Update app.py

app.py

@@ -1,11 +1,8 @@
 import os
 os.environ["OMP_NUM_THREADS"] = "1"
-
-import os
 import subprocess
 from pathlib import Path
 from datetime import datetime
-
 import gradio as gr
 from huggingface_hub import snapshot_download
 
@@ -25,10 +22,10 @@ def setup():
     # Clone LatentSync repo at runtime (won't appear in HF Files tab)
     if not REPO_DIR.exists():
         run(["git", "clone", "--depth", "1", "https://github.com/bytedance/LatentSync.git", str(REPO_DIR)])
-
+    
     CKPT_DIR.mkdir(parents=True, exist_ok=True)
     TEMP_DIR.mkdir(parents=True, exist_ok=True)
-
+    
     # Download all checkpoint files (includes latentsync_unet + whisper tiny/small etc)
     snapshot_download(
         repo_id=HF_CKPT_REPO,
@@ -59,18 +56,19 @@ def make_still_video(image_path: str, audio_path: str, fps: int = 25) -> str:
 
 def generate(avatar_img, audio_wav, steps, guidance, seed, use_deepcache):
     setup()
-
+    
     img_path = str(Path(avatar_img).resolve())
     wav_path = str(Path(audio_wav).resolve())
-
+    
     # Make a temp mp4 from the single image + audio
     video_path = make_still_video(img_path, wav_path, fps=25)
-
+    
     out_path = TEMP_DIR / f"result_{datetime.now().strftime('%Y%m%d_%H%M%S')}.mp4"
-
+    
+    # FIXED: Use correct config path - configs/unet/stage2.yaml instead of configs/unet.yaml
     cmd = [
         "python", "-m", "scripts.inference",
-        "--unet_config_path", "configs/unet.yaml",
+        "--unet_config_path", "configs/unet/stage2.yaml",  # ← FIXED PATH
         "--inference_ckpt_path", "checkpoints/latentsync_unet.pt",
         "--video_path", video_path,
         "--audio_path", wav_path,
@@ -80,28 +78,30 @@ def generate(avatar_img, audio_wav, steps, guidance, seed, use_deepcache):
         "--seed", str(int(seed)),
         "--temp_dir", "temp",
     ]
+    
     if use_deepcache:
         cmd.append("--enable_deepcache")
-
+    
     run(cmd, cwd=str(REPO_DIR))
+    
     return str(out_path)
 
 with gr.Blocks(title="LatentSync (avatar.jpg + audio.wav → lip-sync mp4)") as demo:
-    gr.Markdown("## LatentSync on Hugging Face (T4) — Upload avatar + audio → mp4")
-
+    gr.Markdown("## LatentSync 1.5 on Hugging Face (T4) — Upload avatar + audio → mp4")
+    
     with gr.Row():
         avatar = gr.Image(type="filepath", label="Avatar image (jpg/png)")
         audio = gr.Audio(type="filepath", label="Audio (wav)", format="wav")
-
+    
     with gr.Row():
         steps = gr.Slider(10, 40, value=20, step=1, label="Inference Steps")
         guidance = gr.Slider(0.8, 2.0, value=1.0, step=0.1, label="Guidance Scale")
         seed = gr.Number(value=1247, precision=0, label="Seed")
         deepcache = gr.Checkbox(value=True, label="Enable DeepCache (faster)")
-
+    
     btn = gr.Button("Generate")
     out = gr.Video(label="Output video")
-
+    
     btn.click(generate, inputs=[avatar, audio, steps, guidance, seed, deepcache], outputs=out)
 
-demo.launch()
+demo.launch()