Merge branch 'master' of github.com:comfyanonymous/ComfyUI into merge/0.4.0-snapshot

comfy/api_server/routes/internal/internal_routes.py

@@ -62,8 +62,13 @@ class InternalRoutes:
                 return web.json_response({"error": "Invalid directory type"}, status=400)
 
             directory = get_directory_by_type(directory_type)
+
+            def is_visible_file(entry: os.DirEntry) -> bool:
+                """Filter out hidden files (e.g., .DS_Store on macOS)."""
+                return entry.is_file() and not entry.name.startswith('.')
+
             sorted_files = sorted(
-                (entry for entry in os.scandir(directory) if entry.is_file()),
+                (entry for entry in os.scandir(directory) if is_visible_file(entry)),
                 key=lambda entry: -entry.stat().st_mtime
             )
             return web.json_response([entry.name for entry in sorted_files], status=200)

comfy/model_detection.py

@@ -268,8 +268,10 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
                 dit_config["nerf_tile_size"] = 512
                 dit_config["nerf_final_head_type"] = "conv" if f"{key_prefix}nerf_final_layer_conv.norm.scale" in state_dict_keys else "linear"
                 dit_config["nerf_embedder_dtype"] = torch.float32
-            if "__x0__" in state_dict_keys: # x0 pred
-                dit_config["use_x0"] = True
+                if "__x0__" in state_dict_keys: # x0 pred
+                    dit_config["use_x0"] = True
+                else:
+                    dit_config["use_x0"] = False
         else:
             dit_config["guidance_embed"] = "{}guidance_in.in_layer.weight".format(key_prefix) in state_dict_keys
             dit_config["yak_mlp"] = '{}double_blocks.0.img_mlp.gate_proj.weight'.format(key_prefix) in state_dict_keys

comfy/sd.py

@@ -138,7 +138,7 @@ class CLIP:
 
         self.tokenizer: "sd1_clip.SD1Tokenizer" = tokenizer(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data)
         self.patcher = model_patcher.ModelPatcher(self.cond_stage_model, load_device=load_device, offload_device=offload_device)
-        #Match torch.float32 hardcode upcast in TE implemention
+        # Match torch.float32 hardcode upcast in TE implemention
         self.patcher.set_model_compute_dtype(torch.float32)
         self.patcher.hook_mode = EnumHookMode.MinVram
         self.patcher.is_clip = True
@@ -565,16 +565,10 @@ class VAE:
                     ddconfig = {"dim": dim, "z_dim": self.latent_channels, "dim_mult": [1, 2, 4, 4], "num_res_blocks": 2, "attn_scales": [], "temperal_downsample": [False, True, True], "dropout": 0.0}
                     self.first_stage_model = wan_vae.WanVAE(**ddconfig)
                     self.working_dtypes = [torch.bfloat16, torch.float16, torch.float32]
+                    self.memory_used_encode = lambda shape, dtype: (1500 if shape[2] <= 4 else 6000) * shape[3] * shape[4] * model_management.dtype_size(dtype)
+                    self.memory_used_decode = lambda shape, dtype: (2200 if shape[2] <= 4 else 7000) * shape[3] * shape[4] * (8 * 8) * model_management.dtype_size(dtype)
+
 
-                    # todo: not sure how to detect qwen here
-                    wan_21_decode = 7000
-                    wan_21_encode = wan_21_decode - 1000
-                    qwen_vae_decode = int(wan_21_decode / 3)
-                    qwen_vae_encode = int(wan_21_encode / 3)
-                    encode_const = qwen_vae_encode if "qwen" in self.ckpt_name.lower() else wan_21_encode
-                    decode_const = qwen_vae_decode if "qwen" in self.ckpt_name.lower() else wan_21_decode
-                    self.memory_used_encode = lambda shape, dtype: encode_const * shape[3] * shape[4] * model_management.dtype_size(dtype)
-                    self.memory_used_decode = lambda shape, dtype: decode_const * shape[3] * shape[4] * (8 * 8) * model_management.dtype_size(dtype)
             # Hunyuan 3d v2 2.0 & 2.1
             elif "geo_decoder.cross_attn_decoder.ln_1.bias" in sd:
 

comfy/supported_models.py

@@ -575,7 +575,7 @@ class SD3(supported_models_base.BASE):
     unet_extra_config = {}
     latent_format = latent_formats.SD3
 
-    memory_usage_factor = 1.2
+    memory_usage_factor = 1.6
 
     text_encoder_key_prefix = ["text_encoders."]
 
@@ -1108,7 +1108,7 @@ class ZImage(Lumina2):
         "shift": 3.0,
     }
 
-    memory_usage_factor = 1.7
+    memory_usage_factor = 2.0
 
     supported_inference_dtypes = [torch.bfloat16, torch.float16, torch.float32]
 

comfy_api/latest/_io.py

@@ -1818,7 +1818,7 @@ class NodeOutput(_NodeOutputInternal):
         if "expand" in data:
             expand = data["expand"]
         # todo: this is being called incorrectly, but since it's untested and unused anyway, maybe we'll wait for upstream to fix it
-        return cls(args=args, ui=ui, expand=expand)  # pylint: disable=unexpected-keyword-arg
+        return cls(*args, ui=ui, expand=expand)  # pylint: disable=unexpected-keyword-arg
 
     def __getitem__(self, index) -> Any:
         return self.args[index]

comfy_api_nodes/nodes_kling.py

@@ -2056,7 +2056,7 @@ class KlingExtension(ComfyExtension):
             OmniProImageToVideoNode,
             OmniProVideoToVideoNode,
             OmniProEditVideoNode,
-            # OmniProImageNode,  # need support from backend
+            OmniProImageNode,
         ]