Merge branch 'master' into dr-support-pip-cm

README.md

@@ -71,6 +71,7 @@ See what ComfyUI can do with the [example workflows](https://comfyanonymous.gith
    - [Omnigen 2](https://comfyanonymous.github.io/ComfyUI_examples/omnigen/)
    - [Flux Kontext](https://comfyanonymous.github.io/ComfyUI_examples/flux/#flux-kontext-image-editing-model)
    - [HiDream E1.1](https://comfyanonymous.github.io/ComfyUI_examples/hidream/#hidream-e11)
+   - [Qwen Image Edit](https://comfyanonymous.github.io/ComfyUI_examples/qwen_image/#edit-model)
 - Video Models
    - [Stable Video Diffusion](https://comfyanonymous.github.io/ComfyUI_examples/video/)
    - [Mochi](https://comfyanonymous.github.io/ComfyUI_examples/mochi/)

comfy/ldm/qwen_image/model.py

@@ -347,10 +347,10 @@ class QwenImageTransformer2DModel(nn.Module):
         h_offset = ((h_offset + (patch_size // 2)) // patch_size)
         w_offset = ((w_offset + (patch_size // 2)) // patch_size)
 
-        img_ids = torch.zeros((h_len, w_len, 3), device=x.device, dtype=x.dtype)
+        img_ids = torch.zeros((h_len, w_len, 3), device=x.device)
         img_ids[:, :, 0] = img_ids[:, :, 1] + index
-        img_ids[:, :, 1] = img_ids[:, :, 1] + torch.linspace(h_offset, h_len - 1 + h_offset, steps=h_len, device=x.device, dtype=x.dtype).unsqueeze(1)
-        img_ids[:, :, 2] = img_ids[:, :, 2] + torch.linspace(w_offset, w_len - 1 + w_offset, steps=w_len, device=x.device, dtype=x.dtype).unsqueeze(0)
+        img_ids[:, :, 1] = img_ids[:, :, 1] + torch.linspace(h_offset, h_len - 1 + h_offset, steps=h_len, device=x.device, dtype=x.dtype).unsqueeze(1) - (h_len // 2)
+        img_ids[:, :, 2] = img_ids[:, :, 2] + torch.linspace(w_offset, w_len - 1 + w_offset, steps=w_len, device=x.device, dtype=x.dtype).unsqueeze(0) - (w_len // 2)
         return hidden_states, repeat(img_ids, "h w c -> b (h w) c", b=bs), orig_shape
 
     def forward(
@@ -396,10 +396,11 @@ class QwenImageTransformer2DModel(nn.Module):
                 hidden_states = torch.cat([hidden_states, kontext], dim=1)
                 img_ids = torch.cat([img_ids, kontext_ids], dim=1)
 
-        txt_start = round(max(((x.shape[-1] + (self.patch_size // 2)) // self.patch_size), ((x.shape[-2] + (self.patch_size // 2)) // self.patch_size)))
-        txt_ids = torch.linspace(txt_start, txt_start + context.shape[1], steps=context.shape[1], device=x.device, dtype=x.dtype).reshape(1, -1, 1).repeat(x.shape[0], 1, 3)
+        txt_start = round(max(((x.shape[-1] + (self.patch_size // 2)) // self.patch_size) // 2, ((x.shape[-2] + (self.patch_size // 2)) // self.patch_size) // 2))
+        txt_ids = torch.arange(txt_start, txt_start + context.shape[1], device=x.device).reshape(1, -1, 1).repeat(x.shape[0], 1, 3)
         ids = torch.cat((txt_ids, img_ids), dim=1)
         image_rotary_emb = self.pe_embedder(ids).squeeze(1).unsqueeze(2).to(x.dtype)
+        del ids, txt_ids, img_ids
 
         hidden_states = self.img_in(hidden_states)
         encoder_hidden_states = self.txt_norm(encoder_hidden_states)

comfy/sd1_clip.py

@@ -534,7 +534,10 @@ class SDTokenizer:
         min_padding = tokenizer_options.get("{}_min_padding".format(self.embedding_key), self.min_padding)
 
         text = escape_important(text)
-        parsed_weights = token_weights(text, 1.0)
+        if kwargs.get("disable_weights", False):
+            parsed_weights = [(text, 1.0)]
+        else:
+            parsed_weights = token_weights(text, 1.0)
 
         # tokenize words
         tokens = []

comfy/text_encoders/qwen_image.py

@@ -15,7 +15,7 @@ class QwenImageTokenizer(sd1_clip.SD1Tokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
         super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, name="qwen25_7b", tokenizer=Qwen25_7BVLITokenizer)
         self.llama_template = "<|im_start|>system\nDescribe the image by detailing the color, shape, size, texture, quantity, text, spatial relationships of the objects and background:<|im_end|>\n<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
-        self.llama_template_images = "<|im_start|>system\nDescribe the key features of the input image \\(color, shape, size, texture, objects, background\\), then explain how the user's text instruction should alter or modify the image. Generate a new image that meets the user's requirements while maintaining consistency with the original input where appropriate.<|im_end|>\n<|im_start|>user\n<|vision_start|><|image_pad|><|vision_end|>{}<|im_end|>\n<|im_start|>assistant\n"
+        self.llama_template_images = "<|im_start|>system\nDescribe the key features of the input image (color, shape, size, texture, objects, background), then explain how the user's text instruction should alter or modify the image. Generate a new image that meets the user's requirements while maintaining consistency with the original input where appropriate.<|im_end|>\n<|im_start|>user\n<|vision_start|><|image_pad|><|vision_end|>{}<|im_end|>\n<|im_start|>assistant\n"
 
     def tokenize_with_weights(self, text, return_word_ids=False, llama_template=None, images=[], **kwargs):
         if llama_template is None:
@@ -25,7 +25,7 @@ class QwenImageTokenizer(sd1_clip.SD1Tokenizer):
                 llama_text = self.llama_template.format(text)
         else:
             llama_text = llama_template.format(text)
-        tokens = super().tokenize_with_weights(llama_text, return_word_ids=return_word_ids, **kwargs)
+        tokens = super().tokenize_with_weights(llama_text, return_word_ids=return_word_ids, disable_weights=True, **kwargs)
         key_name = next(iter(tokens))
         embed_count = 0
         qwen_tokens = tokens[key_name]

comfy_extras/nodes_lt.py

@@ -166,7 +166,7 @@ class LTXVAddGuide:
         negative = self.add_keyframe_index(negative, frame_idx, guiding_latent, scale_factors)
 
         mask = torch.full(
-            (noise_mask.shape[0], 1, guiding_latent.shape[2], 1, 1),
+            (noise_mask.shape[0], 1, guiding_latent.shape[2], noise_mask.shape[3], noise_mask.shape[4]),
             1.0 - strength,
             dtype=noise_mask.dtype,
             device=noise_mask.device,

comfyui_version.py

@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.3.50"
+__version__ = "0.3.51"

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.3.50"
+version = "0.3.51"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.9"

requirements.txt

@@ -1,5 +1,5 @@
 comfyui-frontend-package==1.25.9
-comfyui-workflow-templates==0.1.60
+comfyui-workflow-templates==0.1.62
 comfyui-embedded-docs==0.2.6
 comfyui_manager
 torch