Wan2GP / wan /image2video.py
zxymimi23451's picture
Upload 258 files
78360e7 verified
# Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
import gc
import logging
import math
import os
import random
import sys
import types
from contextlib import contextmanager
from functools import partial
import json
import numpy as np
import torch
import torch.cuda.amp as amp
import torch.distributed as dist
import torchvision.transforms.functional as TF
from tqdm import tqdm
from .distributed.fsdp import shard_model
from .modules.clip import CLIPModel
from .modules.model import WanModel
from .modules.t5 import T5EncoderModel
from .modules.vae import WanVAE
from .utils.fm_solvers import (FlowDPMSolverMultistepScheduler,
get_sampling_sigmas, retrieve_timesteps)
from .utils.fm_solvers_unipc import FlowUniPCMultistepScheduler
from wan.modules.posemb_layers import get_rotary_pos_embed
from wan.utils.utils import resize_lanczos, calculate_new_dimensions
def optimized_scale(positive_flat, negative_flat):
# Calculate dot production
dot_product = torch.sum(positive_flat * negative_flat, dim=1, keepdim=True)
# Squared norm of uncondition
squared_norm = torch.sum(negative_flat ** 2, dim=1, keepdim=True) + 1e-8
# st_star = v_cond^T * v_uncond / ||v_uncond||^2
st_star = dot_product / squared_norm
return st_star
class WanI2V:
def __init__(
self,
config,
checkpoint_dir,
model_filename = None,
model_type = None,
base_model_type= None,
text_encoder_filename= None,
quantizeTransformer = False,
dtype = torch.bfloat16,
VAE_dtype = torch.float32,
save_quantized = False,
mixed_precision_transformer = False
):
self.device = torch.device(f"cuda")
self.config = config
self.dtype = dtype
self.VAE_dtype = VAE_dtype
self.num_train_timesteps = config.num_train_timesteps
self.param_dtype = config.param_dtype
# shard_fn = partial(shard_model, device_id=device_id)
self.text_encoder = T5EncoderModel(
text_len=config.text_len,
dtype=config.t5_dtype,
device=torch.device('cpu'),
checkpoint_path=text_encoder_filename,
tokenizer_path=os.path.join(checkpoint_dir, config.t5_tokenizer),
shard_fn=None,
)
self.vae_stride = config.vae_stride
self.patch_size = config.patch_size
self.vae = WanVAE(
vae_pth=os.path.join(checkpoint_dir, config.vae_checkpoint), dtype = VAE_dtype,
device=self.device)
self.clip = CLIPModel(
dtype=config.clip_dtype,
device=self.device,
checkpoint_path=os.path.join(checkpoint_dir ,
config.clip_checkpoint),
tokenizer_path=os.path.join(checkpoint_dir , config.clip_tokenizer))
logging.info(f"Creating WanModel from {model_filename[-1]}")
from mmgp import offload
# fantasy = torch.load("c:/temp/fantasy.ckpt")
# proj_model = fantasy["proj_model"]
# audio_processor = fantasy["audio_processor"]
# offload.safetensors2.torch_write_file(proj_model, "proj_model.safetensors")
# offload.safetensors2.torch_write_file(audio_processor, "audio_processor.safetensors")
# for k,v in audio_processor.items():
# audio_processor[k] = v.to(torch.bfloat16)
# with open("fantasy_config.json", "r", encoding="utf-8") as reader:
# config_text = reader.read()
# config_json = json.loads(config_text)
# offload.safetensors2.torch_write_file(audio_processor, "audio_processor_bf16.safetensors", config=config_json)
# model_filename = [model_filename, "audio_processor_bf16.safetensors"]
# model_filename = "c:/temp/i2v480p/diffusion_pytorch_model-00001-of-00007.safetensors"
# dtype = torch.float16
base_config_file = f"configs/{base_model_type}.json"
forcedConfigPath = base_config_file if len(model_filename) > 1 else None
self.model = offload.fast_load_transformers_model(model_filename, modelClass=WanModel,do_quantize= quantizeTransformer and not save_quantized, writable_tensors= False, defaultConfigPath= base_config_file, forcedConfigPath= forcedConfigPath)
self.model.lock_layers_dtypes(torch.float32 if mixed_precision_transformer else dtype)
offload.change_dtype(self.model, dtype, True)
# offload.save_model(self.model, "wan2.1_image2video_720p_14B_mbf16.safetensors", config_file_path="c:/temp/i2v720p/config.json")
# offload.save_model(self.model, "wan2.1_image2video_720p_14B_quanto_mbf16_int8.safetensors",do_quantize=True, config_file_path="c:/temp/i2v720p/config.json")
# offload.save_model(self.model, "wan2.1_image2video_720p_14B_quanto_mfp16_int8.safetensors",do_quantize=True, config_file_path="c:/temp/i2v720p/config.json")
# offload.save_model(self.model, "wan2.1_Fun_InP_1.3B_bf16_bis.safetensors")
self.model.eval().requires_grad_(False)
if save_quantized:
from wgp import save_quantized_model
save_quantized_model(self.model, model_type, model_filename[0], dtype, base_config_file)
self.sample_neg_prompt = config.sample_neg_prompt
def generate(self,
input_prompt,
image_start,
image_end = None,
height =720,
width = 1280,
fit_into_canvas = True,
frame_num=81,
shift=5.0,
sample_solver='unipc',
sampling_steps=40,
guide_scale=5.0,
n_prompt="",
seed=-1,
callback = None,
enable_RIFLEx = False,
VAE_tile_size= 0,
joint_pass = False,
slg_layers = None,
slg_start = 0.0,
slg_end = 1.0,
cfg_star_switch = True,
cfg_zero_step = 5,
audio_scale=None,
audio_cfg_scale=None,
audio_proj=None,
audio_context_lens=None,
model_filename = None,
**bbargs
):
r"""
Generates video frames from input image and text prompt using diffusion process.
Args:
input_prompt (`str`):
Text prompt for content generation.
image_start (PIL.Image.Image):
Input image tensor. Shape: [3, H, W]
max_area (`int`, *optional*, defaults to 720*1280):
Maximum pixel area for latent space calculation. Controls video resolution scaling
frame_num (`int`, *optional*, defaults to 81):
How many frames to sample from a video. The number should be 4n+1
shift (`float`, *optional*, defaults to 5.0):
Noise schedule shift parameter. Affects temporal dynamics
[NOTE]: If you want to generate a 480p video, it is recommended to set the shift value to 3.0.
sample_solver (`str`, *optional*, defaults to 'unipc'):
Solver used to sample the video.
sampling_steps (`int`, *optional*, defaults to 40):
Number of diffusion sampling steps. Higher values improve quality but slow generation
guide_scale (`float`, *optional*, defaults 5.0):
Classifier-free guidance scale. Controls prompt adherence vs. creativity
n_prompt (`str`, *optional*, defaults to ""):
Negative prompt for content exclusion. If not given, use `config.sample_neg_prompt`
seed (`int`, *optional*, defaults to -1):
Random seed for noise generation. If -1, use random seed
offload_model (`bool`, *optional*, defaults to True):
If True, offloads models to CPU during generation to save VRAM
Returns:
torch.Tensor:
Generated video frames tensor. Dimensions: (C, N H, W) where:
- C: Color channels (3 for RGB)
- N: Number of frames (81)
- H: Frame height (from max_area)
- W: Frame width from max_area)
"""
add_frames_for_end_image = "image2video" in model_filename or "fantasy" in model_filename
image_start = TF.to_tensor(image_start)
lat_frames = int((frame_num - 1) // self.vae_stride[0] + 1)
any_end_frame = image_end !=None
if any_end_frame:
any_end_frame = True
image_end = TF.to_tensor(image_end)
if add_frames_for_end_image:
frame_num +=1
lat_frames = int((frame_num - 2) // self.vae_stride[0] + 2)
h, w = image_start.shape[1:]
h, w = calculate_new_dimensions(height, width, h, w, fit_into_canvas)
lat_h = round(
h // self.vae_stride[1] //
self.patch_size[1] * self.patch_size[1])
lat_w = round(
w // self.vae_stride[2] //
self.patch_size[2] * self.patch_size[2])
h = lat_h * self.vae_stride[1]
w = lat_w * self.vae_stride[2]
clip_image_size = self.clip.model.image_size
img_interpolated = resize_lanczos(image_start, h, w).sub_(0.5).div_(0.5).unsqueeze(0).transpose(0,1).to(self.device) #, self.dtype
image_start = resize_lanczos(image_start, clip_image_size, clip_image_size)
image_start = image_start.sub_(0.5).div_(0.5).to(self.device) #, self.dtype
if image_end!= None:
img_interpolated2 = resize_lanczos(image_end, h, w).sub_(0.5).div_(0.5).unsqueeze(0).transpose(0,1).to(self.device) #, self.dtype
image_end = resize_lanczos(image_end, clip_image_size, clip_image_size)
image_end = image_end.sub_(0.5).div_(0.5).to(self.device) #, self.dtype
max_seq_len = lat_frames * lat_h * lat_w // ( self.patch_size[1] * self.patch_size[2])
seed = seed if seed >= 0 else random.randint(0, sys.maxsize)
seed_g = torch.Generator(device=self.device)
seed_g.manual_seed(seed)
noise = torch.randn(16, lat_frames, lat_h, lat_w, dtype=torch.float32, generator=seed_g, device=self.device)
msk = torch.ones(1, frame_num, lat_h, lat_w, device=self.device)
if any_end_frame:
msk[:, 1: -1] = 0
if add_frames_for_end_image:
msk = torch.concat([ torch.repeat_interleave(msk[:, 0:1], repeats=4, dim=1), msk[:, 1:-1], torch.repeat_interleave(msk[:, -1:], repeats=4, dim=1) ], dim=1)
else:
msk = torch.concat([ torch.repeat_interleave(msk[:, 0:1], repeats=4, dim=1), msk[:, 1:] ], dim=1)
else:
msk[:, 1:] = 0
msk = torch.concat([ torch.repeat_interleave(msk[:, 0:1], repeats=4, dim=1), msk[:, 1:] ], dim=1)
msk = msk.view(1, msk.shape[1] // 4, 4, lat_h, lat_w)
msk = msk.transpose(1, 2)[0]
if n_prompt == "":
n_prompt = self.sample_neg_prompt
if self._interrupt:
return None
# preprocess
context = self.text_encoder([input_prompt], self.device)[0]
context_null = self.text_encoder([n_prompt], self.device)[0]
context = context.to(self.dtype)
context_null = context_null.to(self.dtype)
if self._interrupt:
return None
clip_context = self.clip.visual([image_start[:, None, :, :]])
from mmgp import offload
offload.last_offload_obj.unload_all()
if any_end_frame:
mean2 = 0
enc= torch.concat([
img_interpolated,
torch.full( (3, frame_num-2, h, w), mean2, device=self.device, dtype= self.VAE_dtype),
img_interpolated2,
], dim=1).to(self.device)
else:
enc= torch.concat([
img_interpolated,
torch.zeros(3, frame_num-1, h, w, device=self.device, dtype= self.VAE_dtype)
], dim=1).to(self.device)
image_start, image_end, img_interpolated, img_interpolated2 = None, None, None, None
lat_y = self.vae.encode([enc], VAE_tile_size, any_end_frame= any_end_frame and add_frames_for_end_image)[0]
y = torch.concat([msk, lat_y])
lat_y = None
# evaluation mode
if sample_solver == 'unipc':
sample_scheduler = FlowUniPCMultistepScheduler(
num_train_timesteps=self.num_train_timesteps,
shift=1,
use_dynamic_shifting=False)
sample_scheduler.set_timesteps(
sampling_steps, device=self.device, shift=shift)
timesteps = sample_scheduler.timesteps
elif sample_solver == 'dpm++':
sample_scheduler = FlowDPMSolverMultistepScheduler(
num_train_timesteps=self.num_train_timesteps,
shift=1,
use_dynamic_shifting=False)
sampling_sigmas = get_sampling_sigmas(sampling_steps, shift)
timesteps, _ = retrieve_timesteps(
sample_scheduler,
device=self.device,
sigmas=sampling_sigmas)
else:
raise NotImplementedError("Unsupported solver.")
# sample videos
latent = noise
batch_size = 1
freqs = get_rotary_pos_embed(latent.shape[1:], enable_RIFLEx= enable_RIFLEx)
kwargs = { 'clip_fea': clip_context, 'y': y, 'freqs' : freqs, 'pipeline' : self, 'callback' : callback }
if audio_proj != None:
kwargs.update({
"audio_proj": audio_proj.to(self.dtype),
"audio_context_lens": audio_context_lens,
})
if self.model.enable_cache:
self.model.previous_residual = [None] * (3 if audio_cfg_scale !=None else 2)
self.model.compute_teacache_threshold(self.model.cache_start_step, timesteps, self.model.teacache_multiplier)
# self.model.to(self.device)
if callback != None:
callback(-1, None, True)
latent = latent.to(self.device)
for i, t in enumerate(tqdm(timesteps)):
offload.set_step_no_for_lora(self.model, i)
kwargs["slg_layers"] = slg_layers if int(slg_start * sampling_steps) <= i < int(slg_end * sampling_steps) else None
latent_model_input = latent
timestep = [t]
timestep = torch.stack(timestep).to(self.device)
kwargs.update({
't' :timestep,
'current_step' :i,
})
if guide_scale == 1:
noise_pred = self.model( [latent_model_input], context=[context], audio_scale = None if audio_scale == None else [audio_scale], x_id=0, **kwargs, )[0]
if self._interrupt:
return None
elif joint_pass:
if audio_proj == None:
noise_pred_cond, noise_pred_uncond = self.model(
[latent_model_input, latent_model_input],
context=[context, context_null],
**kwargs)
else:
noise_pred_cond, noise_pred_noaudio, noise_pred_uncond = self.model(
[latent_model_input, latent_model_input, latent_model_input],
context=[context, context, context_null],
audio_scale = [audio_scale, None, None ],
**kwargs)
if self._interrupt:
return None
else:
noise_pred_cond = self.model( [latent_model_input], context=[context], audio_scale = None if audio_scale == None else [audio_scale], x_id=0, **kwargs, )[0]
if self._interrupt:
return None
if audio_proj != None:
noise_pred_noaudio = self.model(
[latent_model_input],
x_id=1,
context=[context],
**kwargs,
)[0]
if self._interrupt:
return None
noise_pred_uncond = self.model(
[latent_model_input],
x_id=1 if audio_scale == None else 2,
context=[context_null],
**kwargs,
)[0]
if self._interrupt:
return None
del latent_model_input
if guide_scale > 1:
# CFG Zero *. Thanks to https://github.com/WeichenFan/CFG-Zero-star/
if cfg_star_switch:
positive_flat = noise_pred_cond.view(batch_size, -1)
negative_flat = noise_pred_uncond.view(batch_size, -1)
alpha = optimized_scale(positive_flat,negative_flat)
alpha = alpha.view(batch_size, 1, 1, 1)
if (i <= cfg_zero_step):
noise_pred = noise_pred_cond*0. # it would be faster not to compute noise_pred...
else:
noise_pred_uncond *= alpha
if audio_scale == None:
noise_pred = noise_pred_uncond + guide_scale * (noise_pred_cond - noise_pred_uncond)
else:
noise_pred = noise_pred_uncond + guide_scale * (noise_pred_noaudio - noise_pred_uncond) + audio_cfg_scale * (noise_pred_cond - noise_pred_noaudio)
noise_pred_uncond, noise_pred_noaudio = None, None
temp_x0 = sample_scheduler.step(
noise_pred.unsqueeze(0),
t,
latent.unsqueeze(0),
return_dict=False,
generator=seed_g)[0]
latent = temp_x0.squeeze(0)
del temp_x0
del timestep
if callback is not None:
callback(i, latent, False)
x0 = [latent]
video = self.vae.decode(x0, VAE_tile_size, any_end_frame= any_end_frame and add_frames_for_end_image)[0]
if any_end_frame and add_frames_for_end_image:
# video[:, -1:] = img_interpolated2
video = video[:, :-1]
del noise, latent
del sample_scheduler
return video