SANA-Video Image to Video pipeline SanaImageToVideoPipeline support (#12634)

* move sana-video to a new dir and add `SanaImageToVideoPipeline` with no modify;

* fix bug and run text/image-to-vidoe success;

* make style; quality; fix-copies;

* add sana image-to-video pipeline in markdown;

* add test case for sana image-to-video;

* make style;

* add a init file in sana-video test dir;

* Update src/diffusers/pipelines/sana_video/pipeline_sana_video_i2v.py

Co-authored-by: dg845 <[email protected]>

* Update tests/pipelines/sana_video/test_sana_video_i2v.py

Co-authored-by: dg845 <[email protected]>

* Update src/diffusers/pipelines/sana_video/pipeline_sana_video_i2v.py

Co-authored-by: dg845 <[email protected]>

* Update src/diffusers/pipelines/sana_video/pipeline_sana_video_i2v.py

Co-authored-by: dg845 <[email protected]>

* Update tests/pipelines/sana_video/test_sana_video_i2v.py

Co-authored-by: dg845 <[email protected]>

* minor update;

* fix bug and skip fp16 save test;

Co-authored-by: Yuyang Zhao <[email protected]>

* Update src/diffusers/pipelines/sana_video/pipeline_sana_video_i2v.py

Co-authored-by: dg845 <[email protected]>

* Update src/diffusers/pipelines/sana_video/pipeline_sana_video_i2v.py

Co-authored-by: dg845 <[email protected]>

* Update src/diffusers/pipelines/sana_video/pipeline_sana_video_i2v.py

Co-authored-by: dg845 <[email protected]>

* Update src/diffusers/pipelines/sana_video/pipeline_sana_video_i2v.py

Co-authored-by: dg845 <[email protected]>

* add copied from for `encode_prompt`

* Apply style fixes

---------

Co-authored-by: dg845 <[email protected]>
Co-authored-by: Yuyang Zhao <[email protected]>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Author: 4 people

docs/source/en/api/pipelines/sana_video.md

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License. -->
 
-# SanaVideoPipeline
+# Sana-Video
 
 <div class="flex flex-wrap space-x-1">
   <img alt="LoRA" src="https://img.shields.io/badge/LoRA-d8b4fe?style=flat"/>
@@ -37,6 +37,85 @@ Refer to [this](https://huggingface.co/collections/Efficient-Large-Model/sana-vi
 
 Note: The recommended dtype mentioned is for the transformer weights. The text encoder and VAE weights must stay in `torch.bfloat16` or `torch.float32` for the model to work correctly. Please refer to the inference example below to see how to load the model with the recommended dtype. 
 
+
+## Generation Pipelines
+
+<hfoptions id="generation pipelines">`
+<hfoption id="Text-to-Video">
+
+The example below demonstrates how to use the text-to-video pipeline to generate a video using a text descriptio and a starting frame.
+
+```python
+model_id = 
+pipe = SanaVideoPipeline.from_pretrained("Efficient-Large-Model/SANA-Video_2B_480p_diffusers", torch_dtype=torch.bfloat16)
+pipe.text_encoder.to(torch.bfloat16)
+pipe.vae.to(torch.float32)
+pipe.to("cuda")
+
+prompt = "A cat and a dog baking a cake together in a kitchen. The cat is carefully measuring flour, while the dog is stirring the batter with a wooden spoon. The kitchen is cozy, with sunlight streaming through the window."
+negative_prompt = "A chaotic sequence with misshapen, deformed limbs in heavy motion blur, sudden disappearance, jump cuts, jerky movements, rapid shot changes, frames out of sync, inconsistent character shapes, temporal artifacts, jitter, and ghosting effects, creating a disorienting visual experience."
+motion_scale = 30
+motion_prompt = f" motion score: {motion_scale}."
+prompt = prompt + motion_prompt
+
+video = pipe(
+    prompt=prompt,
+    negative_prompt=negative_prompt,
+    height=480,
+    width=832,
+    frames=81,
+    guidance_scale=6,
+    num_inference_steps=50,
+    generator=torch.Generator(device="cuda").manual_seed(0),
+).frames[0]
+
+export_to_video(video, "sana_video.mp4", fps=16)
+```
+
+</hfoption>
+<hfoption id="Image-to-Video">
+
+The example below demonstrates how to use the image-to-video pipeline to generate a video using a text descriptio and a starting frame.
+
+```python
+model_id = "Efficient-Large-Model/SANA-Video_2B_480p_diffusers"
+pipe = SanaImageToVideoPipeline.from_pretrained(
+    model_id,
+    torch_dtype=torch.bfloat16,
+)
+pipe.scheduler = FlowMatchEulerDiscreteScheduler.from_config(pipe.scheduler.config, flow_shift=8.0)
+pipe.vae.to(torch.float32)
+pipe.text_encoder.to(torch.bfloat16)
+pipe.to("cuda")
+
+image = load_image("https://raw.githubusercontent.com/NVlabs/Sana/refs/heads/main/asset/samples/i2v-1.png")
+prompt = "A woman stands against a stunning sunset backdrop, her long, wavy brown hair gently blowing in the breeze. She wears a sleeveless, light-colored blouse with a deep V-neckline, which accentuates her graceful posture. The warm hues of the setting sun cast a golden glow across her face and hair, creating a serene and ethereal atmosphere. The background features a blurred landscape with soft, rolling hills and scattered clouds, adding depth to the scene. The camera remains steady, capturing the tranquil moment from a medium close-up angle."
+negative_prompt = "A chaotic sequence with misshapen, deformed limbs in heavy motion blur, sudden disappearance, jump cuts, jerky movements, rapid shot changes, frames out of sync, inconsistent character shapes, temporal artifacts, jitter, and ghosting effects, creating a disorienting visual experience."
+motion_scale = 30
+motion_prompt = f" motion score: {motion_scale}."
+prompt = prompt + motion_prompt
+
+motion_scale = 30.0
+
+video = pipe(
+    image=image,
+    prompt=prompt,
+    negative_prompt=negative_prompt,
+    height=480,
+    width=832,
+    frames=81,
+    guidance_scale=6,
+    num_inference_steps=50,
+    generator=torch.Generator(device="cuda").manual_seed(0),
+).frames[0]
+
+export_to_video(video, "sana-i2v.mp4", fps=16)
+```
+
+</hfoption>
+</hfoptions>
+
+
 ## Quantization
 
 Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
@@ -97,6 +176,13 @@ export_to_video(output, "sana-video-output.mp4", fps=16)
   - __call__
 
 
+## SanaImageToVideoPipeline
+
+[[autodoc]] SanaImageToVideoPipeline
+  - all
+  - __call__
+
+
 ## SanaVideoPipelineOutput
 
-[[autodoc]] pipelines.sana.pipeline_sana_video.SanaVideoPipelineOutput
+[[autodoc]] pipelines.sana_video.pipeline_sana_video.SanaVideoPipelineOutput

scripts/convert_sana_video_to_diffusers.py

@@ -80,6 +80,8 @@ def main(args):
 
     # scheduler
     flow_shift = 8.0
+    if args.task == "i2v":
+        assert args.scheduler_type == "flow-euler", "Scheduler type must be flow-euler for i2v task."
 
     # model config
     layer_num = 20
@@ -312,6 +314,7 @@ def main(args):
         choices=["flow-dpm_solver", "flow-euler", "uni-pc"],
         help="Scheduler type to use.",
     )
+    parser.add_argument("--task", default="t2v", type=str, required=True, help="Task to convert, t2v or i2v.")
     parser.add_argument("--dump_path", default=None, type=str, required=True, help="Path to the output pipeline.")
     parser.add_argument("--save_full_pipeline", action="store_true", help="save all the pipeline elements in one.")
     parser.add_argument("--dtype", default="fp32", type=str, choices=["fp32", "fp16", "bf16"], help="Weight dtype.")

src/diffusers/__init__.py

@@ -545,11 +545,13 @@
             "QwenImagePipeline",
             "ReduxImageEncoder",
             "SanaControlNetPipeline",
+            "SanaImageToVideoPipeline",
             "SanaPAGPipeline",
             "SanaPipeline",
             "SanaSprintImg2ImgPipeline",
             "SanaSprintPipeline",
             "SanaVideoPipeline",
+            "SanaVideoPipeline",
             "SemanticStableDiffusionPipeline",
             "ShapEImg2ImgPipeline",
             "ShapEPipeline",
@@ -1227,6 +1229,7 @@
             QwenImagePipeline,
             ReduxImageEncoder,
             SanaControlNetPipeline,
+            SanaImageToVideoPipeline,
             SanaPAGPipeline,
             SanaPipeline,
             SanaSprintImg2ImgPipeline,

src/diffusers/models/transformers/transformer_sana_video.py

@@ -237,7 +237,6 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
         return freqs_cos, freqs_sin
 
 
-# Copied from diffusers.models.transformers.sana_transformer.SanaModulatedNorm
 class SanaModulatedNorm(nn.Module):
     def __init__(self, dim: int, elementwise_affine: bool = False, eps: float = 1e-6):
         super().__init__()
@@ -247,7 +246,7 @@ def forward(
         self, hidden_states: torch.Tensor, temb: torch.Tensor, scale_shift_table: torch.Tensor
     ) -> torch.Tensor:
         hidden_states = self.norm(hidden_states)
-        shift, scale = (scale_shift_table[None] + temb[:, None].to(scale_shift_table.device)).chunk(2, dim=1)
+        shift, scale = (scale_shift_table[None, None] + temb[:, :, None].to(scale_shift_table.device)).unbind(dim=2)
         hidden_states = hidden_states * (1 + scale) + shift
         return hidden_states
 
@@ -423,8 +422,8 @@ def forward(
 
         # 1. Modulation
         shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = (
-            self.scale_shift_table[None] + timestep.reshape(batch_size, 6, -1)
-        ).chunk(6, dim=1)
+            self.scale_shift_table[None, None] + timestep.reshape(batch_size, timestep.shape[1], 6, -1)
+        ).unbind(dim=2)
 
         # 2. Self Attention
         norm_hidden_states = self.norm1(hidden_states)
@@ -635,13 +634,16 @@ def forward(
 
         if guidance is not None:
             timestep, embedded_timestep = self.time_embed(
-                timestep, guidance=guidance, hidden_dtype=hidden_states.dtype
+                timestep.flatten(), guidance=guidance, hidden_dtype=hidden_states.dtype
             )
         else:
             timestep, embedded_timestep = self.time_embed(
-                timestep, batch_size=batch_size, hidden_dtype=hidden_states.dtype
+                timestep.flatten(), batch_size=batch_size, hidden_dtype=hidden_states.dtype
             )
 
+        timestep = timestep.view(batch_size, -1, timestep.size(-1))
+        embedded_timestep = embedded_timestep.view(batch_size, -1, embedded_timestep.size(-1))
+
         encoder_hidden_states = self.caption_projection(encoder_hidden_states)
         encoder_hidden_states = encoder_hidden_states.view(batch_size, -1, hidden_states.shape[-1])
 

src/diffusers/pipelines/__init__.py

@@ -308,7 +308,10 @@
         "SanaSprintPipeline",
         "SanaControlNetPipeline",
         "SanaSprintImg2ImgPipeline",
+    ]
+    _import_structure["sana_video"] = [
         "SanaVideoPipeline",
+        "SanaImageToVideoPipeline",
     ]
     _import_structure["semantic_stable_diffusion"] = ["SemanticStableDiffusionPipeline"]
     _import_structure["shap_e"] = ["ShapEImg2ImgPipeline", "ShapEPipeline"]
@@ -749,8 +752,8 @@
             SanaPipeline,
             SanaSprintImg2ImgPipeline,
             SanaSprintPipeline,
-            SanaVideoPipeline,
         )
+        from .sana_video import SanaImageToVideoPipeline, SanaVideoPipeline
         from .semantic_stable_diffusion import SemanticStableDiffusionPipeline
         from .shap_e import ShapEImg2ImgPipeline, ShapEPipeline
         from .stable_audio import StableAudioPipeline, StableAudioProjectionModel

src/diffusers/pipelines/sana/__init__.py

@@ -26,7 +26,6 @@
     _import_structure["pipeline_sana_controlnet"] = ["SanaControlNetPipeline"]
     _import_structure["pipeline_sana_sprint"] = ["SanaSprintPipeline"]
     _import_structure["pipeline_sana_sprint_img2img"] = ["SanaSprintImg2ImgPipeline"]
-    _import_structure["pipeline_sana_video"] = ["SanaVideoPipeline"]
 
 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     try:
@@ -40,7 +39,6 @@
         from .pipeline_sana_controlnet import SanaControlNetPipeline
         from .pipeline_sana_sprint import SanaSprintPipeline
         from .pipeline_sana_sprint_img2img import SanaSprintImg2ImgPipeline
-        from .pipeline_sana_video import SanaVideoPipeline
 else:
     import sys
 

src/diffusers/pipelines/sana/pipeline_output.py

@@ -3,7 +3,6 @@
 
 import numpy as np
 import PIL.Image
-import torch
 
 from ...utils import BaseOutput
 
@@ -20,18 +19,3 @@ class SanaPipelineOutput(BaseOutput):
     """
 
     images: Union[List[PIL.Image.Image], np.ndarray]
-
-
-@dataclass
-class SanaVideoPipelineOutput(BaseOutput):
-    r"""
-    Output class for Sana-Video pipelines.
-
-    Args:
-        frames (`torch.Tensor`, `np.ndarray`, or List[List[PIL.Image.Image]]):
-            List of video outputs - It can be a nested list of length `batch_size,` with each sub-list containing
-            denoised PIL image sequences of length `num_frames.` It can also be a NumPy array or Torch tensor of shape
-            `(batch_size, num_frames, channels, height, width)`.
-    """
-
-    frames: torch.Tensor

src/diffusers/pipelines/sana_video/__init__.py

@@ -0,0 +1,49 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+    DIFFUSERS_SLOW_IMPORT,
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    get_objects_from_module,
+    is_torch_available,
+    is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_import_structure = {}
+
+
+try:
+    if not (is_transformers_available() and is_torch_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ...utils import dummy_torch_and_transformers_objects  # noqa F403
+
+    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+    _import_structure["pipeline_sana_video"] = ["SanaVideoPipeline"]
+    _import_structure["pipeline_sana_video_i2v"] = ["SanaImageToVideoPipeline"]
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+    try:
+        if not (is_transformers_available() and is_torch_available()):
+            raise OptionalDependencyNotAvailable()
+
+    except OptionalDependencyNotAvailable:
+        from ...utils.dummy_torch_and_transformers_objects import *
+    else:
+        from .pipeline_sana_video import SanaVideoPipeline
+        from .pipeline_sana_video_i2v import SanaImageToVideoPipeline
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(
+        __name__,
+        globals()["__file__"],
+        _import_structure,
+        module_spec=__spec__,
+    )
+
+    for name, value in _dummy_objects.items():
+        setattr(sys.modules[__name__], name, value)

src/diffusers/pipelines/sana_video/pipeline_output.py

@@ -0,0 +1,20 @@
+from dataclasses import dataclass
+
+import torch
+
+from ...utils import BaseOutput
+
+
+@dataclass
+class SanaVideoPipelineOutput(BaseOutput):
+    r"""
+    Output class for Sana-Video pipelines.
+
+    Args:
+        frames (`torch.Tensor`, `np.ndarray`, or List[List[PIL.Image.Image]]):
+            List of video outputs - It can be a nested list of length `batch_size,` with each sub-list containing
+            denoised PIL image sequences of length `num_frames.` It can also be a NumPy array or Torch tensor of shape
+            `(batch_size, num_frames, channels, height, width)`.
+    """
+
+    frames: torch.Tensor

src/diffusers/pipelines/sana/pipeline_sana_video.py renamed to src/diffusers/pipelines/sana_video/pipeline_sana_video.py

@@ -95,17 +95,16 @@
         >>> from diffusers import SanaVideoPipeline
         >>> from diffusers.utils import export_to_video
 
-        >>> model_id = "Efficient-Large-Model/SANA-Video_2B_480p_diffusers"
-        >>> pipe = SanaVideoPipeline.from_pretrained(model_id)
+        >>> pipe = SanaVideoPipeline.from_pretrained("Efficient-Large-Model/SANA-Video_2B_480p_diffusers")
         >>> pipe.transformer.to(torch.bfloat16)
         >>> pipe.text_encoder.to(torch.bfloat16)
         >>> pipe.vae.to(torch.float32)
         >>> pipe.to("cuda")
-        >>> model_score = 30
+        >>> motion_score = 30
 
         >>> prompt = "Evening, backlight, side lighting, soft light, high contrast, mid-shot, centered composition, clean solo shot, warm color. A young Caucasian man stands in a forest, golden light glimmers on his hair as sunlight filters through the leaves. He wears a light shirt, wind gently blowing his hair and collar, light dances across his face with his movements. The background is blurred, with dappled light and soft tree shadows in the distance. The camera focuses on his lifted gaze, clear and emotional."
         >>> negative_prompt = "A chaotic sequence with misshapen, deformed limbs in heavy motion blur, sudden disappearance, jump cuts, jerky movements, rapid shot changes, frames out of sync, inconsistent character shapes, temporal artifacts, jitter, and ghosting effects, creating a disorienting visual experience."
-        >>> motion_prompt = f" motion score: {model_score}."
+        >>> motion_prompt = f" motion score: {motion_score}."
         >>> prompt = prompt + motion_prompt
 
         >>> output = pipe(
@@ -231,6 +230,7 @@ def __init__(
 
         self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)
 
+    # Copied from diffusers.pipelines.sana.pipeline_sana.SanaPipeline._get_gemma_prompt_embeds
     def _get_gemma_prompt_embeds(
         self,
         prompt: Union[str, List[str]],
@@ -827,9 +827,9 @@ def __call__(
         Examples:
 
         Returns:
-            [`~pipelines.sana.pipeline_output.SanaVideoPipelineOutput`] or `tuple`:
-                If `return_dict` is `True`, [`~pipelines.sana.pipeline_output.SanaVideoPipelineOutput`] is returned,
-                otherwise a `tuple` is returned where the first element is a list with the generated videos
+            [`~pipelines.sana_video.pipeline_output.SanaVideoPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`~pipelines.sana_video.pipeline_output.SanaVideoPipelineOutput`] is
+                returned, otherwise a `tuple` is returned where the first element is a list with the generated videos
         """
 
         if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):