Spaces:
Running
on
L40S
Running
on
L40S
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,173 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from huggingface_hub import snapshot_download
|
2 |
+
|
3 |
+
# Download All Required Models using `snapshot_download`
|
4 |
+
|
5 |
+
# Download Wan2.1-I2V-14B-480P model
|
6 |
+
wan_model_path = snapshot_download(
|
7 |
+
repo_id="Wan-AI/Wan2.1-I2V-14B-480P",
|
8 |
+
local_dir="./weights/Wan2.1-I2V-14B-480P",
|
9 |
+
#local_dir_use_symlinks=False
|
10 |
+
)
|
11 |
+
|
12 |
+
# Download Chinese wav2vec2 model
|
13 |
+
wav2vec_path = snapshot_download(
|
14 |
+
repo_id="TencentGameMate/chinese-wav2vec2-base",
|
15 |
+
local_dir="./weights/chinese-wav2vec2-base",
|
16 |
+
#local_dir_use_symlinks=False
|
17 |
+
)
|
18 |
+
|
19 |
+
# Download MeiGen MultiTalk weights
|
20 |
+
multitalk_path = snapshot_download(
|
21 |
+
repo_id="MeiGen-AI/MeiGen-MultiTalk",
|
22 |
+
local_dir="./weights/MeiGen-MultiTalk",
|
23 |
+
#local_dir_use_symlinks=False
|
24 |
+
)
|
25 |
+
|
26 |
+
|
27 |
+
import os
|
28 |
+
import shutil
|
29 |
+
|
30 |
+
# Define paths
|
31 |
+
base_model_dir = "./weights/Wan2.1-I2V-14B-480P"
|
32 |
+
multitalk_dir = "./weights/MeiGen-MultiTalk"
|
33 |
+
|
34 |
+
# File to rename
|
35 |
+
original_index = os.path.join(base_model_dir, "diffusion_pytorch_model.safetensors.index.json")
|
36 |
+
backup_index = os.path.join(base_model_dir, "diffusion_pytorch_model.safetensors.index.json_old")
|
37 |
+
|
38 |
+
# Rename the original index file
|
39 |
+
if os.path.exists(original_index):
|
40 |
+
os.rename(original_index, backup_index)
|
41 |
+
print("Renamed original index file to .json_old")
|
42 |
+
|
43 |
+
# Copy updated index file from MultiTalk
|
44 |
+
shutil.copy2(
|
45 |
+
os.path.join(multitalk_dir, "diffusion_pytorch_model.safetensors.index.json"),
|
46 |
+
base_model_dir
|
47 |
+
)
|
48 |
+
|
49 |
+
# Copy MultiTalk model weights
|
50 |
+
shutil.copy2(
|
51 |
+
os.path.join(multitalk_dir, "multitalk.safetensors"),
|
52 |
+
base_model_dir
|
53 |
+
)
|
54 |
+
|
55 |
+
print("Copied MultiTalk files into base model directory.")
|
56 |
+
|
57 |
+
|
58 |
+
import torch
|
59 |
+
|
60 |
+
# Check if CUDA-compatible GPU is available
|
61 |
+
if torch.cuda.is_available():
|
62 |
+
# Get current GPU name
|
63 |
+
gpu_name = torch.cuda.get_device_name(torch.cuda.current_device())
|
64 |
+
print(f"Current GPU: {gpu_name}")
|
65 |
+
|
66 |
+
# Enforce GPU requirement
|
67 |
+
if "A100" not in gpu_name and "L4" not in gpu_name:
|
68 |
+
raise RuntimeError(f"This notebook requires an A100 or L4 GPU. Found: {gpu_name}")
|
69 |
+
elif "L4" in gpu_name:
|
70 |
+
print("Warning: L4 is supported, but A100 is recommended for faster inference.")
|
71 |
+
else:
|
72 |
+
raise RuntimeError("No CUDA-compatible GPU found. An A100 or L4 GPU is required.")
|
73 |
+
|
74 |
+
|
75 |
+
GPU_TO_VRAM_PARAMS = {
|
76 |
+
"NVIDIA A100": 11000000000,
|
77 |
+
"NVIDIA A100-SXM4-40GB": 11000000000,
|
78 |
+
"NVIDIA A100-SXM4-80GB": 22000000000,
|
79 |
+
"NVIDIA L4": 5000000000
|
80 |
+
}
|
81 |
+
USED_VRAM_PARAMS = GPU_TO_VRAM_PARAMS[gpu_name]
|
82 |
+
print("Using", USED_VRAM_PARAMS, "for num_persistent_param_in_dit")
|
83 |
+
|
84 |
+
import subprocess
|
85 |
+
|
86 |
+
import json
|
87 |
+
import tempfile
|
88 |
+
#import os
|
89 |
+
|
90 |
+
def create_temp_input_json(prompt: str, cond_image_path: str, cond_audio_path: str) -> str:
|
91 |
+
"""
|
92 |
+
Create a temporary JSON file with the user-provided prompt, image, and audio paths.
|
93 |
+
Returns the path to the temporary JSON file.
|
94 |
+
"""
|
95 |
+
# Structure based on your original JSON format
|
96 |
+
data = {
|
97 |
+
"prompt": prompt,
|
98 |
+
"cond_image": cond_image_path,
|
99 |
+
"cond_audio": {
|
100 |
+
"person1": cond_audio_path
|
101 |
+
}
|
102 |
+
}
|
103 |
+
|
104 |
+
# Create a temp file
|
105 |
+
temp_json = tempfile.NamedTemporaryFile(delete=False, suffix=".json", mode='w', encoding='utf-8')
|
106 |
+
json.dump(data, temp_json, indent=4)
|
107 |
+
temp_json_path = temp_json.name
|
108 |
+
temp_json.close()
|
109 |
+
|
110 |
+
print(f"Temporary input JSON saved to: {temp_json_path}")
|
111 |
+
return temp_json_path
|
112 |
+
|
113 |
+
|
114 |
+
def infer(prompt, cond_image_path, cond_audio_path):
|
115 |
+
|
116 |
+
# Example usage (from user input)
|
117 |
+
prompt = "A woman sings passionately in a dimly lit studio."
|
118 |
+
cond_image_path = "examples/single/single1.png" # Assume uploaded via Gradio
|
119 |
+
cond_audio_path = "examples/single/1.wav" # Assume uploaded via Gradio
|
120 |
+
|
121 |
+
input_json_path = create_temp_input_json(prompt, cond_image_path, cond_audio_path)
|
122 |
+
|
123 |
+
cmd = [
|
124 |
+
"python3", "generate_multitalk.py",
|
125 |
+
"--ckpt_dir", "weights/Wan2.1-I2V-14B-480P",
|
126 |
+
"--wav2vec_dir", "weights/chinese-wav2vec2-base",
|
127 |
+
"--input_json", "./examples/single_example_1.json",
|
128 |
+
"--sample_steps", "20",
|
129 |
+
"--num_persistent_param_in_dit", USED_VRAM_PARAMS,
|
130 |
+
"--mode", "streaming",
|
131 |
+
"--use_teacache",
|
132 |
+
"--save_file", "multi_long_mediumvram_exp"
|
133 |
+
]
|
134 |
+
|
135 |
+
subprocess.run(cmd, check=True)
|
136 |
+
|
137 |
+
return "multi_long_mediumvra_exp.mp4"
|
138 |
+
|
139 |
+
import gradio as gr
|
140 |
+
|
141 |
+
with gr.Blocks(title="MultiTalk Inference") as demo:
|
142 |
+
gr.Markdown("## 🎤 MultiTalk Inference Demo")
|
143 |
+
|
144 |
+
with gr.Row():
|
145 |
+
with gr.Column():
|
146 |
+
prompt_input = gr.Textbox(
|
147 |
+
label="Text Prompt",
|
148 |
+
placeholder="Describe the scene...",
|
149 |
+
lines=4
|
150 |
+
)
|
151 |
+
|
152 |
+
image_input = gr.Image(
|
153 |
+
type="filepath",
|
154 |
+
label="Conditioning Image"
|
155 |
+
)
|
156 |
+
|
157 |
+
audio_input = gr.Audio(
|
158 |
+
type="filepath",
|
159 |
+
label="Conditioning Audio (.wav)"
|
160 |
+
)
|
161 |
+
|
162 |
+
submit_btn = gr.Button("Generate")
|
163 |
+
|
164 |
+
with gr.Column():
|
165 |
+
output_video = gr.Video(label="Generated Video")
|
166 |
+
|
167 |
+
submit_btn.click(
|
168 |
+
fn=infer,
|
169 |
+
inputs=[prompt_input, image_input, audio_input],
|
170 |
+
outputs=output_video
|
171 |
+
)
|
172 |
+
|
173 |
+
demo.launch()
|