items-center justify-center gap-8 p-8" };
 const _hoisted_2 = { class: "animated-gradient-text text-glow select-none" };
 const _sfc_main = /* @__PURE__ */ defineComponent({
@@ -36,4 +41,8 @@ const WelcomeView = /* @__PURE__ */ _export_sfc(_sfc_main, [["__scopeId", "data-
 export {
   WelcomeView as default
 };
+<<<<<<<< HEAD:comfy/web/assets/WelcomeView-CXVMqRFA.js
 //# sourceMappingURL=WelcomeView-CXVMqRFA.js.map
+========
+//# sourceMappingURL=WelcomeView-DIFvbWc2.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/WelcomeView-DIFvbWc2.js

comfy/web/assets/index-CBxvvAzM.css

@@ -306,6 +306,7 @@
 .litegraph .dialog .dialog-footer {
   height: 50px;
   padding: 10px;
+  margin: 0;
   border-top: 1px solid #1a1a1a;
 }
 
@@ -442,63 +443,6 @@
   color: black;
 }
 
-.litegraph .subgraph_property {
-  padding: 4px;
-}
-
-.litegraph .subgraph_property:hover {
-  background-color: #333;
-}
-
-.litegraph .subgraph_property.extra {
-  margin-top: 8px;
-}
-
-.litegraph .subgraph_property span.name {
-  font-size: 1.3em;
-  padding-left: 4px;
-}
-
-.litegraph .subgraph_property span.type {
-  opacity: 0.5;
-  margin-right: 20px;
-  padding-left: 4px;
-}
-
-.litegraph .subgraph_property span.label {
-  display: inline-block;
-  width: 60px;
-  padding: 0px 10px;
-}
-
-.litegraph .subgraph_property input {
-  width: 140px;
-  color: #999;
-  background-color: #1a1a1a;
-  border-radius: 4px;
-  border: 0;
-  margin-right: 10px;
-  padding: 4px;
-  padding-left: 10px;
-}
-
-.litegraph .subgraph_property button {
-  background-color: #1c1c1c;
-  color: #aaa;
-  border: 0;
-  border-radius: 2px;
-  padding: 4px 10px;
-  cursor: pointer;
-}
-
-.litegraph .subgraph_property.extra {
-  color: #ccc;
-}
-
-.litegraph .subgraph_property.extra input {
-  background-color: #111;
-}
-
 .litegraph .bullet_icon {
   margin-left: 10px;
   border-radius: 10px;
@@ -661,21 +605,6 @@
 .litegraph .dialog .dialog-content {
   display: block;
 }
-.litegraph .dialog .dialog-content .subgraph_property {
-  padding: 5px;
-}
-.litegraph .dialog .dialog-footer {
-  margin: 0;
-}
-.litegraph .dialog .dialog-footer .subgraph_property {
-  margin-top: 0;
-  display: flex;
-  align-items: center;
-  padding: 5px;
-}
-.litegraph .dialog .dialog-footer .subgraph_property .name {
-  flex: 1;
-}
 .litegraph .graphdialog {
   display: flex;
   align-items: center;
@@ -2110,6 +2039,12 @@
   .-right-4{
     right: -1rem;
   }
+<<<<<<<< HEAD:comfy/web/assets/index-ChXzdVeQ.css
+========
+  .bottom-0{
+    bottom: 0px;
+  }
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/index-CBxvvAzM.css
   .bottom-\[10px\]{
     bottom: 10px;
   }
@@ -2119,6 +2054,15 @@
   .left-0{
     left: 0px;
   }
+  .left-1\/2{
+    left: 50%;
+  }
+  .left-12{
+    left: 3rem;
+  }
+  .left-2{
+    left: 0.5rem;
+  }
   .left-\[-350px\]{
     left: -350px;
   }
@@ -2128,6 +2072,9 @@
   .top-0{
     top: 0px;
   }
+  .top-2{
+    top: 0.5rem;
+  }
   .top-\[50px\]{
     top: 50px;
   }
@@ -2137,6 +2084,9 @@
   .z-10{
     z-index: 10;
   }
+  .z-20{
+    z-index: 20;
+  }
   .z-\[1000\]{
     z-index: 1000;
   }
@@ -2192,6 +2142,10 @@
     margin-top: 1rem;
     margin-bottom: 1rem;
   }
+  .my-8{
+    margin-top: 2rem;
+    margin-bottom: 2rem;
+  }
   .mb-2{
     margin-bottom: 0.5rem;
   }
@@ -2240,6 +2194,9 @@
   .mt-5{
     margin-top: 1.25rem;
   }
+  .mt-6{
+    margin-top: 1.5rem;
+  }
   .block{
     display: block;
   }
@@ -2279,6 +2236,9 @@
   .h-16{
     height: 4rem;
   }
+  .h-48{
+    height: 12rem;
+  }
   .h-6{
     height: 1.5rem;
   }
@@ -2324,6 +2284,9 @@
   .min-h-screen{
     min-height: 100vh;
   }
+  .w-0{
+    width: 0px;
+  }
   .w-1\/2{
     width: 50%;
   }
@@ -2336,12 +2299,21 @@
   .w-16{
     width: 4rem;
   }
+  .w-24{
+    width: 6rem;
+  }
   .w-28{
     width: 7rem;
   }
+  .w-3{
+    width: 0.75rem;
+  }
   .w-3\/12{
     width: 25%;
   }
+  .w-32{
+    width: 8rem;
+  }
   .w-44{
     width: 11rem;
   }
@@ -2451,6 +2423,9 @@
   .cursor-pointer{
     cursor: pointer;
   }
+  .touch-none{
+    touch-action: none;
+  }
   .select-none{
     -webkit-user-select: none;
        -moz-user-select: none;
@@ -2633,7 +2608,11 @@
   }
   .border-neutral-700{
     --tw-border-opacity: 1;
+<<<<<<<< HEAD:comfy/web/assets/index-ChXzdVeQ.css
     border-color: rgb(64 64 64 / var(--tw-border-opacity, 1));
+========
+    border-color: rgb(64 64 64 / var(--tw-border-opacity));
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/index-CBxvvAzM.css
   }
   .bg-\[var\(--comfy-menu-bg\)\]{
     background-color: var(--comfy-menu-bg);
@@ -2886,6 +2865,10 @@
     --tw-text-opacity: 1;
     color: rgb(239 68 68 / var(--tw-text-opacity, 1));
   }
+  .text-white{
+    --tw-text-opacity: 1;
+    color: rgb(255 255 255 / var(--tw-text-opacity));
+  }
   .underline{
     text-decoration-line: underline;
   }
@@ -2967,6 +2950,9 @@
   .duration-100{
     transition-duration: 100ms;
   }
+  .duration-200{
+    transition-duration: 200ms;
+  }
   .duration-300{
     transition-duration: 300ms;
   }
@@ -3025,8 +3011,6 @@ body {
   height: 100vh;
   margin: 0;
   overflow: hidden;
-  grid-template-columns: auto 1fr auto;
-  grid-template-rows: auto 1fr auto;
   background: var(--bg-color) var(--bg-img);
   color: var(--fg-color);
   min-height: -webkit-fill-available;
@@ -3036,87 +3020,6 @@ body {
   font-family: Arial, sans-serif;
 }
 
-/**
-  +------------------+------------------+------------------+
-  |                                                        |
-  |  .comfyui-body-                                        |
-  |       top                                              |
-  | (spans all cols)                                       |
-  |                                                        |
-  +------------------+------------------+------------------+
-  |                  |                  |                  |
-  | .comfyui-body-   |   #graph-canvas  | .comfyui-body-   |
-  |      left        |                  |      right       |
-  |                  |                  |                  |
-  |                  |                  |                  |
-  +------------------+------------------+------------------+
-  |                                                        |
-  |  .comfyui-body-                                        |
-  |      bottom                                            |
-  | (spans all cols)                                       |
-  |                                                        |
-  +------------------+------------------+------------------+
-*/
-
-.comfyui-body-top {
-  order: -5;
-  /* Span across all columns */
-  grid-column: 1/-1;
-  /* Position at the first row */
-  grid-row: 1;
-  /* Top menu bar dropdown needs to be above of graph canvas splitter overlay which is z-index: 999 */
-  /* Top menu bar z-index needs to be higher than bottom menu bar z-index as by default
-  pysssss's image feed is located at body-bottom, and it can overlap with the queue button, which
-  is located in body-top. */
-  z-index: 1001;
-  display: flex;
-  flex-direction: column;
-}
-
-.comfyui-body-left {
-  order: -4;
-  /* Position in the first column */
-  grid-column: 1;
-  /* Position below the top element */
-  grid-row: 2;
-  z-index: 10;
-  display: flex;
-}
-
-.graph-canvas-container {
-  width: 100%;
-  height: 100%;
-  order: -3;
-  grid-column: 2;
-  grid-row: 2;
-  position: relative;
-  overflow: hidden;
-}
-
-#graph-canvas {
-  width: 100%;
-  height: 100%;
-  touch-action: none;
-}
-
-.comfyui-body-right {
-  order: -2;
-  z-index: 10;
-  grid-column: 3;
-  grid-row: 2;
-}
-
-.comfyui-body-bottom {
-  order: 4;
-  /* Span across all columns */
-  grid-column: 1/-1;
-  grid-row: 3;
-  /* Bottom menu bar dropdown needs to be above of graph canvas splitter overlay which is z-index: 999 */
-  z-index: 1000;
-  display: flex;
-  flex-direction: column;
-}
-
 .comfy-multiline-input {
   background-color: var(--comfy-input-bg);
   color: var(--input-text);
@@ -3531,84 +3434,6 @@ dialog::backdrop {
   justify-content: center;
 }
 
-#comfy-settings-dialog {
-  padding: 0;
-  width: 41rem;
-}
-
-#comfy-settings-dialog tr > td:first-child {
-  text-align: right;
-}
-
-#comfy-settings-dialog tbody button,
-#comfy-settings-dialog table > button {
-  background-color: var(--bg-color);
-  border: 1px var(--border-color) solid;
-  border-radius: 0;
-  color: var(--input-text);
-  font-size: 1rem;
-  padding: 0.5rem;
-}
-
-#comfy-settings-dialog button:hover {
-  background-color: var(--tr-odd-bg-color);
-}
-
-/* General CSS for tables */
-
-.comfy-table {
-  border-collapse: collapse;
-  color: var(--input-text);
-  font-family: Arial, sans-serif;
-  width: 100%;
-}
-
-.comfy-table caption {
-  position: sticky;
-  top: 0;
-  background-color: var(--bg-color);
-  color: var(--input-text);
-  font-size: 1rem;
-  font-weight: bold;
-  padding: 8px;
-  text-align: center;
-  border-bottom: 1px solid var(--border-color);
-}
-
-.comfy-table caption .comfy-btn {
-  position: absolute;
-  top: -2px;
-  right: 0;
-  bottom: 0;
-  cursor: pointer;
-  border: none;
-  height: 100%;
-  border-radius: 0;
-  aspect-ratio: 1/1;
-  -webkit-user-select: none;
-     -moz-user-select: none;
-          user-select: none;
-  font-size: 20px;
-}
-
-.comfy-table caption .comfy-btn:focus {
-  outline: none;
-}
-
-.comfy-table tr:nth-child(even) {
-  background-color: var(--tr-even-bg-color);
-}
-
-.comfy-table tr:nth-child(odd) {
-  background-color: var(--tr-odd-bg-color);
-}
-
-.comfy-table td,
-.comfy-table th {
-  border: 1px solid var(--border-color);
-  padding: 8px;
-}
-
 /* Context menu */
 
 .litegraph .dialog {
@@ -3708,24 +3533,6 @@ dialog::backdrop {
   will-change: transform;
 }
 
-@media only screen and (max-width: 450px) {
-  #comfy-settings-dialog .comfy-table tbody {
-    display: grid;
-  }
-  #comfy-settings-dialog .comfy-table tr {
-    display: grid;
-  }
-  #comfy-settings-dialog tr > td:first-child {
-    text-align: center;
-    border-bottom: none;
-    padding-bottom: 0;
-  }
-  #comfy-settings-dialog tr > td:not(:first-child) {
-    text-align: center;
-    border-top: none;
-  }
-}
-
 audio.comfy-audio.empty-audio-widget {
   display: none;
 }
@@ -3736,7 +3543,6 @@ audio.comfy-audio.empty-audio-widget {
   left: 0;
   width: 100%;
   height: 100%;
-  pointer-events: none;
 }
 
 /* Set auto complete panel's width as it is not accessible within vue-root */
@@ -3799,6 +3605,39 @@ audio.comfy-audio.empty-audio-widget {
 .hover\:opacity-100:hover{
   opacity: 1;
 }
+<<<<<<<< HEAD:comfy/web/assets/index-ChXzdVeQ.css
+========
+@media (prefers-reduced-motion: no-preference){
+
+  .motion-safe\:w-0{
+    width: 0px;
+  }
+
+  .motion-safe\:opacity-0{
+    opacity: 0;
+  }
+
+  .group\/sidebar-tab:focus-within .motion-safe\:group-focus-within\/sidebar-tab\:w-auto{
+    width: auto;
+  }
+
+  .group\/sidebar-tab:focus-within .motion-safe\:group-focus-within\/sidebar-tab\:opacity-100{
+    opacity: 1;
+  }
+
+  .group\/sidebar-tab:hover .motion-safe\:group-hover\/sidebar-tab\:w-auto{
+    width: auto;
+  }
+
+  .group\/sidebar-tab:hover .motion-safe\:group-hover\/sidebar-tab\:opacity-100{
+    opacity: 1;
+  }
+
+  .group\/tree-node:hover .motion-safe\:group-hover\/tree-node\:opacity-100{
+    opacity: 1;
+  }
+}
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/index-CBxvvAzM.css
 @media not all and (min-width: 640px){
 
   .max-sm\:hidden{
@@ -3886,7 +3725,7 @@ audio.comfy-audio.empty-audio-widget {
     padding-top: 0px
 }
 
-.prompt-dialog-content[data-v-3df70997] {
+.prompt-dialog-content[data-v-4f1e3bbe] {
   white-space: pre-wrap;
 }
 
@@ -3904,17 +3743,29 @@ audio.comfy-audio.empty-audio-widget {
   margin-bottom: 1rem;
 }
 
+<<<<<<<< HEAD:comfy/web/assets/index-ChXzdVeQ.css
 .comfy-error-report[data-v-3faf7785] {
+========
+.comfy-error-report[data-v-e5000be2] {
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/index-CBxvvAzM.css
   display: flex;
   flex-direction: column;
   gap: 1rem;
 }
+<<<<<<<< HEAD:comfy/web/assets/index-ChXzdVeQ.css
 .action-container[data-v-3faf7785] {
+========
+.action-container[data-v-e5000be2] {
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/index-CBxvvAzM.css
   display: flex;
   gap: 1rem;
   justify-content: flex-end;
 }
+<<<<<<<< HEAD:comfy/web/assets/index-ChXzdVeQ.css
 .wrapper-pre[data-v-3faf7785] {
+========
+.wrapper-pre[data-v-e5000be2] {
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/index-CBxvvAzM.css
   white-space: pre-wrap;
   word-wrap: break-word;
 }
@@ -3983,6 +3834,7 @@ audio.comfy-audio.empty-audio-widget {
   padding: 0px;
 }
 
+<<<<<<<< HEAD:comfy/web/assets/index-ChXzdVeQ.css
 .form-input[data-v-1451da7b] .input-slider .p-inputnumber input,
 .form-input[data-v-1451da7b] .input-slider .slider-part {
 
@@ -3990,6 +3842,15 @@ audio.comfy-audio.empty-audio-widget {
 }
 .form-input[data-v-1451da7b] .p-inputtext,
 .form-input[data-v-1451da7b] .p-select {
+========
+.form-input[data-v-a29c257f] .input-slider .p-inputnumber input,
+.form-input[data-v-a29c257f] .input-slider .slider-part {
+
+    width: 5rem
+}
+.form-input[data-v-a29c257f] .p-inputtext,
+.form-input[data-v-a29c257f] .p-select {
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/index-CBxvvAzM.css
 
     width: 11rem
 }
@@ -4279,26 +4140,26 @@ audio.comfy-audio.empty-audio-widget {
     position: relative;
 }
 
-[data-v-250ab9af] .p-terminal .xterm {
+[data-v-873a313f] .p-terminal .xterm {
   overflow-x: auto;
 }
-[data-v-250ab9af] .p-terminal .xterm-screen {
+[data-v-873a313f] .p-terminal .xterm-screen {
   background-color: black;
   overflow-y: hidden;
 }
 
-[data-v-90a7f075] .p-terminal .xterm {
+[data-v-14fef2e4] .p-terminal .xterm {
   overflow-x: auto;
 }
-[data-v-90a7f075] .p-terminal .xterm-screen {
+[data-v-14fef2e4] .p-terminal .xterm-screen {
   background-color: black;
   overflow-y: hidden;
 }
 
-[data-v-03daf1c8] .p-terminal .xterm {
+[data-v-cf0c7d52] .p-terminal .xterm {
   overflow-x: auto;
 }
-[data-v-03daf1c8] .p-terminal .xterm-screen {
+[data-v-cf0c7d52] .p-terminal .xterm-screen {
   background-color: black;
   overflow-y: hidden;
 }
@@ -4610,28 +4471,32 @@ audio.comfy-audio.empty-audio-widget {
   box-sizing: border-box;
 }
 
-.tree-node[data-v-a6457774] {
+.tree-node[data-v-a945b5a8] {
   width: 100%;
   display: flex;
   align-items: center;
   justify-content: space-between;
 }
-.leaf-count-badge[data-v-a6457774] {
+.leaf-count-badge[data-v-a945b5a8] {
   margin-left: 0.5rem;
 }
-.node-content[data-v-a6457774] {
+.node-content[data-v-a945b5a8] {
   display: flex;
   align-items: center;
   flex-grow: 1;
 }
-.leaf-label[data-v-a6457774] {
+.leaf-label[data-v-a945b5a8] {
   margin-left: 0.5rem;
 }
-[data-v-a6457774] .editable-text span {
+[data-v-a945b5a8] .editable-text span {
   word-break: break-all;
 }
 
+<<<<<<<< HEAD:comfy/web/assets/index-ChXzdVeQ.css
 [data-v-243f3ee3] .tree-explorer-node-label {
+========
+[data-v-e3a237e6] .tree-explorer-node-label {
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/index-CBxvvAzM.css
   width: 100%;
   display: flex;
   align-items: center;
@@ -4644,10 +4509,17 @@ audio.comfy-audio.empty-audio-widget {
  * By setting the position to relative on the parent and using an absolutely positioned pseudo-element,
  * we can create a visual indicator for the drop target without affecting the layout of other elements.
  */
+<<<<<<<< HEAD:comfy/web/assets/index-ChXzdVeQ.css
 [data-v-243f3ee3] .p-tree-node-content:has(.tree-folder) {
   position: relative;
 }
 [data-v-243f3ee3] .p-tree-node-content:has(.tree-folder.can-drop)::after {
+========
+[data-v-e3a237e6] .p-tree-node-content:has(.tree-folder) {
+  position: relative;
+}
+[data-v-e3a237e6] .p-tree-node-content:has(.tree-folder.can-drop)::after {
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/index-CBxvvAzM.css
   content: '';
   position: absolute;
   top: 0;
@@ -4658,21 +4530,21 @@ audio.comfy-audio.empty-audio-widget {
   pointer-events: none;
 }
 
-[data-v-5e759e25] .p-toolbar-end .p-button {
+[data-v-0061c432] .p-toolbar-end .p-button {
 
     padding-top: 0.25rem;
 
     padding-bottom: 0.25rem
 }
 @media (min-width: 1536px) {
-[data-v-5e759e25] .p-toolbar-end .p-button {
+[data-v-0061c432] .p-toolbar-end .p-button {
 
         padding-top: 0.5rem;
 
         padding-bottom: 0.5rem
 }
 }
-[data-v-5e759e25] .p-toolbar-start {
+[data-v-0061c432] .p-toolbar-start {
 
     min-width: 0px;
 
@@ -4750,36 +4622,11 @@ audio.comfy-audio.empty-audio-widget {
   vertical-align: top;
 }
 
-[data-v-0bb2ac55] .pi-fake-spacer {
+[data-v-3be51840] .pi-fake-spacer {
   height: 1px;
   width: 16px;
 }
 
-._content[data-v-c4279e6b] {
-
-    display: flex;
-
-    flex-direction: column
-}
-._content[data-v-c4279e6b] > :not([hidden]) ~ :not([hidden]) {
-
-    --tw-space-y-reverse: 0;
-
-    margin-top: calc(0.5rem * calc(1 - var(--tw-space-y-reverse)));
-
-    margin-bottom: calc(0.5rem * var(--tw-space-y-reverse))
-}
-._footer[data-v-c4279e6b] {
-
-    display: flex;
-
-    flex-direction: column;
-
-    align-items: flex-end;
-
-    padding-top: 1rem
-}
-
 .slot_row[data-v-d9792337] {
   padding: 2px;
 }
@@ -4907,6 +4754,31 @@ audio.comfy-audio.empty-audio-widget {
   color: var(--error-text);
 }
 
+._content[data-v-c4279e6b] {
+
+    display: flex;
+
+    flex-direction: column
+}
+._content[data-v-c4279e6b] > :not([hidden]) ~ :not([hidden]) {
+
+    --tw-space-y-reverse: 0;
+
+    margin-top: calc(0.5rem * calc(1 - var(--tw-space-y-reverse)));
+
+    margin-bottom: calc(0.5rem * var(--tw-space-y-reverse))
+}
+._footer[data-v-c4279e6b] {
+
+    display: flex;
+
+    flex-direction: column;
+
+    align-items: flex-end;
+
+    padding-top: 1rem
+}
+
 .node-lib-node-container[data-v-da9a8962] {
     height: 100%;
     width: 100%

comfy/web/assets/index-CgMyWf7n.js

@@ -1,7 +1,12 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/index-Br6dw1F6.js
 import { bt as BaseStyle, bu as script$s, bT as script$t, o as openBlock, f as createElementBlock, as as mergeProps, m as createBaseVNode, E as toDisplayString, bM as Ripple, r as resolveDirective, i as withDirectives, y as createBlock, C as resolveDynamicComponent, bi as script$u, bD as resolveComponent, ai as normalizeClass, ci as createSlots, z as withCtx, aU as script$v, c9 as script$w, F as Fragment, D as renderList, a7 as createTextVNode, c3 as setAttribute, cp as normalizeProps, A as renderSlot, B as createCommentVNode, bU as script$x, c8 as equals, cu as script$y, br as script$z, cy as getFirstFocusableElement, c2 as OverlayEventBus, cO as getVNodeProp, c6 as resolveFieldData, dl as invokeElementMethod, bJ as getAttribute, cP as getNextElementSibling, bZ as getOuterWidth, cQ as getPreviousElementSibling, l as script$A, bL as script$B, bO as script$C, bC as script$E, c7 as isNotEmpty, ar as withModifiers, c$ as getOuterHeight, bN as UniqueComponentId, cS as _default, bv as ZIndex, bx as focus, bV as addStyle, b_ as absolutePosition, bW as ConnectedOverlayScrollHandler, bX as isTouchDevice, dm as FilterOperator, bB as script$F, cm as script$G, bA as FocusTrap, k as createVNode, bE as Transition, bf as withKeys, c0 as getIndex, co as script$H, cR as isClickable, cT as clearSelection, c4 as localeComparator, ch as sort, cA as FilterService, df as FilterMatchMode, bI as findSingle, cD as findIndexInList, b$ as find, dn as exportCSV, cL as getOffset, cU as isRTL, dp as getHiddenElementOuterWidth, dq as getHiddenElementOuterHeight, dr as reorderArray, bQ as removeClass, bw as addClass, cc as isEmpty, cB as script$I, ce as script$J } from "./index-BsGgXmrT.js";
 import { s as script$D } from "./index-COyiXDAn.js";
+========
+import { bG as BaseStyle, bH as script$s, bX as script$t, o as openBlock, f as createElementBlock, at as mergeProps, m as createBaseVNode, E as toDisplayString, bO as Ripple, r as resolveDirective, i as withDirectives, y as createBlock, C as resolveDynamicComponent, bm as script$u, bR as resolveComponent, aj as normalizeClass, cp as createSlots, z as withCtx, aY as script$v, cf as script$w, F as Fragment, D as renderList, a8 as createTextVNode, c8 as setAttribute, cx as normalizeProps, A as renderSlot, B as createCommentVNode, bY as script$x, ce as equals, cF as script$y, bv as script$z, cJ as getFirstFocusableElement, c7 as OverlayEventBus, cZ as getVNodeProp, cc as resolveFieldData, dD as invokeElementMethod, bK as getAttribute, c_ as getNextElementSibling, c2 as getOuterWidth, c$ as getPreviousElementSibling, l as script$A, bN as script$B, bQ as script$C, cl as script$E, cd as isNotEmpty, as as withModifiers, da as getOuterHeight, bP as UniqueComponentId, d1 as _default, bZ as ZIndex, bL as focus, b_ as addStyle, c3 as absolutePosition, b$ as ConnectedOverlayScrollHandler, c0 as isTouchDevice, dE as FilterOperator, ca as script$F, ct as script$G, cB as FocusTrap, k as createVNode, bI as Transition, bj as withKeys, c5 as getIndex, cv as script$H, d0 as isClickable, d2 as clearSelection, c9 as localeComparator, co as sort, cL as FilterService, dx as FilterMatchMode, bJ as findSingle, cO as findIndexInList, c4 as find, dF as exportCSV, cW as getOffset, d3 as isRTL, dG as getHiddenElementOuterWidth, dH as getHiddenElementOuterHeight, dI as reorderArray, bT as removeClass, bU as addClass, ci as isEmpty, cM as script$I, ck as script$J } from "./index-Bv0b06LE.js";
+import { s as script$D } from "./index-Dzu9WL4p.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/index-CgMyWf7n.js
 var ColumnStyle = BaseStyle.extend({
   name: "column"
 });
@@ -8787,4 +8792,8 @@ export {
   script as h,
   script$l as s
 };
+<<<<<<<< HEAD:comfy/web/assets/index-Br6dw1F6.js
 //# sourceMappingURL=index-Br6dw1F6.js.map
+========
+//# sourceMappingURL=index-CgMyWf7n.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/index-CgMyWf7n.js

comfy/web/assets/index-Dzu9WL4p.js

@@ -0,0 +1,27 @@
+var __defProp = Object.defineProperty;
+var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+import { bX as script$1, o as openBlock, f as createElementBlock, at as mergeProps, m as createBaseVNode } from "./index-Bv0b06LE.js";
+var script = {
+  name: "BarsIcon",
+  "extends": script$1
+};
+function render(_ctx, _cache, $props, $setup, $data, $options) {
+  return openBlock(), createElementBlock("svg", mergeProps({
+    width: "14",
+    height: "14",
+    viewBox: "0 0 14 14",
+    fill: "none",
+    xmlns: "http://www.w3.org/2000/svg"
+  }, _ctx.pti()), _cache[0] || (_cache[0] = [createBaseVNode("path", {
+    "fill-rule": "evenodd",
+    "clip-rule": "evenodd",
+    d: "M13.3226 3.6129H0.677419C0.497757 3.6129 0.325452 3.54152 0.198411 3.41448C0.0713707 3.28744 0 3.11514 0 2.93548C0 2.75581 0.0713707 2.58351 0.198411 2.45647C0.325452 2.32943 0.497757 2.25806 0.677419 2.25806H13.3226C13.5022 2.25806 13.6745 2.32943 13.8016 2.45647C13.9286 2.58351 14 2.75581 14 2.93548C14 3.11514 13.9286 3.28744 13.8016 3.41448C13.6745 3.54152 13.5022 3.6129 13.3226 3.6129ZM13.3226 7.67741H0.677419C0.497757 7.67741 0.325452 7.60604 0.198411 7.479C0.0713707 7.35196 0 7.17965 0 6.99999C0 6.82033 0.0713707 6.64802 0.198411 6.52098C0.325452 6.39394 0.497757 6.32257 0.677419 6.32257H13.3226C13.5022 6.32257 13.6745 6.39394 13.8016 6.52098C13.9286 6.64802 14 6.82033 14 6.99999C14 7.17965 13.9286 7.35196 13.8016 7.479C13.6745 7.60604 13.5022 7.67741 13.3226 7.67741ZM0.677419 11.7419H13.3226C13.5022 11.7419 13.6745 11.6706 13.8016 11.5435C13.9286 11.4165 14 11.2442 14 11.0645C14 10.8848 13.9286 10.7125 13.8016 10.5855C13.6745 10.4585 13.5022 10.3871 13.3226 10.3871H0.677419C0.497757 10.3871 0.325452 10.4585 0.198411 10.5855C0.0713707 10.7125 0 10.8848 0 11.0645C0 11.2442 0.0713707 11.4165 0.198411 11.5435C0.325452 11.6706 0.497757 11.7419 0.677419 11.7419Z",
+    fill: "currentColor"
+  }, null, -1)]), 16);
+}
+__name(render, "render");
+script.render = render;
+export {
+  script as s
+};
+//# sourceMappingURL=index-Dzu9WL4p.js.map

comfy/web/assets/keybindingService-DyjX-nxF.js

@@ -1,6 +1,10 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/keybindingService-DoUb2RT6.js
 import { an as useKeybindingStore, L as useCommandStore, a as useSettingStore, di as KeyComboImpl, dj as KeybindingImpl } from "./index-BsGgXmrT.js";
+========
+import { ao as useKeybindingStore, J as useCommandStore, a as useSettingStore, dA as KeyComboImpl, dB as KeybindingImpl } from "./index-Bv0b06LE.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/keybindingService-DyjX-nxF.js
 const CORE_KEYBINDINGS = [
   {
     combo: {
@@ -186,7 +190,7 @@ const useKeybindingService = /* @__PURE__ */ __name(() => {
       return;
     }
     const target = event.composedPath()[0];
-    if (!keyCombo.hasModifier && (target.tagName === "TEXTAREA" || target.tagName === "INPUT" || target.tagName === "SPAN" && target.classList.contains("property_value"))) {
+    if (keyCombo.isReservedByTextInput && (target.tagName === "TEXTAREA" || target.tagName === "INPUT" || target.tagName === "SPAN" && target.classList.contains("property_value"))) {
       return;
     }
     const keybinding = keybindingStore.getKeybinding(keyCombo);
@@ -247,4 +251,8 @@ const useKeybindingService = /* @__PURE__ */ __name(() => {
 export {
   useKeybindingService as u
 };
+<<<<<<<< HEAD:comfy/web/assets/keybindingService-DoUb2RT6.js
 //# sourceMappingURL=keybindingService-DoUb2RT6.js.map
+========
+//# sourceMappingURL=keybindingService-DyjX-nxF.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/keybindingService-DyjX-nxF.js

comfy/web/assets/serverConfigStore-D2Vr0L0h.js

@@ -1,6 +1,10 @@
 var __defProp = Object.defineProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
+<<<<<<<< HEAD:comfy/web/assets/serverConfigStore-B9riwnSX.js
 import { I as defineStore, U as ref, c as computed } from "./index-BsGgXmrT.js";
+========
+import { a1 as defineStore, T as ref, c as computed } from "./index-Bv0b06LE.js";
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/serverConfigStore-D2Vr0L0h.js
 const useServerConfigStore = defineStore("serverConfig", () => {
   const serverConfigById = ref({});
   const serverConfigs = computed(() => {
@@ -87,4 +91,8 @@ const useServerConfigStore = defineStore("serverConfig", () => {
 export {
   useServerConfigStore as u
 };
+<<<<<<<< HEAD:comfy/web/assets/serverConfigStore-B9riwnSX.js
 //# sourceMappingURL=serverConfigStore-B9riwnSX.js.map
+========
+//# sourceMappingURL=serverConfigStore-D2Vr0L0h.js.map
+>>>>>>>> 96d891cb94d90f220e066cebad349887137f07a6:comfy/web/assets/serverConfigStore-D2Vr0L0h.js

comfy/web/assets/uvMirrors-B-HKMf6X.js

@@ -0,0 +1,16 @@
+const PYTHON_MIRROR = {
+  settingId: "Comfy-Desktop.UV.PythonInstallMirror",
+  mirror: "https://github.com/astral-sh/python-build-standalone/releases/download",
+  fallbackMirror: "https://bgithub.xyz/astral-sh/python-build-standalone/releases/download",
+  validationPathSuffix: "/20250115/cpython-3.10.16+20250115-aarch64-apple-darwin-debug-full.tar.zst.sha256"
+};
+const PYPI_MIRROR = {
+  settingId: "Comfy-Desktop.UV.PypiInstallMirror",
+  mirror: "https://pypi.org/simple/",
+  fallbackMirror: "https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
+};
+export {
+  PYTHON_MIRROR as P,
+  PYPI_MIRROR as a
+};
+//# sourceMappingURL=uvMirrors-B-HKMf6X.js.map

comfy/web/index.html

@@ -1,15 +1,15 @@
-<!DOCTYPE html>
-<html lang="en">
-  <head>
-    <meta charset="UTF-8">
-    <title>ComfyUI</title>
-    <meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=no">
-    <link rel="stylesheet" type="text/css" href="user.css" />
-    <link rel="stylesheet" type="text/css" href="materialdesignicons.min.css" />
-    <script type="module" crossorigin src="./assets/index-BsGgXmrT.js"></script>
-    <link rel="stylesheet" crossorigin href="./assets/index-ChXzdVeQ.css">
-  </head>
-  <body class="litegraph grid">
-    <div id="vue-app"></div>
-  </body>
-</html>
+<!DOCTYPE html>
+<html lang="en">
+  <head>
+    <meta charset="UTF-8">
+    <title>ComfyUI</title>
+    <meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=no">
+    <link rel="stylesheet" type="text/css" href="user.css" />
+    <link rel="stylesheet" type="text/css" href="materialdesignicons.min.css" />
+    <script type="module" crossorigin src="./assets/index-Bv0b06LE.js"></script>
+    <link rel="stylesheet" crossorigin href="./assets/index-CBxvvAzM.css">
+  </head>
+  <body class="litegraph grid">
+    <div id="vue-app"></div>
+  </body>
+</html>

comfy/web/scripts/domWidget.js

@@ -0,0 +1,2 @@
+// Shim for scripts/domWidget.ts
+export const DOMWidgetImpl = window.comfyAPI.domWidget.DOMWidgetImpl;

comfy/web/templates/image2image.json

@@ -330,7 +330,7 @@
         "Node name for S&R": "CheckpointLoaderSimple"
       },
       "widgets_values": [
-        "v1-5-pruned-emaonly.safetensors"
+        "v1-5-pruned-emaonly-fp16.safetensors"
       ]
     }
   ],
@@ -440,8 +440,8 @@
   "extra": {},
   "version": 0.4,
   "models": [{
-    "name": "v1-5-pruned-emaonly.safetensors",
-    "url": "https://huggingface.co/Comfy-Org/stable-diffusion-v1-5-archive/resolve/main/v1-5-pruned-emaonly.safetensors?download=true",
+    "name": "v1-5-pruned-emaonly-fp16.safetensors",
+    "url": "https://huggingface.co/Comfy-Org/stable-diffusion-v1-5-archive/resolve/main/v1-5-pruned-emaonly-fp16.safetensors?download=true",
     "directory": "checkpoints"
   }]
 }

comfy_extras/nodes/nodes_load_3d.py

@@ -23,10 +23,7 @@ class Load3D():
             "width": ("INT", {"default": 1024, "min": 1, "max": 4096, "step": 1}),
             "height": ("INT", {"default": 1024, "min": 1, "max": 4096, "step": 1}),
             "material": (["original", "normal", "wireframe", "depth"],),
-            "bg_color": ("STRING", {"default": "#000000", "multiline": False}),
-            "light_intensity": ("INT", {"default": 10, "min": 1, "max": 20, "step": 1}),
             "up_direction": (["original", "-x", "+x", "-y", "+y", "-z", "+z"],),
-            "fov": ("INT", {"default": 75, "min": 10, "max": 150, "step": 1}),
         }}
 
     RETURN_TYPES = ("IMAGE", "MASK", "STRING")
@@ -38,22 +35,14 @@ class Load3D():
     CATEGORY = "3d"
 
     def process(self, model_file, image, **kwargs):
-        if isinstance(image, dict):
-            image_path = folder_paths.get_annotated_filepath(image['image'])
-            mask_path = folder_paths.get_annotated_filepath(image['mask'])
+        image_path = folder_paths.get_annotated_filepath(image['image'])
+        mask_path = folder_paths.get_annotated_filepath(image['mask'])
 
-            load_image_node = nodes.LoadImage()
-            output_image, ignore_mask = load_image_node.load_image(image=image_path)
-            ignore_image, output_mask = load_image_node.load_image(image=mask_path)
+        load_image_node = nodes.LoadImage()
+        output_image, ignore_mask = load_image_node.load_image(image=image_path)
+        ignore_image, output_mask = load_image_node.load_image(image=mask_path)
 
-            return output_image, output_mask, model_file,
-        else:
-            # to avoid the format is not dict which will happen the FE code is not compatibility to core,
-            # we need to this to double-check, it can be removed after merged FE into the core
-            image_path = folder_paths.get_annotated_filepath(image)
-            load_image_node = nodes.LoadImage()
-            output_image, output_mask = load_image_node.load_image(image=image_path)
-            return output_image, output_mask, model_file,
+        return output_image, output_mask, model_file,
 
 
 class Load3DAnimation():
@@ -71,11 +60,7 @@ class Load3DAnimation():
             "width": ("INT", {"default": 1024, "min": 1, "max": 4096, "step": 1}),
             "height": ("INT", {"default": 1024, "min": 1, "max": 4096, "step": 1}),
             "material": (["original", "normal", "wireframe", "depth"],),
-            "bg_color": ("STRING", {"default": "#000000", "multiline": False}),
-            "light_intensity": ("INT", {"default": 10, "min": 1, "max": 20, "step": 1}),
             "up_direction": (["original", "-x", "+x", "-y", "+y", "-z", "+z"],),
-            "animation_speed": (["0.1", "0.5", "1", "1.5", "2"], {"default": "1"}),
-            "fov": ("INT", {"default": 75, "min": 10, "max": 150, "step": 1}),
         }}
 
     RETURN_TYPES = ("IMAGE", "MASK", "STRING")
@@ -87,20 +72,14 @@ class Load3DAnimation():
     CATEGORY = "3d"
 
     def process(self, model_file, image, **kwargs):
-        if isinstance(image, dict):
-            image_path = folder_paths.get_annotated_filepath(image['image'])
-            mask_path = folder_paths.get_annotated_filepath(image['mask'])
+        image_path = folder_paths.get_annotated_filepath(image['image'])
+        mask_path = folder_paths.get_annotated_filepath(image['mask'])
 
-            load_image_node = nodes.LoadImage()
-            output_image, ignore_mask = load_image_node.load_image(image=image_path)
-            ignore_image, output_mask = load_image_node.load_image(image=mask_path)
+        load_image_node = nodes.LoadImage()
+        output_image, ignore_mask = load_image_node.load_image(image=image_path)
+        ignore_image, output_mask = load_image_node.load_image(image=mask_path)
 
-            return output_image, output_mask, model_file,
-        else:
-            image_path = folder_paths.get_annotated_filepath(image)
-            load_image_node = nodes.LoadImage()
-            output_image, output_mask = load_image_node.load_image(image=image_path)
-            return output_image, output_mask, model_file,
+        return output_image, output_mask, model_file,
 
 
 class Preview3D():
@@ -109,10 +88,27 @@ class Preview3D():
         return {"required": {
             "model_file": ("STRING", {"default": "", "multiline": False}),
             "material": (["original", "normal", "wireframe", "depth"],),
-            "bg_color": ("STRING", {"default": "#000000", "multiline": False}),
-            "light_intensity": ("INT", {"default": 10, "min": 1, "max": 20, "step": 1}),
             "up_direction": (["original", "-x", "+x", "-y", "+y", "-z", "+z"],),
-            "fov": ("INT", {"default": 75, "min": 10, "max": 150, "step": 1}),
+        }}
+
+    OUTPUT_NODE = True
+    RETURN_TYPES = ()
+
+    CATEGORY = "3d"
+
+    FUNCTION = "process"
+    EXPERIMENTAL = True
+
+    def process(self, model_file, **kwargs):
+        return {"ui": {"model_file": [model_file]}, "result": ()}
+
+class Preview3DAnimation():
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {
+            "model_file": ("STRING", {"default": "", "multiline": False}),
+            "material": (["original", "normal", "wireframe", "depth"],),
+            "up_direction": (["original", "-x", "+x", "-y", "+y", "-z", "+z"],),
         }}
 
     OUTPUT_NODE = True
@@ -130,11 +126,13 @@ class Preview3D():
 NODE_CLASS_MAPPINGS = {
     "Load3D": Load3D,
     "Load3DAnimation": Load3DAnimation,
-    "Preview3D": Preview3D
+    "Preview3D": Preview3D,
+    "Preview3DAnimation": Preview3DAnimation
 }
 
 NODE_DISPLAY_NAME_MAPPINGS = {
     "Load3D": "Load 3D",
     "Load3DAnimation": "Load 3D - Animation",
-    "Preview3D": "Preview 3D"
+    "Preview3D": "Preview 3D",
+    "Preview3DAnimation": "Preview 3D - Animation"
 }

comfy_extras/nodes/nodes_model_advanced.py

@@ -2,6 +2,8 @@ import comfy.sd
 import comfy.model_sampling
 import comfy.latent_formats
 import torch
+import node_helpers
+
 
 from comfy.nodes.common import MAX_RESOLUTION
 
@@ -302,6 +304,24 @@ class RescaleCFG:
         m.set_model_sampler_cfg_function(rescale_cfg)
         return (m, )
 
+class ModelComputeDtype:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                              "dtype": (["default", "fp32", "fp16", "bf16"],),
+                              }}
+
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "advanced/debug/model"
+
+    def patch(self, model, dtype):
+        m = model.clone()
+        m.set_model_compute_dtype(node_helpers.string_to_torch_dtype(dtype))
+        return (m, )
+
+
 NODE_CLASS_MAPPINGS = {
     "ModelSamplingDiscrete": ModelSamplingDiscrete,
     "ModelSamplingContinuousEDM": ModelSamplingContinuousEDM,
@@ -311,4 +331,5 @@ NODE_CLASS_MAPPINGS = {
     "ModelSamplingAuraFlow": ModelSamplingAuraFlow,
     "ModelSamplingFlux": ModelSamplingFlux,
     "RescaleCFG": RescaleCFG,
+    "ModelComputeDtype": ModelComputeDtype,
 }

comfy_extras/nodes/nodes_model_merging_model_specific.py

@@ -205,6 +205,54 @@ class ModelMergeLTXV(nodes_model_merging.ModelMergeBlocks):
         return {"required": arg_dict}
 
 
+class ModelMergeCosmos7B(nodes_model_merging.ModelMergeBlocks):
+    CATEGORY = "advanced/model_merging/model_specific"
+
+    @classmethod
+    def INPUT_TYPES(s):
+        arg_dict = {"model1": ("MODEL",),
+                    "model2": ("MODEL",)}
+
+        argument = ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01})
+
+        arg_dict["pos_embedder."] = argument
+        arg_dict["extra_pos_embedder."] = argument
+        arg_dict["x_embedder."] = argument
+        arg_dict["t_embedder."] = argument
+        arg_dict["affline_norm."] = argument
+
+        for i in range(28):
+            arg_dict["blocks.block{}.".format(i)] = argument
+
+        arg_dict["final_layer."] = argument
+
+        return {"required": arg_dict}
+
+
+class ModelMergeCosmos14B(nodes_model_merging.ModelMergeBlocks):
+    CATEGORY = "advanced/model_merging/model_specific"
+
+    @classmethod
+    def INPUT_TYPES(s):
+        arg_dict = {"model1": ("MODEL",),
+                    "model2": ("MODEL",)}
+
+        argument = ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01})
+
+        arg_dict["pos_embedder."] = argument
+        arg_dict["extra_pos_embedder."] = argument
+        arg_dict["x_embedder."] = argument
+        arg_dict["t_embedder."] = argument
+        arg_dict["affline_norm."] = argument
+
+        for i in range(36):
+            arg_dict["blocks.block{}.".format(i)] = argument
+
+        arg_dict["final_layer."] = argument
+
+        return {"required": arg_dict}
+
+
 NODE_CLASS_MAPPINGS = {
     "ModelMergeSD1": ModelMergeSD1,
     "ModelMergeSD2": ModelMergeSD1,  # SD1 and SD2 have the same blocks
@@ -215,4 +263,6 @@ NODE_CLASS_MAPPINGS = {
     "ModelMergeSD35_Large": ModelMergeSD35_Large,
     "ModelMergeMochiPreview": ModelMergeMochiPreview,
     "ModelMergeLTXV": ModelMergeLTXV,
+    "ModelMergeCosmos7B": ModelMergeCosmos7B,
+    "ModelMergeCosmos14B": ModelMergeCosmos14B,
 }

comfy_extras/nodes_lumina2.py

@@ -0,0 +1,104 @@
+from comfy.comfy_types import IO, ComfyNodeABC, InputTypeDict
+import torch
+
+
+class RenormCFG:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "model": ("MODEL",),
+                              "cfg_trunc": ("FLOAT", {"default": 100, "min": 0.0, "max": 100.0, "step": 0.01}),
+                              "renorm_cfg": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step": 0.01}),
+                              }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "advanced/model"
+
+    def patch(self, model, cfg_trunc, renorm_cfg):
+        def renorm_cfg_func(args):
+            cond_denoised = args["cond_denoised"]
+            uncond_denoised = args["uncond_denoised"]
+            cond_scale = args["cond_scale"]
+            timestep = args["timestep"]
+            x_orig = args["input"]
+            in_channels = model.model.diffusion_model.in_channels
+
+            if timestep[0] < cfg_trunc:
+                cond_eps, uncond_eps = cond_denoised[:, :in_channels], uncond_denoised[:, :in_channels]
+                cond_rest, _ = cond_denoised[:, in_channels:], uncond_denoised[:, in_channels:]
+                half_eps = uncond_eps + cond_scale * (cond_eps - uncond_eps)
+                half_rest = cond_rest
+
+                if float(renorm_cfg) > 0.0:
+                    ori_pos_norm = torch.linalg.vector_norm(cond_eps
+                            , dim=tuple(range(1, len(cond_eps.shape))), keepdim=True
+                    )
+                    max_new_norm = ori_pos_norm * float(renorm_cfg)
+                    new_pos_norm = torch.linalg.vector_norm(
+                            half_eps, dim=tuple(range(1, len(half_eps.shape))), keepdim=True
+                        )
+                    if new_pos_norm >= max_new_norm:
+                        half_eps = half_eps * (max_new_norm / new_pos_norm)
+            else:
+                cond_eps, uncond_eps = cond_denoised[:, :in_channels], uncond_denoised[:, :in_channels]
+                cond_rest, _ = cond_denoised[:, in_channels:], uncond_denoised[:, in_channels:]
+                half_eps = cond_eps
+                half_rest = cond_rest
+
+            cfg_result = torch.cat([half_eps, half_rest], dim=1)
+
+            # cfg_result = uncond_denoised + (cond_denoised - uncond_denoised) * cond_scale
+
+            return x_orig - cfg_result
+
+        m = model.clone()
+        m.set_model_sampler_cfg_function(renorm_cfg_func)
+        return (m, )
+
+
+class CLIPTextEncodeLumina2(ComfyNodeABC):
+    SYSTEM_PROMPT = {
+        "superior": "You are an assistant designed to generate superior images with the superior "\
+            "degree of image-text alignment based on textual prompts or user prompts.",
+        "alignment": "You are an assistant designed to generate high-quality images with the "\
+            "highest degree of image-text alignment based on textual prompts."
+    }
+    SYSTEM_PROMPT_TIP = "Lumina2 provide two types of system prompts:" \
+        "Superior: You are an assistant designed to generate superior images with the superior "\
+        "degree of image-text alignment based on textual prompts or user prompts. "\
+        "Alignment: You are an assistant designed to generate high-quality images with the highest "\
+        "degree of image-text alignment based on textual prompts."
+    @classmethod
+    def INPUT_TYPES(s) -> InputTypeDict:
+        return {
+            "required": {
+                "system_prompt": (list(CLIPTextEncodeLumina2.SYSTEM_PROMPT.keys()), {"tooltip": CLIPTextEncodeLumina2.SYSTEM_PROMPT_TIP}),
+                "user_prompt": (IO.STRING, {"multiline": True, "dynamicPrompts": True, "tooltip": "The text to be encoded."}),
+                "clip": (IO.CLIP, {"tooltip": "The CLIP model used for encoding the text."})
+            }
+        }
+    RETURN_TYPES = (IO.CONDITIONING,)
+    OUTPUT_TOOLTIPS = ("A conditioning containing the embedded text used to guide the diffusion model.",)
+    FUNCTION = "encode"
+
+    CATEGORY = "conditioning"
+    DESCRIPTION = "Encodes a system prompt and a user prompt using a CLIP model into an embedding that can be used to guide the diffusion model towards generating specific images."
+
+    def encode(self, clip, user_prompt, system_prompt):
+        if clip is None:
+            raise RuntimeError("ERROR: clip input is invalid: None\n\nIf the clip is from a checkpoint loader node your checkpoint does not contain a valid clip or text encoder model.")
+        system_prompt = CLIPTextEncodeLumina2.SYSTEM_PROMPT[system_prompt]
+        prompt = f'{system_prompt} <Prompt Start> {user_prompt}'
+        tokens = clip.tokenize(prompt)
+        return (clip.encode_from_tokens_scheduled(tokens), )
+
+
+NODE_CLASS_MAPPINGS = {
+    "CLIPTextEncodeLumina2": CLIPTextEncodeLumina2,
+    "RenormCFG": RenormCFG
+}
+
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "CLIPTextEncodeLumina2": "CLIP Text Encode for Lumina2",
+}

comfy_extras/nodes_video.py

@@ -0,0 +1,75 @@
+import os
+import av
+import torch
+import folder_paths
+import json
+from fractions import Fraction
+
+
+class SaveWEBM:
+    def __init__(self):
+        self.output_dir = folder_paths.get_output_directory()
+        self.type = "output"
+        self.prefix_append = ""
+
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required":
+                    {"images": ("IMAGE", ),
+                     "filename_prefix": ("STRING", {"default": "ComfyUI"}),
+                     "codec": (["vp9", "av1"],),
+                     "fps": ("FLOAT", {"default": 24.0, "min": 0.01, "max": 1000.0, "step": 0.01}),
+                     "crf": ("FLOAT", {"default": 32.0, "min": 0, "max": 63.0, "step": 1, "tooltip": "Higher crf means lower quality with a smaller file size, lower crf means higher quality higher filesize."}),
+                     },
+                "hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},
+                }
+
+    RETURN_TYPES = ()
+    FUNCTION = "save_images"
+
+    OUTPUT_NODE = True
+
+    CATEGORY = "image/video"
+
+    EXPERIMENTAL = True
+
+    def save_images(self, images, codec, fps, filename_prefix, crf, prompt=None, extra_pnginfo=None):
+        filename_prefix += self.prefix_append
+        full_output_folder, filename, counter, subfolder, filename_prefix = folder_paths.get_save_image_path(filename_prefix, self.output_dir, images[0].shape[1], images[0].shape[0])
+
+        file = f"{filename}_{counter:05}_.webm"
+        container = av.open(os.path.join(full_output_folder, file), mode="w")
+
+        if prompt is not None:
+            container.metadata["prompt"] = json.dumps(prompt)
+
+        if extra_pnginfo is not None:
+            for x in extra_pnginfo:
+                container.metadata[x] = json.dumps(extra_pnginfo[x])
+
+        codec_map = {"vp9": "libvpx-vp9", "av1": "libaom-av1"}
+        stream = container.add_stream(codec_map[codec], rate=Fraction(round(fps * 1000), 1000))
+        stream.width = images.shape[-2]
+        stream.height = images.shape[-3]
+        stream.pix_fmt = "yuv420p"
+        stream.bit_rate = 0
+        stream.options = {'crf': str(crf)}
+
+        for frame in images:
+            frame = av.VideoFrame.from_ndarray(torch.clamp(frame[..., :3] * 255, min=0, max=255).to(device=torch.device("cpu"), dtype=torch.uint8).numpy(), format="rgb24")
+            for packet in stream.encode(frame):
+                container.mux(packet)
+        container.close()
+
+        results = [{
+            "filename": file,
+            "subfolder": subfolder,
+            "type": self.type
+        }]
+
+        return {"ui": {"images": results, "animated": (True,)}}  # TODO: frontend side
+
+
+NODE_CLASS_MAPPINGS = {
+    "SaveWEBM": SaveWEBM,
+}

requirements.txt

@@ -13,7 +13,8 @@ torchinfo
 safetensors>=0.4.2
 bitsandbytes>=0.43.0 ;platform_system != 'Darwin'
 bitsandbytes ;platform_system == 'Darwin'
-aiohttp>=3.8.4
+aiohttp>=3.11.8
+yarl>=1.18.0
 accelerate>=0.25.0
 pyyaml>=6.0
 scikit-image>=0.20.0
@@ -73,4 +74,5 @@ humanize
 lightning
 flax
 jax
-colour
+colour
+av

setup.py

@@ -23,7 +23,7 @@ package_name = "comfyui"
 """
 The current version.
 """
-version = "0.3.11"
+version = "0.3.15"
 
 """
 The package index to the torch built with AMD ROCm.

tests/unit/comfy_test/folder_path_test.py

@@ -26,7 +26,18 @@ def temp_dir():
         yield tmpdirname
 
 
-def test_get_directory_by_type():
+@pytest.fixture
+def set_base_dir():
+    fn = FolderNames()
+    with context_folder_names_and_paths(FolderNames()):
+        def _set_base_dir(base_dir):
+            fn.base_paths.clear()
+            fn.add_base_path(Path(base_dir))
+
+        yield _set_base_dir
+
+
+def test_get_directory_by_type(clear_folder_paths):
     test_dir = "/test/dir"
     folder_paths.set_output_directory(test_dir)
     assert folder_paths.get_directory_by_type("output") == test_dir
@@ -118,3 +129,49 @@ def test_add_output_path_absolute(temp_dir):
         assert len(mp.additional_absolute_directory_paths) == 0
         assert len(mp.additional_relative_directory_paths) == 1
         assert list(mp.additional_relative_directory_paths)[0] == (Path("output") / "diffusion_models")
+
+
+def test_base_path_changes(set_base_dir):
+    test_dir = os.path.abspath("/test/dir")
+    set_base_dir(test_dir)
+
+    assert folder_paths.base_path == test_dir
+    assert folder_paths.models_dir == os.path.join(test_dir, "models")
+    assert folder_paths.input_directory == os.path.join(test_dir, "input")
+    assert folder_paths.output_directory == os.path.join(test_dir, "output")
+    assert folder_paths.temp_directory == os.path.join(test_dir, "temp")
+    assert folder_paths.user_directory == os.path.join(test_dir, "user")
+
+    assert os.path.join(test_dir, "custom_nodes") in folder_paths.get_folder_paths("custom_nodes")
+
+    for name in ["checkpoints", "loras", "vae", "configs", "embeddings", "controlnet", "classifiers"]:
+        assert folder_paths.get_folder_paths(name)[0] == os.path.join(test_dir, "models", name)
+
+
+def test_base_path_change_clears_old(set_base_dir):
+    test_dir = os.path.abspath("/test/dir")
+    set_base_dir(test_dir)
+
+    assert len(folder_paths.get_folder_paths("custom_nodes")) == 1
+
+    single_model_paths = [
+        "checkpoints",
+        "loras",
+        "vae",
+        "configs",
+        "clip_vision",
+        "style_models",
+        "diffusers",
+        "vae_approx",
+        "gligen",
+        "upscale_models",
+        "embeddings",
+        "hypernetworks",
+        "photomaker",
+        "classifiers",
+    ]
+    for name in single_model_paths:
+        assert len(folder_paths.get_folder_paths(name)) == 1
+
+    for name in ["controlnet", "diffusion_models", "text_encoders"]:
+        assert len(folder_paths.get_folder_paths(name)) == 2

tests/unit/server/routes/internal_routes_test.py

@@ -1,115 +0,0 @@
-import pytest
-from aiohttp import web
-from unittest.mock import MagicMock, patch
-from comfy.api_server.routes.internal.internal_routes import InternalRoutes
-from comfy.api_server.services.file_service import FileService
-from comfy.cmd.folder_paths import models_dir, user_directory, output_directory
-
-
-@pytest.fixture
-def internal_routes():
-    return InternalRoutes(None)
-
-@pytest.fixture
-def aiohttp_client_factory(aiohttp_client, internal_routes):
-    async def _get_client():
-        app = internal_routes.get_app()
-        return await aiohttp_client(app)
-    return _get_client
-
-@pytest.mark.asyncio
-async def test_list_files_valid_directory(aiohttp_client_factory, internal_routes):
-    mock_file_list = [
-        {"name": "file1.txt", "path": "file1.txt", "type": "file", "size": 100},
-        {"name": "dir1", "path": "dir1", "type": "directory"}
-    ]
-    internal_routes.file_service.list_files = MagicMock(return_value=mock_file_list)
-    client = await aiohttp_client_factory()
-    resp = await client.get('/files?directory=models')
-    assert resp.status == 200
-    data = await resp.json()
-    assert 'files' in data
-    assert len(data['files']) == 2
-    assert data['files'] == mock_file_list
-
-    # Check other valid directories
-    resp = await client.get('/files?directory=user')
-    assert resp.status == 200
-    resp = await client.get('/files?directory=output')
-    assert resp.status == 200
-
-@pytest.mark.asyncio
-async def test_list_files_invalid_directory(aiohttp_client_factory, internal_routes):
-    internal_routes.file_service.list_files = MagicMock(side_effect=ValueError("Invalid directory key"))
-    client = await aiohttp_client_factory()
-    resp = await client.get('/files?directory=invalid')
-    assert resp.status == 400
-    data = await resp.json()
-    assert 'error' in data
-    assert data['error'] == "Invalid directory key"
-
-@pytest.mark.asyncio
-async def test_list_files_exception(aiohttp_client_factory, internal_routes):
-    internal_routes.file_service.list_files = MagicMock(side_effect=Exception("Unexpected error"))
-    client = await aiohttp_client_factory()
-    resp = await client.get('/files?directory=models')
-    assert resp.status == 500
-    data = await resp.json()
-    assert 'error' in data
-    assert data['error'] == "Unexpected error"
-
-@pytest.mark.asyncio
-async def test_list_files_no_directory_param(aiohttp_client_factory, internal_routes):
-    mock_file_list = []
-    internal_routes.file_service.list_files = MagicMock(return_value=mock_file_list)
-    client = await aiohttp_client_factory()
-    resp = await client.get('/files')
-    assert resp.status == 200
-    data = await resp.json()
-    assert 'files' in data
-    assert len(data['files']) == 0
-
-def test_setup_routes(internal_routes):
-    internal_routes.setup_routes()
-    routes = internal_routes.routes
-    assert any(route.method == 'GET' and str(route.path) == '/files' for route in routes)
-
-def test_get_app(internal_routes):
-    app = internal_routes.get_app()
-    assert isinstance(app, web.Application)
-    assert internal_routes._app is not None
-
-def test_get_app_reuse(internal_routes):
-    app1 = internal_routes.get_app()
-    app2 = internal_routes.get_app()
-    assert app1 is app2
-
-@pytest.mark.asyncio
-async def test_routes_added_to_app(aiohttp_client_factory, internal_routes):
-    client = await aiohttp_client_factory()
-    try:
-        resp = await client.get('/files')
-        print(f"Response received: status {resp.status}")  # noqa: T201
-    except Exception as e:
-        print(f"Exception occurred during GET request: {e}")  # noqa: T201
-        raise
-
-    assert resp.status != 404, "Route /files does not exist"
-
-@pytest.mark.asyncio
-async def test_file_service_initialization():
-    with patch('comfy.api_server.routes.internal.internal_routes.FileService') as MockFileService:
-        # Create a mock instance
-        mock_file_service_instance = MagicMock(spec=FileService)
-        MockFileService.return_value = mock_file_service_instance
-        internal_routes = InternalRoutes(None)
-
-        # Check if FileService was initialized with the correct parameters
-        MockFileService.assert_called_once_with({
-            "models": models_dir,
-            "user": user_directory,
-            "output": output_directory
-        })
-
-        # Verify that the file_service attribute of InternalRoutes is set
-        assert internal_routes.file_service == mock_file_service_instance

tests/unit/server/services/file_service_test.py

@@ -1,62 +0,0 @@
-from unittest.mock import MagicMock
-
-import pytest
-
-from comfy.api_server.services.file_service import FileService
-
-
-@pytest.fixture
-def mock_file_system_ops():
-    return MagicMock()
-
-
-@pytest.fixture
-def file_service(mock_file_system_ops):
-    allowed_directories = {
-        "models": "/path/to/models",
-        "user": "/path/to/user",
-        "output": "/path/to/output"
-    }
-    return FileService(allowed_directories, file_system_ops=mock_file_system_ops)
-
-
-def test_list_files_valid_directory(file_service, mock_file_system_ops):
-    mock_file_system_ops.walk_directory.return_value = [
-        {"name": "file1.txt", "path": "file1.txt", "type": "file", "size": 100},
-        {"name": "dir1", "path": "dir1", "type": "directory"}
-    ]
-
-    result = file_service.list_files("models")
-
-    assert len(result) == 2
-    assert result[0]["name"] == "file1.txt"
-    assert result[1]["name"] == "dir1"
-    mock_file_system_ops.walk_directory.assert_called_once_with("/path/to/models")
-
-
-def test_list_files_invalid_directory(file_service):
-    # Does not support walking directories outside of the allowed directories
-    with pytest.raises(ValueError, match="Invalid directory key"):
-        file_service.list_files("invalid_key")
-
-
-def test_list_files_empty_directory(file_service, mock_file_system_ops):
-    mock_file_system_ops.walk_directory.return_value = []
-
-    result = file_service.list_files("models")
-
-    assert len(result) == 0
-    mock_file_system_ops.walk_directory.assert_called_once_with("/path/to/models")
-
-
-@pytest.mark.parametrize("directory_key", ["models", "user", "output"])
-def test_list_files_all_allowed_directories(file_service, mock_file_system_ops, directory_key):
-    mock_file_system_ops.walk_directory.return_value = [
-        {"name": f"file_{directory_key}.txt", "path": f"file_{directory_key}.txt", "type": "file", "size": 100}
-    ]
-
-    result = file_service.list_files(directory_key)
-
-    assert len(result) == 1
-    assert result[0]["name"] == f"file_{directory_key}.txt"
-    mock_file_system_ops.walk_directory.assert_called_once_with(f"/path/to/{directory_key}")

tests/unit/utils/extra_config_test.py

@@ -118,7 +118,7 @@ def test_load_extra_model_paths_expands_userpath(
     mock_yaml_safe_load.assert_called_once()
 
     # Check if open was called with the correct file path
-    mock_file.assert_called_once_with(dummy_yaml_file_name, 'r')
+    mock_file.assert_called_once_with(dummy_yaml_file_name, 'r', encoding='utf-8')
 
 
 @patch('builtins.open', new_callable=mock_open)
@@ -149,7 +149,7 @@ def test_load_extra_model_paths_expands_appdata(
     else:
         expected_base_path = '/Users/TestUser/AppData/Roaming/ComfyUI'
     expected_calls = [
-        ('checkpoints', os.path.join(expected_base_path, 'models/checkpoints'), False),
+        ('checkpoints', os.path.normpath(os.path.join(expected_base_path, 'models/checkpoints')), False),
     ]
 
     assert mock_add_model_folder_path.call_count == len(expected_calls)