Image to Video Generation with Stable Video Diffusion#
This Jupyter notebook can be launched after a local installation only.
Stable Video Diffusion (SVD) Image-to-Video is a diffusion model that takes in a still image as a conditioning frame, and generates a video from it. In this tutorial we consider how to convert and run Stable Video Diffusion using OpenVINO. We will use stable-video-diffusion-img2video-xt model as example. Additionally, to speedup video generation process we apply AnimateLCM LoRA weights and run optimization with NNCF.
Table of contents:
Installation Instructions#
This is a self-contained example that relies solely on its own code.
We recommend running the notebook in a virtual environment. You only need a Jupyter server to start. For details, please refer to Installation Guide.
Prerequisites#
%pip install -q "torch>=2.1" "diffusers>=0.25" "peft>=0.6.2" "transformers" "openvino>=2024.1.0" Pillow opencv-python tqdm "gradio>=4.19" safetensors --extra-index-url https://download.pytorch.org/whl/cpu
%pip install -q datasets "nncf>=2.10.0"
Download PyTorch Model#
The code below load Stable Video Diffusion XT model using Diffusers library and apply Consistency Distilled AnimateLCM weights.
import torch
from pathlib import Path
from diffusers import StableVideoDiffusionPipeline
from diffusers.utils import load_image, export_to_video
from diffusers.models.attention_processor import AttnProcessor
from safetensors import safe_open
import gc
import requests
lcm_scheduler_url = "https://huggingface.co/spaces/wangfuyun/AnimateLCM-SVD/raw/main/lcm_scheduler.py"
r = requests.get(lcm_scheduler_url)
with open("lcm_scheduler.py", "w") as f:
f.write(r.text)
from lcm_scheduler import AnimateLCMSVDStochasticIterativeScheduler
from huggingface_hub import hf_hub_download
MODEL_DIR = Path("model")
IMAGE_ENCODER_PATH = MODEL_DIR / "image_encoder.xml"
VAE_ENCODER_PATH = MODEL_DIR / "vae_encoder.xml"
VAE_DECODER_PATH = MODEL_DIR / "vae_decoder.xml"
UNET_PATH = MODEL_DIR / "unet.xml"
load_pt_pipeline = not (VAE_ENCODER_PATH.exists() and VAE_DECODER_PATH.exists() and UNET_PATH.exists() and IMAGE_ENCODER_PATH.exists())
unet, vae, image_encoder = None, None, None
if load_pt_pipeline:
noise_scheduler = AnimateLCMSVDStochasticIterativeScheduler(
num_train_timesteps=40,
sigma_min=0.002,
sigma_max=700.0,
sigma_data=1.0,
s_noise=1.0,
rho=7,
clip_denoised=False,
)
pipe = StableVideoDiffusionPipeline.from_pretrained(
"stabilityai/stable-video-diffusion-img2vid-xt",
variant="fp16",
scheduler=noise_scheduler,
)
pipe.unet.set_attn_processor(AttnProcessor())
hf_hub_download(
repo_id="wangfuyun/AnimateLCM-SVD-xt",
filename="AnimateLCM-SVD-xt.safetensors",
local_dir="./checkpoints",
)
state_dict = {}
LCM_LORA_PATH = Path(
"checkpoints/AnimateLCM-SVD-xt.safetensors",
)
with safe_open(LCM_LORA_PATH, framework="pt", device="cpu") as f:
for key in f.keys():
state_dict[key] = f.get_tensor(key)
missing, unexpected = pipe.unet.load_state_dict(state_dict, strict=True)
pipe.scheduler.save_pretrained(MODEL_DIR / "scheduler")
pipe.feature_extractor.save_pretrained(MODEL_DIR / "feature_extractor")
unet = pipe.unet
unet.eval()
vae = pipe.vae
vae.eval()
image_encoder = pipe.image_encoder
image_encoder.eval()
del pipe
gc.collect()
# Load the conditioning image
image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/rocket.png?download=true")
image = image.resize((512, 256))
Convert Model to OpenVINO Intermediate Representation#
OpenVINO supports PyTorch models via conversion into Intermediate
Representation (IR) format. We need to provide a model object, input
data for model tracing to ov.convert_model function to obtain
OpenVINO ov.Model object instance. Model can be saved on disk for
next deployment using ov.save_model function.
Stable Video Diffusion consists of 3 parts:
Image Encoder for extraction embeddings from the input image.
U-Net for step-by-step denoising video clip.
VAE for encoding input image into latent space and decoding generated video.
Let’s convert each part.
Image Encoder#
import openvino as ov
def cleanup_torchscript_cache():
"""
Helper for removing cached model representation
"""
torch._C._jit_clear_class_registry()
torch.jit._recursive.concrete_type_store = torch.jit._recursive.ConcreteTypeStore()
torch.jit._state._clear_class_state()
if not IMAGE_ENCODER_PATH.exists():
with torch.no_grad():
ov_model = ov.convert_model(
image_encoder,
example_input=torch.zeros((1, 3, 224, 224)),
input=[-1, 3, 224, 224],
)
ov.save_model(ov_model, IMAGE_ENCODER_PATH)
del ov_model
cleanup_torchscript_cache()
print(f"Image Encoder successfully converted to IR and saved to {IMAGE_ENCODER_PATH}")
del image_encoder
gc.collect();
U-net#
if not UNET_PATH.exists():
unet_inputs = {
"sample": torch.ones([2, 2, 8, 32, 32]),
"timestep": torch.tensor(1.256),
"encoder_hidden_states": torch.zeros([2, 1, 1024]),
"added_time_ids": torch.ones([2, 3]),
}
with torch.no_grad():
ov_model = ov.convert_model(unet, example_input=unet_inputs)
ov.save_model(ov_model, UNET_PATH)
del ov_model
cleanup_torchscript_cache()
print(f"UNet successfully converted to IR and saved to {UNET_PATH}")
del unet
gc.collect();
VAE Encoder and Decoder#
As discussed above VAE model used for encoding initial image and decoding generated video. Encoding and Decoding happen on different pipeline stages, so for convenient usage we separate VAE on 2 parts: Encoder and Decoder.
class VAEEncoderWrapper(torch.nn.Module):
def __init__(self, vae):
super().__init__()
self.vae = vae
def forward(self, image):
return self.vae.encode(x=image)["latent_dist"].sample()
class VAEDecoderWrapper(torch.nn.Module):
def __init__(self, vae):
super().__init__()
self.vae = vae
def forward(self, latents, num_frames: int):
return self.vae.decode(latents, num_frames=num_frames)
if not VAE_ENCODER_PATH.exists():
vae_encoder = VAEEncoderWrapper(vae)
with torch.no_grad():
ov_model = ov.convert_model(vae_encoder, example_input=torch.zeros((1, 3, 576, 1024)))
ov.save_model(ov_model, VAE_ENCODER_PATH)
cleanup_torchscript_cache()
print(f"VAE Encoder successfully converted to IR and saved to {VAE_ENCODER_PATH}")
del vae_encoder
gc.collect()
if not VAE_DECODER_PATH.exists():
vae_decoder = VAEDecoderWrapper(vae)
with torch.no_grad():
ov_model = ov.convert_model(vae_decoder, example_input=(torch.zeros((8, 4, 72, 128)), torch.tensor(8)))
ov.save_model(ov_model, VAE_DECODER_PATH)
cleanup_torchscript_cache()
print(f"VAE Decoder successfully converted to IR and saved to {VAE_ENCODER_PATH}")
del vae_decoder
gc.collect()
del vae
gc.collect();
Prepare Inference Pipeline#
The code bellow implements OVStableVideoDiffusionPipeline class for
running video generation using OpenVINO. The pipeline accepts input
image and returns the sequence of generated frames The diagram below
represents a simplified pipeline workflow.
svd#
The pipeline is very similar to Stable Diffusion Image to Image Generation pipeline with the only difference that Image Encoder is used instead of Text Encoder. Model takes input image and random seed as initial prompt. Then image encoded into embeddings space using Image Encoder and into latent space using VAE Encoder and passed as input to U-Net model. Next, the U-Net iteratively denoises the random latent video representations while being conditioned on the image embeddings. The output of the U-Net, being the noise residual, is used to compute a denoised latent image representation via a scheduler algorithm for next iteration in generation cycle. This process repeats the given number of times and, finally, VAE decoder converts denoised latents into sequence of video frames.
from diffusers.pipelines.pipeline_utils import DiffusionPipeline
import PIL.Image
from diffusers.image_processor import VaeImageProcessor
from diffusers.utils.torch_utils import randn_tensor
from typing import Callable, Dict, List, Optional, Union
from diffusers.pipelines.stable_video_diffusion import (
StableVideoDiffusionPipelineOutput,
)
def _append_dims(x, target_dims):
"""Appends dimensions to the end of a tensor until it has target_dims dimensions."""
dims_to_append = target_dims - x.ndim
if dims_to_append < 0:
raise ValueError(f"input has {x.ndim} dims but target_dims is {target_dims}, which is less")
return x[(...,) + (None,) * dims_to_append]
def tensor2vid(video: torch.Tensor, processor, output_type="np"):
# Based on:
# https://github.com/modelscope/modelscope/blob/1509fdb973e5871f37148a4b5e5964cafd43e64d/modelscope/pipelines/multi_modal/text_to_video_synthesis_pipeline.py#L78
batch_size, channels, num_frames, height, width = video.shape
outputs = []
for batch_idx in range(batch_size):
batch_vid = video[batch_idx].permute(1, 0, 2, 3)
batch_output = processor.postprocess(batch_vid, output_type)
outputs.append(batch_output)
return outputs
class OVStableVideoDiffusionPipeline(DiffusionPipeline):
r"""
Pipeline to generate video from an input image using Stable Video Diffusion.
This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
implemented for all pipelines (downloading, saving, running on a particular device, etc.).
Args:
vae ([`AutoencoderKL`]):
Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
image_encoder ([`~transformers.CLIPVisionModelWithProjection`]):
Frozen CLIP image-encoder ([laion/CLIP-ViT-H-14-laion2B-s32B-b79K](https://huggingface.co/laion/CLIP-ViT-H-14-laion2B-s32B-b79K)).
unet ([`UNetSpatioTemporalConditionModel`]):
A `UNetSpatioTemporalConditionModel` to denoise the encoded image latents.
scheduler ([`EulerDiscreteScheduler`]):
A scheduler to be used in combination with `unet` to denoise the encoded image latents.
feature_extractor ([`~transformers.CLIPImageProcessor`]):
A `CLIPImageProcessor` to extract features from generated images.
"""
def __init__(
self,
vae_encoder,
image_encoder,
unet,
vae_decoder,
scheduler,
feature_extractor,
):
super().__init__()
self.vae_encoder = vae_encoder
self.vae_decoder = vae_decoder
self.image_encoder = image_encoder
self.register_to_config(unet=unet)
self.scheduler = scheduler
self.feature_extractor = feature_extractor
self.vae_scale_factor = 2 ** (4 - 1)
self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
def _encode_image(self, image, device, num_videos_per_prompt, do_classifier_free_guidance):
dtype = torch.float32
if not isinstance(image, torch.Tensor):
image = self.image_processor.pil_to_numpy(image)
image = self.image_processor.numpy_to_pt(image)
# We normalize the image before resizing to match with the original implementation.
# Then we unnormalize it after resizing.
image = image * 2.0 - 1.0
image = _resize_with_antialiasing(image, (224, 224))
image = (image + 1.0) / 2.0
# Normalize the image with for CLIP input
image = self.feature_extractor(
images=image,
do_normalize=True,
do_center_crop=False,
do_resize=False,
do_rescale=False,
return_tensors="pt",
).pixel_values
image = image.to(device=device, dtype=dtype)
image_embeddings = torch.from_numpy(self.image_encoder(image)[0])
image_embeddings = image_embeddings.unsqueeze(1)
# duplicate image embeddings for each generation per prompt, using mps friendly method
bs_embed, seq_len, _ = image_embeddings.shape
image_embeddings = image_embeddings.repeat(1, num_videos_per_prompt, 1)
image_embeddings = image_embeddings.view(bs_embed * num_videos_per_prompt, seq_len, -1)
if do_classifier_free_guidance:
negative_image_embeddings = torch.zeros_like(image_embeddings)
# For classifier free guidance, we need to do two forward passes.
# Here we concatenate the unconditional and text embeddings into a single batch
# to avoid doing two forward passes
image_embeddings = torch.cat([negative_image_embeddings, image_embeddings])
return image_embeddings
def _encode_vae_image(
self,
image: torch.Tensor,
device,
num_videos_per_prompt,
do_classifier_free_guidance,
):
image_latents = torch.from_numpy(self.vae_encoder(image)[0])
if do_classifier_free_guidance:
negative_image_latents = torch.zeros_like(image_latents)
# For classifier free guidance, we need to do two forward passes.
# Here we concatenate the unconditional and text embeddings into a single batch
# to avoid doing two forward passes
image_latents = torch.cat([negative_image_latents, image_latents])
# duplicate image_latents for each generation per prompt, using mps friendly method
image_latents = image_latents.repeat(num_videos_per_prompt, 1, 1, 1)
return image_latents
def _get_add_time_ids(
self,
fps,
motion_bucket_id,
noise_aug_strength,
dtype,
batch_size,
num_videos_per_prompt,
do_classifier_free_guidance,
):
add_time_ids = [fps, motion_bucket_id, noise_aug_strength]
passed_add_embed_dim = 256 * len(add_time_ids)
expected_add_embed_dim = 3 * 256
if expected_add_embed_dim != passed_add_embed_dim:
raise ValueError(
f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`."
)
add_time_ids = torch.tensor([add_time_ids], dtype=dtype)
add_time_ids = add_time_ids.repeat(batch_size * num_videos_per_prompt, 1)
if do_classifier_free_guidance:
add_time_ids = torch.cat([add_time_ids, add_time_ids])
return add_time_ids
def decode_latents(self, latents, num_frames, decode_chunk_size=14):
# [batch, frames, channels, height, width] -> [batch*frames, channels, height, width]
latents = latents.flatten(0, 1)
latents = 1 / 0.18215 * latents
# decode decode_chunk_size frames at a time to avoid OOM
frames = []
for i in range(0, latents.shape[0], decode_chunk_size):
frame = torch.from_numpy(self.vae_decoder([latents[i : i + decode_chunk_size], num_frames])[0])
frames.append(frame)
frames = torch.cat(frames, dim=0)
# [batch*frames, channels, height, width] -> [batch, channels, frames, height, width]
frames = frames.reshape(-1, num_frames, *frames.shape[1:]).permute(0, 2, 1, 3, 4)
# we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
frames = frames.float()
return frames
def check_inputs(self, image, height, width):
if not isinstance(image, torch.Tensor) and not isinstance(image, PIL.Image.Image) and not isinstance(image, list):
raise ValueError("`image` has to be of type `torch.FloatTensor` or `PIL.Image.Image` or `List[PIL.Image.Image]` but is" f" {type(image)}")
if height % 8 != 0 or width % 8 != 0:
raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
def prepare_latents(
self,
batch_size,
num_frames,
num_channels_latents,
height,
width,
dtype,
device,
generator,
latents=None,
):
shape = (
batch_size,
num_frames,
num_channels_latents // 2,
height // self.vae_scale_factor,
width // self.vae_scale_factor,
)
if isinstance(generator, list) and len(generator) != batch_size:
raise ValueError(
f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
f" size of {batch_size}. Make sure the batch size matches the length of the generators."
)
if latents is None:
latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
else:
latents = latents.to(device)
# scale the initial noise by the standard deviation required by the scheduler
latents = latents * self.scheduler.init_noise_sigma
return latents
@torch.no_grad()
def __call__(
self,
image: Union[PIL.Image.Image, List[PIL.Image.Image], torch.FloatTensor],
height: int = 320,
width: int = 512,
num_frames: Optional[int] = 8,
num_inference_steps: int = 4,
min_guidance_scale: float = 1.0,
max_guidance_scale: float = 1.2,
fps: int = 7,
motion_bucket_id: int = 80,
noise_aug_strength: int = 0.01,
decode_chunk_size: Optional[int] = None,
num_videos_per_prompt: Optional[int] = 1,
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
latents: Optional[torch.FloatTensor] = None,
output_type: Optional[str] = "pil",
callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
callback_on_step_end_tensor_inputs: List[str] = ["latents"],
return_dict: bool = True,
):
r"""
The call function to the pipeline for generation.
Args:discussed
image (`PIL.Image.Image` or `List[PIL.Image.Image]` or `torch.FloatTensor`):
Image or images to guide image generation. If you provide a tensor, it needs to be compatible with
[`CLIPImageProcessor`](https://huggingface.co/lambdalabs/sd-image-variations-diffusers/blob/main/feature_extractor/preprocessor_config.json).
height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
The height in pixels of the generated image.
width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
The width in pixels of the generated image.
num_frames (`int`, *optional*):
The number of video frames to generate. Defaults to 14 for `stable-video-diffusion-img2vid` and to 25 for `stable-video-diffusion-img2vid-xt`
num_inference_steps (`int`, *optional*, defaults to 25):
The number of denoising steps. More denoising steps usually lead to a higher quality image at the
expense of slower inference. This parameter is modulated by `strength`.
min_guidance_scale (`float`, *optional*, defaults to 1.0):
The minimum guidance scale. Used for the classifier free guidance with first frame.
max_guidance_scale (`float`, *optional*, defaults to 3.0):
The maximum guidance scale. Used for the classifier free guidance with last frame.
fps (`int`, *optional*, defaults to 7):
Frames per second. The rate at which the generated images shall be exported to a video after generation.
Note that Stable Diffusion Video's UNet was micro-conditioned on fps-1 during training.
motion_bucket_id (`int`, *optional*, defaults to 127):
The motion bucket ID. Used as conditioning for the generation. The higher the number the more motion will be in the video.
noise_aug_strength (`int`, *optional*, defaults to 0.02):
The amount of noise added to the init image, the higher it is the less the video will look like the init image. Increase it for more motion.
decode_chunk_size (`int`, *optional*):
The number of frames to decode at a time. The higher the chunk size, the higher the temporal consistency
between frames, but also the higher the memory consumption. By default, the decoder will decode all frames at once
for maximal quality. Reduce `decode_chunk_size` to reduce memory usage.
num_videos_per_prompt (`int`, *optional*, defaults to 1):
The number of images to generate per prompt.
generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
generation deterministic.
latents (`torch.FloatTensor`, *optional*):
Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
tensor is generated by sampling using the supplied random `generator`.
output_type (`str`, *optional*, defaults to `"pil"`):
The output format of the generated image. Choose between `PIL.Image` or `np.array`.
callback_on_step_end (`Callable`, *optional*):
A function that calls at the end of each denoising steps during the inference. The function is called
with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
`callback_on_step_end_tensor_inputs`.
callback_on_step_end_tensor_inputs (`List`, *optional*):
The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
`._callback_tensor_inputs` attribute of your pipeline class.
return_dict (`bool`, *optional*, defaults to `True`):
Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
plain tuple.
Returns:
[`~pipelines.stable_diffusion.StableVideoDiffusionPipelineOutput`] or `tuple`:
If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableVideoDiffusionPipelineOutput`] is returned,
otherwise a `tuple` is returned where the first element is a list of list with the generated frames.
Examples:
```py
from diffusers import StableVideoDiffusionPipeline
from diffusers.utils import load_image, export_to_video
pipe = StableVideoDiffusionPipeline.from_pretrained("stabilityai/stable-video-diffusion-img2vid-xt", torch_dtype=torch.float16, variant="fp16")
pipe.to("cuda")
image = load_image("https://lh3.googleusercontent.com/y-iFOHfLTwkuQSUegpwDdgKmOjRSTvPxat63dQLB25xkTs4lhIbRUFeNBWZzYf370g=s1200")
image = image.resize((1024, 576))
frames = pipe(image, num_frames=25, decode_chunk_size=8).frames[0]
export_to_video(frames, "generated.mp4", fps=7)
```
"""
# 0. Default height and width to unet
height = height or 96 * self.vae_scale_factor
width = width or 96 * self.vae_scale_factor
num_frames = num_frames if num_frames is not None else 25
decode_chunk_size = decode_chunk_size if decode_chunk_size is not None else num_frames
# 1. Check inputs. Raise error if not correct
self.check_inputs(image, height, width)
# 2. Define call parameters
if isinstance(image, PIL.Image.Image):
batch_size = 1
elif isinstance(image, list):
batch_size = len(image)
else:
batch_size = image.shape[0]
device = torch.device("cpu")
# here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
# of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
# corresponds to doing no classifier free guidance.
do_classifier_free_guidance = max_guidance_scale > 1.0
# 3. Encode input image
image_embeddings = self._encode_image(image, device, num_videos_per_prompt, do_classifier_free_guidance)
# NOTE: Stable Diffusion Video was conditioned on fps - 1, which
# is why it is reduced here.
# See: https://github.com/Stability-AI/generative-models/blob/ed0997173f98eaf8f4edf7ba5fe8f15c6b877fd3/scripts/sampling/simple_video_sample.py#L188
fps = fps - 1
# 4. Encode input image using VAE
image = self.image_processor.preprocess(image, height=height, width=width)
noise = randn_tensor(image.shape, generator=generator, device=image.device, dtype=image.dtype)
image = image + noise_aug_strength * noise
image_latents = self._encode_vae_image(image, device, num_videos_per_prompt, do_classifier_free_guidance)
image_latents = image_latents.to(image_embeddings.dtype)
# Repeat the image latents for each frame so we can concatenate them with the noise
# image_latents [batch, channels, height, width] ->[batch, num_frames, channels, height, width]
image_latents = image_latents.unsqueeze(1).repeat(1, num_frames, 1, 1, 1)
# 5. Get Added Time IDs
added_time_ids = self._get_add_time_ids(
fps,
motion_bucket_id,
noise_aug_strength,
image_embeddings.dtype,
batch_size,
num_videos_per_prompt,
do_classifier_free_guidance,
)
added_time_ids = added_time_ids
# 4. Prepare timesteps
self.scheduler.set_timesteps(num_inference_steps, device=device)
timesteps = self.scheduler.timesteps
# 5. Prepare latent variables
num_channels_latents = 8
latents = self.prepare_latents(
batch_size * num_videos_per_prompt,
num_frames,
num_channels_latents,
height,
width,
image_embeddings.dtype,
device,
generator,
latents,
)
# 7. Prepare guidance scale
guidance_scale = torch.linspace(min_guidance_scale, max_guidance_scale, num_frames).unsqueeze(0)
guidance_scale = guidance_scale.to(device, latents.dtype)
guidance_scale = guidance_scale.repeat(batch_size * num_videos_per_prompt, 1)
guidance_scale = _append_dims(guidance_scale, latents.ndim)
# 8. Denoising loop
num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
num_timesteps = len(timesteps)
with self.progress_bar(total=num_inference_steps) as progress_bar:
for i, t in enumerate(timesteps):
# expand the latents if we are doing classifier free guidance
latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
# Concatenate image_latents over channels dimention
latent_model_input = torch.cat([latent_model_input, image_latents], dim=2)
# predict the noise residual
noise_pred = torch.from_numpy(
self.unet(
[
latent_model_input,
t,
image_embeddings,
added_time_ids,
]
)[0]
)
# perform guidance
if do_classifier_free_guidance:
noise_pred_uncond, noise_pred_cond = noise_pred.chunk(2)
noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_cond - noise_pred_uncond)
# compute the previous noisy sample x_t -> x_t-1
latents = self.scheduler.step(noise_pred, t, latents).prev_sample
if callback_on_step_end is not None:
callback_kwargs = {}
for k in callback_on_step_end_tensor_inputs:
callback_kwargs[k] = locals()[k]
callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
latents = callback_outputs.pop("latents", latents)
if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
progress_bar.update()
if not output_type == "latent":
frames = self.decode_latents(latents, num_frames, decode_chunk_size)
frames = tensor2vid(frames, self.image_processor, output_type=output_type)
else:
frames = latents
if not return_dict:
return frames
return StableVideoDiffusionPipelineOutput(frames=frames)
# resizing utils
def _resize_with_antialiasing(input, size, interpolation="bicubic", align_corners=True):
h, w = input.shape[-2:]
factors = (h / size[0], w / size[1])
# First, we have to determine sigma
# Taken from skimage: https://github.com/scikit-image/scikit-image/blob/v0.19.2/skimage/transform/_warps.py#L171
sigmas = (
max((factors[0] - 1.0) / 2.0, 0.001),
max((factors[1] - 1.0) / 2.0, 0.001),
)
# Now kernel size. Good results are for 3 sigma, but that is kind of slow. Pillow uses 1 sigma
# https://github.com/python-pillow/Pillow/blob/master/src/libImaging/Resample.c#L206
# But they do it in the 2 passes, which gives better results. Let's try 2 sigmas for now
ks = int(max(2.0 * 2 * sigmas[0], 3)), int(max(2.0 * 2 * sigmas[1], 3))
# Make sure it is odd
if (ks[0] % 2) == 0:
ks = ks[0] + 1, ks[1]
if (ks[1] % 2) == 0:
ks = ks[0], ks[1] + 1
input = _gaussian_blur2d(input, ks, sigmas)
output = torch.nn.functional.interpolate(input, size=size, mode=interpolation, align_corners=align_corners)
return output
def _compute_padding(kernel_size):
"""Compute padding tuple."""
# 4 or 6 ints: (padding_left, padding_right,padding_top,padding_bottom)
# https://pytorch.org/docs/stable/nn.html#torch.nn.functional.pad
if len(kernel_size) < 2:
raise AssertionError(kernel_size)
computed = [k - 1 for k in kernel_size]
# for even kernels we need to do asymmetric padding :(
out_padding = 2 * len(kernel_size) * [0]
for i in range(len(kernel_size)):
computed_tmp = computed[-(i + 1)]
pad_front = computed_tmp // 2
pad_rear = computed_tmp - pad_front
out_padding[2 * i + 0] = pad_front
out_padding[2 * i + 1] = pad_rear
return out_padding
def _filter2d(input, kernel):
# prepare kernel
b, c, h, w = input.shape
tmp_kernel = kernel[:, None, ...].to(device=input.device, dtype=input.dtype)
tmp_kernel = tmp_kernel.expand(-1, c, -1, -1)
height, width = tmp_kernel.shape[-2:]
padding_shape: list[int] = _compute_padding([height, width])
input = torch.nn.functional.pad(input, padding_shape, mode="reflect")
# kernel and input tensor reshape to align element-wise or batch-wise params
tmp_kernel = tmp_kernel.reshape(-1, 1, height, width)
input = input.view(-1, tmp_kernel.size(0), input.size(-2), input.size(-1))
# convolve the tensor with the kernel.
output = torch.nn.functional.conv2d(input, tmp_kernel, groups=tmp_kernel.size(0), padding=0, stride=1)
out = output.view(b, c, h, w)
return out
def _gaussian(window_size: int, sigma):
if isinstance(sigma, float):
sigma = torch.tensor([[sigma]])
batch_size = sigma.shape[0]
x = (torch.arange(window_size, device=sigma.device, dtype=sigma.dtype) - window_size // 2).expand(batch_size, -1)
if window_size % 2 == 0:
x = x + 0.5
gauss = torch.exp(-x.pow(2.0) / (2 * sigma.pow(2.0)))
return gauss / gauss.sum(-1, keepdim=True)
def _gaussian_blur2d(input, kernel_size, sigma):
if isinstance(sigma, tuple):
sigma = torch.tensor([sigma], dtype=input.dtype)
else:
sigma = sigma.to(dtype=input.dtype)
ky, kx = int(kernel_size[0]), int(kernel_size[1])
bs = sigma.shape[0]
kernel_x = _gaussian(kx, sigma[:, 1].view(bs, 1))
kernel_y = _gaussian(ky, sigma[:, 0].view(bs, 1))
out_x = _filter2d(input, kernel_x[..., None, :])
out = _filter2d(out_x, kernel_y[..., None])
return out
Run Video Generation#
Select Inference Device#
import requests
r = requests.get(
url="https://raw.githubusercontent.com/openvinotoolkit/openvino_notebooks/latest/utils/notebook_utils.py",
)
open("notebook_utils.py", "w").write(r.text)
from notebook_utils import device_widget
device = device_widget()
device
Dropdown(description='Device:', index=1, options=('CPU', 'AUTO'), value='AUTO')
from transformers import CLIPImageProcessor
core = ov.Core()
vae_encoder = core.compile_model(VAE_ENCODER_PATH, device.value)
image_encoder = core.compile_model(IMAGE_ENCODER_PATH, device.value)
unet = core.compile_model(UNET_PATH, device.value)
vae_decoder = core.compile_model(VAE_DECODER_PATH, device.value)
scheduler = AnimateLCMSVDStochasticIterativeScheduler.from_pretrained(MODEL_DIR / "scheduler")
feature_extractor = CLIPImageProcessor.from_pretrained(MODEL_DIR / "feature_extractor")
Now, let’s see model in action. > Please, note, video generation is
memory and time consuming process. For reducing memory consumption, we
decreased input video resolution to 576x320 and number of generated
frames that may affect quality of generated video. You can change these
settings manually providing height, width and num_frames
parameters into pipeline.
ov_pipe = OVStableVideoDiffusionPipeline(vae_encoder, image_encoder, unet, vae_decoder, scheduler, feature_extractor)
frames = ov_pipe(
image,
num_inference_steps=4,
motion_bucket_id=60,
num_frames=8,
height=320,
width=512,
generator=torch.manual_seed(12342),
).frames[0]
out_path = Path("generated.mp4")
export_to_video(frames, str(out_path), fps=7)
frames[0].save(
"generated.gif",
save_all=True,
append_images=frames[1:],
optimize=False,
duration=120,
loop=0,
)
from IPython.display import HTML
HTML('<img src="generated.gif">')
Quantization#
NNCF enables
post-training quantization by adding quantization layers into model
graph and then using a subset of the training dataset to initialize the
parameters of these additional quantization layers. Quantized operations
are executed in INT8 instead of FP32/FP16 making model
inference faster.
According to OVStableVideoDiffusionPipeline structure, the diffusion
model takes up significant portion of the overall pipeline execution
time. Now we will show you how to optimize the UNet part using
NNCF to reduce
computation cost and speed up the pipeline. Quantizing the rest of the
pipeline does not significantly improve inference performance but can
lead to a substantial degradation of accuracy. That’s why we use only
weight compression for the vae encoder and vae decoder to reduce
the memory footprint.
For the UNet model we apply quantization in hybrid mode which means that we quantize: (1) weights of MatMul and Embedding layers and (2) activations of other layers. The steps are the following:
Create a calibration dataset for quantization.
Collect operations with weights.
Run
nncf.compress_model()to compress only the model weights.Run
nncf.quantize()on the compressed model with weighted operations ignored by providingignored_scopeparameter.Save the
INT8model usingopenvino.save_model()function.
Please select below whether you would like to run quantization to improve model inference speed.
NOTE: Quantization is time and memory consuming operation. Running quantization code below may take some time.
from notebook_utils import quantization_widget
to_quantize = quantization_widget()
to_quantize
# Fetch `skip_kernel_extension` module
import requests
r = requests.get(
url="https://raw.githubusercontent.com/openvinotoolkit/openvino_notebooks/latest/utils/skip_kernel_extension.py",
)
open("skip_kernel_extension.py", "w").write(r.text)
ov_int8_pipeline = None
OV_INT8_UNET_PATH = MODEL_DIR / "unet_int8.xml"
OV_INT8_VAE_ENCODER_PATH = MODEL_DIR / "vae_encoder_int8.xml"
OV_INT8_VAE_DECODER_PATH = MODEL_DIR / "vae_decoder_int8.xml"
%load_ext skip_kernel_extension
Prepare calibration dataset#
We use a portion of
fusing/instructpix2pix-1000-samples
dataset from Hugging Face as calibration data. To collect intermediate
model inputs for UNet optimization we should customize
CompiledModel.
%%skip not $to_quantize.value
from typing import Any
import datasets
import numpy as np
from tqdm.notebook import tqdm
from IPython.utils import io
class CompiledModelDecorator(ov.CompiledModel):
def __init__(self, compiled_model: ov.CompiledModel, data_cache: List[Any] = None, keep_prob: float = 0.5):
super().__init__(compiled_model)
self.data_cache = data_cache if data_cache is not None else []
self.keep_prob = keep_prob
def __call__(self, *args, **kwargs):
if np.random.rand() <= self.keep_prob:
self.data_cache.append(*args)
return super().__call__(*args, **kwargs)
def collect_calibration_data(ov_pipe, calibration_dataset_size: int, num_inference_steps: int = 50) -> List[Dict]:
original_unet = ov_pipe.unet
calibration_data = []
ov_pipe.unet = CompiledModelDecorator(original_unet, calibration_data, keep_prob=1)
dataset = datasets.load_dataset("fusing/instructpix2pix-1000-samples", split="train", streaming=False).shuffle(seed=42)
# Run inference for data collection
pbar = tqdm(total=calibration_dataset_size)
for batch in dataset:
image = batch["input_image"]
with io.capture_output() as captured:
ov_pipe(
image,
num_inference_steps=4,
motion_bucket_id=60,
num_frames=8,
height=256,
width=256,
generator=torch.manual_seed(12342),
)
pbar.update(len(calibration_data) - pbar.n)
if len(calibration_data) >= calibration_dataset_size:
break
ov_pipe.unet = original_unet
return calibration_data[:calibration_dataset_size]
%%skip not $to_quantize.value
if not OV_INT8_UNET_PATH.exists():
subset_size = 200
calibration_data = collect_calibration_data(ov_pipe, calibration_dataset_size=subset_size)
Run Hybrid Model Quantization#
%%skip not $to_quantize.value
from collections import deque
def get_operation_const_op(operation, const_port_id: int):
node = operation.input_value(const_port_id).get_node()
queue = deque([node])
constant_node = None
allowed_propagation_types_list = ["Convert", "FakeQuantize", "Reshape"]
while len(queue) != 0:
curr_node = queue.popleft()
if curr_node.get_type_name() == "Constant":
constant_node = curr_node
break
if len(curr_node.inputs()) == 0:
break
if curr_node.get_type_name() in allowed_propagation_types_list:
queue.append(curr_node.input_value(0).get_node())
return constant_node
def is_embedding(node) -> bool:
allowed_types_list = ["f16", "f32", "f64"]
const_port_id = 0
input_tensor = node.input_value(const_port_id)
if input_tensor.get_element_type().get_type_name() in allowed_types_list:
const_node = get_operation_const_op(node, const_port_id)
if const_node is not None:
return True
return False
def collect_ops_with_weights(model):
ops_with_weights = []
for op in model.get_ops():
if op.get_type_name() == "MatMul":
constant_node_0 = get_operation_const_op(op, const_port_id=0)
constant_node_1 = get_operation_const_op(op, const_port_id=1)
if constant_node_0 or constant_node_1:
ops_with_weights.append(op.get_friendly_name())
if op.get_type_name() == "Gather" and is_embedding(op):
ops_with_weights.append(op.get_friendly_name())
return ops_with_weights
%%skip not $to_quantize.value
import nncf
import logging
from nncf.quantization.advanced_parameters import AdvancedSmoothQuantParameters
nncf.set_log_level(logging.ERROR)
if not OV_INT8_UNET_PATH.exists():
diffusion_model = core.read_model(UNET_PATH)
unet_ignored_scope = collect_ops_with_weights(diffusion_model)
compressed_diffusion_model = nncf.compress_weights(diffusion_model, ignored_scope=nncf.IgnoredScope(types=['Convolution']))
quantized_diffusion_model = nncf.quantize(
model=compressed_diffusion_model,
calibration_dataset=nncf.Dataset(calibration_data),
subset_size=subset_size,
model_type=nncf.ModelType.TRANSFORMER,
# We additionally ignore the first convolution to improve the quality of generations
ignored_scope=nncf.IgnoredScope(names=unet_ignored_scope + ["__module.conv_in/aten::_convolution/Convolution"]),
advanced_parameters=nncf.AdvancedQuantizationParameters(smooth_quant_alphas=AdvancedSmoothQuantParameters(matmul=-1))
)
ov.save_model(quantized_diffusion_model, OV_INT8_UNET_PATH)
Run Weight Compression#
Quantizing of the vae encoder and vae decoder does not
significantly improve inference performance but can lead to a
substantial degradation of accuracy. Only weight compression will be
applied for footprint reduction.
%%skip not $to_quantize.value
nncf.set_log_level(logging.INFO)
if not OV_INT8_VAE_ENCODER_PATH.exists():
text_encoder_model = core.read_model(VAE_ENCODER_PATH)
compressed_text_encoder_model = nncf.compress_weights(text_encoder_model, mode=nncf.CompressWeightsMode.INT4_SYM, group_size=64)
ov.save_model(compressed_text_encoder_model, OV_INT8_VAE_ENCODER_PATH)
if not OV_INT8_VAE_DECODER_PATH.exists():
decoder_model = core.read_model(VAE_DECODER_PATH)
compressed_decoder_model = nncf.compress_weights(decoder_model, mode=nncf.CompressWeightsMode.INT4_SYM, group_size=64)
ov.save_model(compressed_decoder_model, OV_INT8_VAE_DECODER_PATH)
Let’s compare the video generated by the original and optimized
pipelines. Dynamic quantization should be disabled for UNet model
because it introduces a performance overhead when applied to Diffusion
models that have been quantized using a Hybrid approach.
%%skip not $to_quantize.value
ov_int8_vae_encoder = core.compile_model(OV_INT8_VAE_ENCODER_PATH, device.value)
ov_int8_unet = core.compile_model(OV_INT8_UNET_PATH, device.value, config={"DYNAMIC_QUANTIZATION_GROUP_SIZE":"0"})
ov_int8_decoder = core.compile_model(OV_INT8_VAE_DECODER_PATH, device.value)
ov_int8_pipeline = OVStableVideoDiffusionPipeline(
ov_int8_vae_encoder, image_encoder, ov_int8_unet, ov_int8_decoder, scheduler, feature_extractor
)
int8_frames = ov_int8_pipeline(
image,
num_inference_steps=4,
motion_bucket_id=60,
num_frames=8,
height=320,
width=512,
generator=torch.manual_seed(12342),
).frames[0]
%%skip not $to_quantize.value
from IPython.display import display
int8_out_path = Path("generated_int8.mp4")
export_to_video(int8_frames, str(int8_out_path), fps=7)
int8_frames[0].save(
"generated_int8.gif",
save_all=True,
append_images=int8_frames[1:],
optimize=False,
duration=120,
loop=0,
)
display(HTML('<img src="generated_int8.gif">'))
Compare model file sizes#
%%skip not $to_quantize.value
fp16_model_paths = [VAE_ENCODER_PATH, UNET_PATH, VAE_DECODER_PATH]
int8_model_paths = [OV_INT8_VAE_ENCODER_PATH, OV_INT8_UNET_PATH, OV_INT8_VAE_DECODER_PATH]
for fp16_path, int8_path in zip(fp16_model_paths, int8_model_paths):
fp16_ir_model_size = fp16_path.with_suffix(".bin").stat().st_size
int8_model_size = int8_path.with_suffix(".bin").stat().st_size
print(f"{fp16_path.stem} compression rate: {fp16_ir_model_size / int8_model_size:.3f}")
vae_encoder compression rate: 2.018
unet compression rate: 1.996
vae_decoder compression rate: 2.007
Compare inference time of the FP16 and INT8 pipelines#
To measure the inference performance of the FP16 and INT8
pipelines, we use median inference time on calibration subset.
NOTE: For the most accurate performance estimation, it is recommended to run
benchmark_appin a terminal/command prompt after closing other applications.
%%skip not $to_quantize.value
import time
def calculate_inference_time(pipeline, validation_data):
inference_time = []
for prompt in validation_data:
start = time.perf_counter()
with io.capture_output() as captured:
_ = pipeline(
image,
num_inference_steps=4,
motion_bucket_id=60,
num_frames=8,
height=320,
width=512,
generator=torch.manual_seed(12342),
)
end = time.perf_counter()
delta = end - start
inference_time.append(delta)
return np.median(inference_time)
%%skip not $to_quantize.value
validation_size = 3
validation_dataset = datasets.load_dataset("fusing/instructpix2pix-1000-samples", split="train", streaming=True).shuffle(seed=42).take(validation_size)
validation_data = [data["input_image"] for data in validation_dataset]
fp_latency = calculate_inference_time(ov_pipe, validation_data)
int8_latency = calculate_inference_time(ov_int8_pipeline, validation_data)
print(f"Performance speed-up: {fp_latency / int8_latency:.3f}")
Performance speed-up: 1.243
Interactive Demo#
Please select below whether you would like to use the quantized model to launch the interactive demo.
import ipywidgets as widgets
quantized_model_present = ov_int8_pipeline is not None
use_quantized_model = widgets.Checkbox(
value=quantized_model_present,
description="Use quantized model",
disabled=not quantized_model_present,
)
use_quantized_model
Checkbox(value=True, description='Use quantized model')
if not Path("gradio_helper.py").exists():
r = requests.get(url="https://raw.githubusercontent.com/openvinotoolkit/openvino_notebooks/latest/notebooks/stable-video-diffusion/gradio_helper.py")
open("gradio_helper.py", "w").write(r.text)
from gradio_helper import make_demo
pipeline = ov_int8_pipeline if use_quantized_model.value else ov_pipe
demo = make_demo(pipeline)
try:
demo.queue().launch(debug=False)
except Exception:
demo.queue().launch(debug=False, share=True)
# if you are launching remotely, specify server_name and server_port
# demo.launch(server_name='your server name', server_port='server port in int')
# Read more in the docs: https://gradio.app/docs/