fix IP2V -node

hyvideo/diffusion/pipelines/pipeline_hunyuan_video.py

@@ -251,6 +251,7 @@ class HunyuanVideoPipeline(DiffusionPipeline):
                 f" size of {batch_size}. Make sure the batch size matches the length of the generators."
             )
         noise = randn_tensor(shape, generator=generator, device=device, dtype=self.base_dtype)
+    
         if freenoise:
             logger.info("Applying FreeNoise")
             # code and comments from AnimateDiff-Evolved by Kosinkadink (https://github.com/Kosinkadink/ComfyUI-AnimateDiff-Evolved)
@@ -285,28 +286,26 @@ class HunyuanVideoPipeline(DiffusionPipeline):
                 # apply shuffled indexes
                 #print("place_idx:", place_idx, "delta:", delta, "list_idx:", list_idx)
                 noise[:, :, place_idx:place_idx + delta, :, :] = noise[:, :, list_idx, :, :]
+
         i2v_mask = None
+        if official_i2v:
+            # Create mask
+            i2v_mask = torch.zeros(shape[0], 1, shape[2], shape[3], shape[4], device=device)
+            i2v_mask[:, :, 0, ...] = 1.0
+
         if image_cond_latents is not None:
             if image_cond_latents.shape[2] == 1:      
                 padding = torch.zeros(shape, device=device)
                 padding[:, :, 0:1, :, :] = image_cond_latents
                 image_cond_latents = padding
-        if official_i2v:
-            # Create mask
-            i2v_mask = torch.zeros(shape[0], 1, shape[2], shape[3], shape[4], device=device)
-            i2v_mask[:, :, 0, ...] = 1.0
-            t = torch.tensor([0.999]).to(device=device)
-            latents = noise * t + image_cond_latents * (1 - t)
-            latents = latents.to(dtype=self.base_dtype)
-        elif latents is None:
-            print("No latents provided, generating noise and using it as latents")
-            latents = noise        
-        elif denoise_strength < 1.0:
-            latents = latents.to(device)
+
+        if denoise_strength < 1.0:
+            if official_i2v:
+                latents = torch.cat((latents[:,:,0].unsqueeze(2), latents), dim=2)
             timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, denoise_strength, device)
             latent_timestep = timesteps[:1]
-            frames_needed = noise.shape[1]
-            current_frames = latents.shape[1]
+            frames_needed = noise.shape[2]
+            current_frames = latents.shape[2]
             
             if frames_needed > current_frames:
                 repeat_factor = frames_needed - current_frames
@@ -316,8 +315,13 @@ class HunyuanVideoPipeline(DiffusionPipeline):
             elif frames_needed < current_frames:
                 latents = latents[:, :frames_needed, :, :, :]
             latents = latents * (1 - latent_timestep / 1000) + latent_timestep / 1000 * noise
+            print("latents shape:", latents.shape)
+        elif official_i2v:
+            t = torch.tensor([0.999]).to(device=device)
+            latents = noise * t + image_cond_latents * (1 - t)
+            latents = latents.to(dtype=self.base_dtype)
         else:
-            latents = latents.to(device)
+            latents = noise
 
         # Check existence to make it compatible with FlowMatchEulerDiscreteScheduler
         if hasattr(self.scheduler, "init_noise_sigma"):

hyvideo/text_encoder/__init__.py

@@ -6,6 +6,7 @@ import torch
 import torch.nn as nn
 from transformers import CLIPTextModel, CLIPTokenizer, AutoTokenizer, AutoModel, AutoProcessor, CLIPImageProcessor #LlavaForConditionalGeneration
 from .modeling_llava import LlavaForConditionalGeneration
+from .processing_llava import LlavaProcessor
 from transformers.utils import ModelOutput
 
 from ..constants import TEXT_ENCODER_PATH, TOKENIZER_PATH
@@ -166,8 +167,11 @@ class TextEncoder(nn.Module):
         elif "llm" in text_encoder_type or "glm" in text_encoder_type or "vlm" in text_encoder_type:
             self.output_key = output_key or "last_hidden_state"
             if "glm" in text_encoder_type or "vlm" in text_encoder_type:
-                #self.processor = AutoProcessor.from_pretrained(text_encoder_path, device=device)
-                self.processor = CLIPImageProcessor.from_pretrained(text_encoder_path, use_fast=False)
+                self.processor_ip2v = LlavaProcessor.from_pretrained(text_encoder_path, device=device)
+                self.processor_ip2v.patch_size = None
+                self.processor_ip2v.vision_feature_select_strategy = None
+                
+                self.processor = CLIPImageProcessor.from_pretrained(text_encoder_path, device=device)
                 self.processor.patch_size = None
                 self.processor.vision_feature_select_strategy = None
         else:
@@ -259,7 +263,7 @@ class TextEncoder(nn.Module):
                     raw_images.append(image1.squeeze(0)*255)
                 if image2 is not None:
                     raw_images.append(image2.squeeze(0)*255)
-                text_tokens = self.processor(
+                text_tokens = self.processor_ip2v(
                     raw_images, 
                     text, 
                     **kwargs,

hyvideo/text_encoder/processing_llava.py

@@ -0,0 +1,203 @@
+# coding=utf-8
+# Copyright 2023 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Processor class for Llava.
+"""
+
+from typing import List, Union
+
+from transformers.feature_extraction_utils import BatchFeature
+from transformers.image_utils import ImageInput, get_image_size, to_numpy_array
+from transformers.processing_utils import ProcessingKwargs, ProcessorMixin, Unpack, _validate_images_text_input_order
+from transformers.tokenization_utils_base import PreTokenizedInput, TextInput
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+
+class LlavaProcessorKwargs(ProcessingKwargs, total=False):
+    _defaults = {
+        "text_kwargs": {
+            "padding": False,
+        },
+        "images_kwargs": {},
+    }
+
+
+class LlavaProcessor(ProcessorMixin):
+    r"""
+    Constructs a Llava processor which wraps a Llava image processor and a Llava tokenizer into a single processor.
+
+    [`LlavaProcessor`] offers all the functionalities of [`CLIPImageProcessor`] and [`LlamaTokenizerFast`]. See the
+    [`~LlavaProcessor.__call__`] and [`~LlavaProcessor.decode`] for more information.
+
+    Args:
+        image_processor ([`CLIPImageProcessor`], *optional*):
+            The image processor is a required input.
+        tokenizer ([`LlamaTokenizerFast`], *optional*):
+            The tokenizer is a required input.
+        patch_size (`int`, *optional*):
+            Patch size from the vision tower.
+        vision_feature_select_strategy (`str`, *optional*):
+            The feature selection strategy used to select the vision feature from the vision backbone.
+            Shoudl be same as in model's config
+        chat_template (`str`, *optional*): A Jinja template which will be used to convert lists of messages
+            in a chat into a tokenizable string.
+        image_token (`str`, *optional*, defaults to `"<image>"`):
+            Special token used to denote image location.
+        num_additional_image_tokens (`int`, *optional*, defaults to 0):
+            Number of additional tokens added to the image embeddings, such as CLS (+1). If the backbone has no CLS or other
+            extra tokens appended, no need to set this arg.
+    """
+
+    attributes = ["image_processor", "tokenizer"]
+    valid_kwargs = [
+        "chat_template",
+        "patch_size",
+        "vision_feature_select_strategy",
+        "image_token",
+        "num_additional_image_tokens",
+    ]
+    image_processor_class = "AutoImageProcessor"
+    tokenizer_class = "AutoTokenizer"
+
+    def __init__(
+        self,
+        image_processor=None,
+        tokenizer=None,
+        patch_size=None,
+        vision_feature_select_strategy=None,
+        chat_template=None,
+        image_token="<image>",  # set the default and let users change if they have peculiar special tokens in rare cases
+        num_additional_image_tokens=0,
+        **kwargs,
+    ):
+        self.patch_size = patch_size
+        self.num_additional_image_tokens = num_additional_image_tokens
+        self.vision_feature_select_strategy = vision_feature_select_strategy
+        self.image_token = tokenizer.image_token if hasattr(tokenizer, "image_token") else image_token
+        super().__init__(image_processor, tokenizer, chat_template=chat_template)
+
+    def __call__(
+        self,
+        images: ImageInput = None,
+        text: Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]] = None,
+        audio=None,
+        videos=None,
+        **kwargs: Unpack[LlavaProcessorKwargs],
+    ) -> BatchFeature:
+        """
+        Main method to prepare for the model one or several sequences(s) and image(s). This method forwards the `text`
+        and `kwargs` arguments to LlamaTokenizerFast's [`~LlamaTokenizerFast.__call__`] if `text` is not `None` to encode
+        the text. To prepare the image(s), this method forwards the `images` and `kwrags` arguments to
+        CLIPImageProcessor's [`~CLIPImageProcessor.__call__`] if `images` is not `None`. Please refer to the doctsring
+        of the above two methods for more information.
+
+        Args:
+            images (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `List[PIL.Image.Image]`, `List[np.ndarray]`, `List[torch.Tensor]`):
+                The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch
+                tensor. Both channels-first and channels-last formats are supported.
+            text (`str`, `List[str]`, `List[List[str]]`):
+                The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings
+                (pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set
+                `is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
+            return_tensors (`str` or [`~utils.TensorType`], *optional*):
+                If set, will return tensors of a particular framework. Acceptable values are:
+                - `'tf'`: Return TensorFlow `tf.constant` objects.
+                - `'pt'`: Return PyTorch `torch.Tensor` objects.
+                - `'np'`: Return NumPy `np.ndarray` objects.
+                - `'jax'`: Return JAX `jnp.ndarray` objects.
+
+        Returns:
+            [`BatchFeature`]: A [`BatchFeature`] with the following fields:
+
+            - **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`.
+            - **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
+              `return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not
+              `None`).
+            - **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
+        """
+        if images is None and text is None:
+            raise ValueError("You have to specify at least one of `images` or `text`.")
+
+        # check if images and text inputs are reversed for BC
+        images, text = _validate_images_text_input_order(images, text)
+
+        output_kwargs = self._merge_kwargs(
+            LlavaProcessorKwargs,
+            tokenizer_init_kwargs=self.tokenizer.init_kwargs,
+            **kwargs,
+        )
+        if images is not None:
+            image_inputs = self.image_processor(images, **output_kwargs["images_kwargs"])
+        else:
+            image_inputs = {}
+
+        if isinstance(text, str):
+            text = [text]
+        elif not isinstance(text, list) and not isinstance(text[0], str):
+            raise ValueError("Invalid input text. Please provide a string, or a list of strings")
+
+        # try to expand inputs in processing if we have the necessary parts
+        prompt_strings = text
+        if image_inputs.get("pixel_values") is not None:
+            if self.patch_size is not None and self.vision_feature_select_strategy is not None:
+                # Replace the image token with the expanded image token sequence
+                pixel_values = image_inputs["pixel_values"]
+                height, width = get_image_size(to_numpy_array(pixel_values[0]))
+                num_image_tokens = (height // self.patch_size) * (
+                    width // self.patch_size
+                ) + self.num_additional_image_tokens
+                if self.vision_feature_select_strategy == "default":
+                    num_image_tokens -= self.num_additional_image_tokens
+
+                prompt_strings = []
+                for sample in text:
+                    sample = sample.replace(self.image_token, self.image_token * num_image_tokens)
+                    prompt_strings.append(sample)
+            else:
+                logger.warning_once(
+                    "Expanding inputs for image tokens in LLaVa should be done in processing. "
+                    "Please add `patch_size` and `vision_feature_select_strategy` to the model's processing config or set directly "
+                    "with `processor.patch_size = {{patch_size}}` and processor.vision_feature_select_strategy = {{vision_feature_select_strategy}}`. "
+                    "Using processors without these attributes in the config is deprecated and will throw an error in v4.50."
+                )
+
+        text_inputs = self.tokenizer(prompt_strings, **output_kwargs["text_kwargs"])
+        return BatchFeature(data={**text_inputs, **image_inputs})
+
+    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.batch_decode with CLIP->Llama
+    def batch_decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to LlamaTokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please
+        refer to the docstring of this method for more information.
+        """
+        return self.tokenizer.batch_decode(*args, **kwargs)
+
+    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.decode with CLIP->Llama
+    def decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to LlamaTokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to
+        the docstring of this method for more information.
+        """
+        return self.tokenizer.decode(*args, **kwargs)
+
+    @property
+    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.model_input_names
+    def model_input_names(self):
+        tokenizer_input_names = self.tokenizer.model_input_names
+        image_processor_input_names = self.image_processor.model_input_names
+        return list(dict.fromkeys(tokenizer_input_names + image_processor_input_names))