File size: 6,013 Bytes
0dcccdd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
import json
import os

import cv2
import numpy as np
import torch
from diffusers import (AutoencoderKL, CogVideoXDDIMScheduler, DDIMScheduler,
                       DPMSolverMultistepScheduler,
                       EulerAncestralDiscreteScheduler, EulerDiscreteScheduler,
                       PNDMScheduler)
from omegaconf import OmegaConf
from PIL import Image
from transformers import (CLIPImageProcessor, CLIPVisionModelWithProjection,
                          T5EncoderModel, T5Tokenizer)

from cogvideox.models.autoencoder_magvit import AutoencoderKLCogVideoX
from cogvideox.models.transformer3d import CogVideoXTransformer3DModel
from cogvideox.pipeline.pipeline_cogvideox import CogVideoX_Fun_Pipeline
from cogvideox.pipeline.pipeline_cogvideox_control import \
    CogVideoX_Fun_Pipeline_Control
from cogvideox.utils.lora_utils import merge_lora, unmerge_lora
from cogvideox.utils.utils import get_video_to_video_latent, save_videos_grid

# Low gpu memory mode, this is used when the GPU memory is under 16GB
low_gpu_memory_mode = False

# model path
model_name          = "models/Diffusion_Transformer/CogVideoX-Fun-V1.1-2b-Pose"

# Choose the sampler in "Euler" "Euler A" "DPM++" "PNDM" and "DDIM"
sampler_name        = "DDIM_Origin"

# Load pretrained model if need
transformer_path    = None
vae_path            = None
lora_path           = None
# Other params
sample_size         = [672, 384]
video_length        = 49
fps                 = 8

# Use torch.float16 if GPU does not support torch.bfloat16
# ome graphics cards, such as v100, 2080ti, do not support torch.bfloat16
weight_dtype            = torch.bfloat16
control_video           = "asset/pose.mp4"

# prompts
prompt                  = "A person wearing a knee-length white sleeveless dress and white high-heeled sandals performs a dance in a well-lit room with wooden flooring. The room's background features a closed door, a shelf displaying clear glass bottles of alcoholic beverages, and a partially visible dark-colored sofa. "
negative_prompt         = "The video is not of a high quality, it has a low resolution. Watermark present in each frame. The background is solid. Strange body and strange trajectory. Distortion. "
guidance_scale          = 6.0
seed                    = 43
num_inference_steps     = 50
lora_weight             = 0.55
save_path               = "samples/cogvideox-fun-videos_control"

transformer = CogVideoXTransformer3DModel.from_pretrained_2d(
    model_name, 
    subfolder="transformer",
).to(weight_dtype)

if transformer_path is not None:
    print(f"From checkpoint: {transformer_path}")
    if transformer_path.endswith("safetensors"):
        from safetensors.torch import load_file, safe_open
        state_dict = load_file(transformer_path)
    else:
        state_dict = torch.load(transformer_path, map_location="cpu")
    state_dict = state_dict["state_dict"] if "state_dict" in state_dict else state_dict

    m, u = transformer.load_state_dict(state_dict, strict=False)
    print(f"missing keys: {len(m)}, unexpected keys: {len(u)}")

# Get Vae
vae = AutoencoderKLCogVideoX.from_pretrained(
    model_name, 
    subfolder="vae"
).to(weight_dtype)

if vae_path is not None:
    print(f"From checkpoint: {vae_path}")
    if vae_path.endswith("safetensors"):
        from safetensors.torch import load_file, safe_open
        state_dict = load_file(vae_path)
    else:
        state_dict = torch.load(vae_path, map_location="cpu")
    state_dict = state_dict["state_dict"] if "state_dict" in state_dict else state_dict

    m, u = vae.load_state_dict(state_dict, strict=False)
    print(f"missing keys: {len(m)}, unexpected keys: {len(u)}")

text_encoder = T5EncoderModel.from_pretrained(
    model_name, subfolder="text_encoder", torch_dtype=weight_dtype
)
# Get Scheduler
Choosen_Scheduler = scheduler_dict = {
    "Euler": EulerDiscreteScheduler,
    "Euler A": EulerAncestralDiscreteScheduler,
    "DPM++": DPMSolverMultistepScheduler, 
    "PNDM": PNDMScheduler,
    "DDIM_Cog": CogVideoXDDIMScheduler,
    "DDIM_Origin": DDIMScheduler,
}[sampler_name]
scheduler = Choosen_Scheduler.from_pretrained(
    model_name, 
    subfolder="scheduler"
)

pipeline = CogVideoX_Fun_Pipeline_Control.from_pretrained(
    model_name,
    vae=vae,
    text_encoder=text_encoder,
    transformer=transformer,
    scheduler=scheduler,
    torch_dtype=weight_dtype
)

if low_gpu_memory_mode:
    pipeline.enable_sequential_cpu_offload()
else:
    pipeline.enable_model_cpu_offload()

generator = torch.Generator(device="cuda").manual_seed(seed)

if lora_path is not None:
    pipeline = merge_lora(pipeline, lora_path, lora_weight, "cuda")

video_length = int((video_length - 1) // vae.config.temporal_compression_ratio * vae.config.temporal_compression_ratio) + 1 if video_length != 1 else 1
input_video, input_video_mask, clip_image = get_video_to_video_latent(control_video, video_length=video_length, sample_size=sample_size, fps=fps)

with torch.no_grad():
    sample = pipeline(
        prompt, 
        num_frames = video_length,
        negative_prompt = negative_prompt,
        height      = sample_size[0],
        width       = sample_size[1],
        generator   = generator,
        guidance_scale = guidance_scale,
        num_inference_steps = num_inference_steps,

        control_video = input_video,
    ).videos

if lora_path is not None:
    pipeline = unmerge_lora(pipeline, lora_path, lora_weight, "cuda")
    
if not os.path.exists(save_path):
    os.makedirs(save_path, exist_ok=True)

index = len([path for path in os.listdir(save_path)]) + 1
prefix = str(index).zfill(8)
    
if video_length == 1:
    save_sample_path = os.path.join(save_path, prefix + f".png")

    image = sample[0, :, 0]
    image = image.transpose(0, 1).transpose(1, 2)
    image = (image * 255).numpy().astype(np.uint8)
    image = Image.fromarray(image)
    image.save(save_sample_path)
else:
    video_path = os.path.join(save_path, prefix + ".mp4")
    save_videos_grid(sample, video_path, fps=fps)