Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,215 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# app.py
|
2 |
+
|
3 |
+
import os
|
4 |
+
import sys
|
5 |
+
import time
|
6 |
+
import gradio as gr
|
7 |
+
from huggingface_hub import snapshot_download
|
8 |
+
from huggingface_hub.utils import GatedRepoError, RepositoryNotFoundError, RevisionNotFoundError
|
9 |
+
from pathlib import Path
|
10 |
+
import spaces
|
11 |
+
|
12 |
+
# Add the src directory to the system path to allow for local imports
|
13 |
+
sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), 'src')))
|
14 |
+
|
15 |
+
from models.inference.moda_test import LiveVASAPipeline, emo_map, set_seed
|
16 |
+
|
17 |
+
# --- Configuration ---
|
18 |
+
# Set seed for reproducibility
|
19 |
+
set_seed(42)
|
20 |
+
|
21 |
+
# Paths and constants for the Gradio demo
|
22 |
+
DEFAULT_CFG_PATH = "configs/audio2motion/inference/inference.yaml"
|
23 |
+
DEFAULT_MOTION_MEAN_STD_PATH = "src/datasets/mean.pt"
|
24 |
+
DEFAULT_SILENT_AUDIO_PATH = "src/examples/silent-audio.wav"
|
25 |
+
OUTPUT_DIR = "gradio_output"
|
26 |
+
WEIGHTS_DIR = "pretrain_weights"
|
27 |
+
REPO_ID = "lixinyizju/moda"
|
28 |
+
|
29 |
+
# --- Download Pre-trained Weights from Hugging Face Hub ---
|
30 |
+
def download_weights():
|
31 |
+
"""
|
32 |
+
Downloads pre-trained weights from Hugging Face Hub if they don't exist locally.
|
33 |
+
"""
|
34 |
+
# A simple check for a key file to see if the download is likely complete
|
35 |
+
motion_model_file = os.path.join(WEIGHTS_DIR, "moda", "net-200.pth")
|
36 |
+
|
37 |
+
if not os.path.exists(motion_model_file):
|
38 |
+
print(f"Weights not found locally. Downloading from Hugging Face Hub repo '{REPO_ID}'...")
|
39 |
+
print(f"This may take a while depending on your internet connection.")
|
40 |
+
try:
|
41 |
+
snapshot_download(
|
42 |
+
repo_id=REPO_ID,
|
43 |
+
local_dir=WEIGHTS_DIR,
|
44 |
+
local_dir_use_symlinks=False, # Use False to copy files directly; safer for Windows
|
45 |
+
resume_download=True,
|
46 |
+
)
|
47 |
+
print("Weights downloaded successfully.")
|
48 |
+
except GatedRepoError:
|
49 |
+
raise gr.Error(f"Access to the repository '{REPO_ID}' is gated. Please visit https://huggingface.co/{REPO_ID} to request access.")
|
50 |
+
except (RepositoryNotFoundError, RevisionNotFoundError):
|
51 |
+
raise gr.Error(f"The repository '{REPO_ID}' was not found. Please check the repository ID.")
|
52 |
+
except Exception as e:
|
53 |
+
print(f"An error occurred during download: {e}")
|
54 |
+
raise gr.Error(f"Failed to download models. Please check your internet connection and try again. Error: {e}")
|
55 |
+
else:
|
56 |
+
print(f"Found existing weights at '{WEIGHTS_DIR}'. Skipping download.")
|
57 |
+
|
58 |
+
|
59 |
+
# --- Initialization ---
|
60 |
+
# Create output directory if it doesn't exist
|
61 |
+
os.makedirs(OUTPUT_DIR, exist_ok=True)
|
62 |
+
|
63 |
+
# Download weights before initializing the pipeline
|
64 |
+
download_weights()
|
65 |
+
|
66 |
+
# Instantiate the pipeline once to avoid reloading models on every request
|
67 |
+
print("Initializing MoDA pipeline...")
|
68 |
+
try:
|
69 |
+
pipeline = LiveVASAPipeline(
|
70 |
+
cfg_path=DEFAULT_CFG_PATH,
|
71 |
+
motion_mean_std_path=DEFAULT_MOTION_MEAN_STD_PATH
|
72 |
+
)
|
73 |
+
print("MoDA pipeline initialized successfully.")
|
74 |
+
except Exception as e:
|
75 |
+
print(f"Error initializing pipeline: {e}")
|
76 |
+
pipeline = None
|
77 |
+
|
78 |
+
# Invert the emo_map for easy lookup from the dropdown value
|
79 |
+
emo_name_to_id = {v: k for k, v in emo_map.items()}
|
80 |
+
|
81 |
+
# --- Core Generation Function ---
|
82 |
+
@spaces.GPU
|
83 |
+
def generate_motion(source_image_path, driving_audio_path, emotion_name, cfg_scale, progress=gr.Progress(track_tqdm=True)):
|
84 |
+
"""
|
85 |
+
The main function that takes Gradio inputs and generates the talking head video.
|
86 |
+
"""
|
87 |
+
if pipeline is None:
|
88 |
+
raise gr.Error("Pipeline failed to initialize. Check the console logs for details.")
|
89 |
+
|
90 |
+
if source_image_path is None:
|
91 |
+
raise gr.Error("Please upload a source image.")
|
92 |
+
if driving_audio_path is None:
|
93 |
+
raise gr.Error("Please upload a driving audio file.")
|
94 |
+
|
95 |
+
start_time = time.time()
|
96 |
+
|
97 |
+
# Create a unique subdirectory for this run
|
98 |
+
timestamp = time.strftime("%Y%m%d-%H%M%S")
|
99 |
+
run_output_dir = os.path.join(OUTPUT_DIR, timestamp)
|
100 |
+
os.makedirs(run_output_dir, exist_ok=True)
|
101 |
+
|
102 |
+
# Get emotion ID from its name
|
103 |
+
emotion_id = emo_name_to_id.get(emotion_name, 8) # Default to 'None' (ID 8) if not found
|
104 |
+
|
105 |
+
print(f"Starting generation with the following parameters:")
|
106 |
+
print(f" Source Image: {source_image_path}")
|
107 |
+
print(f" Driving Audio: {driving_audio_path}")
|
108 |
+
print(f" Emotion: {emotion_name} (ID: {emotion_id})")
|
109 |
+
print(f" CFG Scale: {cfg_scale}")
|
110 |
+
|
111 |
+
try:
|
112 |
+
# Call the pipeline's inference method
|
113 |
+
result_video_path = pipeline.driven_sample(
|
114 |
+
image_path=source_image_path,
|
115 |
+
audio_path=driving_audio_path,
|
116 |
+
cfg_scale=float(cfg_scale),
|
117 |
+
emo=emotion_id,
|
118 |
+
save_dir=".",
|
119 |
+
smooth=False, # Smoothing can be slow, disable for a faster demo
|
120 |
+
silent_audio_path=DEFAULT_SILENT_AUDIO_PATH,
|
121 |
+
)
|
122 |
+
except Exception as e:
|
123 |
+
print(f"An error occurred during video generation: {e}")
|
124 |
+
import traceback
|
125 |
+
traceback.print_exc()
|
126 |
+
raise gr.Error(f"An unexpected error occurred: {str(e)}. Please check the console for details.")
|
127 |
+
|
128 |
+
end_time = time.time()
|
129 |
+
|
130 |
+
processing_time = end_time - start_time
|
131 |
+
|
132 |
+
result_video_path = Path(result_video_path)
|
133 |
+
final_path = result_video_path.with_name(f"final_{result_video_path.stem}{result_video_path.suffix}")
|
134 |
+
|
135 |
+
print(f"Video generated successfully at: {final_path}")
|
136 |
+
print(f"Processing time: {processing_time:.2f} seconds.")
|
137 |
+
|
138 |
+
return final_path
|
139 |
+
|
140 |
+
# --- Gradio UI Definition ---
|
141 |
+
with gr.Blocks(theme=gr.themes.Soft(), css=".gradio-container {max-width: 90% !important;}") as demo:
|
142 |
+
gr.HTML(
|
143 |
+
"""
|
144 |
+
<div align='center'>
|
145 |
+
<h1>MoDA: Multi-modal Diffusion Architecture for Talking Head Generation</h1>
|
146 |
+
<p style="display:flex">
|
147 |
+
<a href='https://lixinyyang.github.io/MoDA.github.io/'><img src='https://img.shields.io/badge/Project-Page-blue'></a>
|
148 |
+
<a href='https://arxiv.org/abs/2507.03256'><img src='https://img.shields.io/badge/Paper-Arxiv-red'></a>
|
149 |
+
<a href='https://github.com/lixinyyang/MoDA/'><img src='https://img.shields.io/badge/Code-Github-green'></a>
|
150 |
+
</p>
|
151 |
+
<p>
|
152 |
+
This demo allows you to generate a talking head video from a source image and a driving audio file.
|
153 |
+
</p>
|
154 |
+
</div>
|
155 |
+
"""
|
156 |
+
)
|
157 |
+
|
158 |
+
with gr.Row(variant="panel"):
|
159 |
+
with gr.Column(scale=1):
|
160 |
+
with gr.Row():
|
161 |
+
source_image = gr.Image(label="Source Image", type="filepath", value="src/examples/reference_images/6.jpg")
|
162 |
+
|
163 |
+
with gr.Row():
|
164 |
+
driving_audio = gr.Audio(label="Driving Audio", type="filepath", value="src/examples/driving_audios/5.wav")
|
165 |
+
|
166 |
+
with gr.Row():
|
167 |
+
emotion_dropdown = gr.Dropdown(
|
168 |
+
label="Emotion",
|
169 |
+
choices=list(emo_map.values()),
|
170 |
+
value="None"
|
171 |
+
)
|
172 |
+
|
173 |
+
with gr.Row():
|
174 |
+
cfg_slider = gr.Slider(
|
175 |
+
label="CFG Scale",
|
176 |
+
minimum=1.0,
|
177 |
+
maximum=3.0,
|
178 |
+
step=0.05,
|
179 |
+
value=1.2
|
180 |
+
)
|
181 |
+
|
182 |
+
submit_button = gr.Button("Generate Video", variant="primary")
|
183 |
+
|
184 |
+
with gr.Column(scale=1):
|
185 |
+
output_video = gr.Video(label="Generated Video")
|
186 |
+
|
187 |
+
gr.Markdown("## Examples")
|
188 |
+
gr.Examples(
|
189 |
+
examples=[
|
190 |
+
["src/examples/reference_images/monalisa.jpg", "src/examples/driving_audios/monalisa.wav", "None", 1.2],
|
191 |
+
["src/examples/reference_images/girl.png", "src/examples/driving_audios/girl.wav", "Happiness", 1.25],
|
192 |
+
["src/examples/reference_images/jobs.jpg", "src/examples/driving_audios/jobs.wav", "Neutral", 1.15],
|
193 |
+
],
|
194 |
+
inputs=[source_image, driving_audio, emotion_dropdown, cfg_slider],
|
195 |
+
outputs=output_video,
|
196 |
+
fn=generate_motion,
|
197 |
+
cache_examples=False,
|
198 |
+
)
|
199 |
+
|
200 |
+
gr.Markdown(
|
201 |
+
"""
|
202 |
+
---
|
203 |
+
### **Disclaimer**
|
204 |
+
This project is intended for academic research, and we explicitly disclaim any responsibility for user-generated content. Users are solely liable for their actions while using this generative model.
|
205 |
+
"""
|
206 |
+
)
|
207 |
+
|
208 |
+
submit_button.click(
|
209 |
+
fn=generate_motion,
|
210 |
+
inputs=[source_image, driving_audio, emotion_dropdown, cfg_slider],
|
211 |
+
outputs=output_video
|
212 |
+
)
|
213 |
+
|
214 |
+
if __name__ == "__main__":
|
215 |
+
demo.launch(share=True)
|