fix: remove GFPGAN download, skip enhancer (not needed)

app.py

@@ -1,35 +1,49 @@
 """
 AIBRUH/ditto — Amanda's talking head engine.
-SadTalker-backed portrait animation: image + audio → MP4.
-ZeroGPU A100 Space.
+SadTalker ZeroGPU A100: portrait image + audio → MP4 talking head video.
 """
 import gradio as gr
 import spaces
-import os
-import sys
-import subprocess
-import tempfile
+import os, sys, subprocess, tempfile
 
-# ── One-time setup ──────────────────────────────────────────────────────────
 SADTALKER = "/tmp/SadTalker"
 CKPT = f"{SADTALKER}/checkpoints"
-GFPGAN = f"{SADTALKER}/gfpgan/weights"
 
 def _setup():
-    if not os.path.exists(f"{CKPT}/SadTalker_V0.0.2_256.safetensors"):
-        subprocess.run(["git", "clone", "--depth=1",
-                        "https://github.com/OpenTalker/SadTalker.git", SADTALKER], check=True)
-        os.makedirs(CKPT, exist_ok=True)
-        os.makedirs(GFPGAN, exist_ok=True)
-        from huggingface_hub import hf_hub_download, snapshot_download
-        snapshot_download("vinthony/SadTalker-V002rc", local_dir=CKPT, ignore_patterns=["*.md"])
-        hf_hub_download("tencent/GFPGANv1.4", filename="GFPGANv1.4.pth",
-                        local_dir=GFPGAN)
-    if SADTALKER not in sys.path:
-        sys.path.insert(0, SADTALKER)
+    if os.path.exists(f"{CKPT}/SadTalker_V0.0.2_256.safetensors"):
+        return  # already done
+
+    # Clone SadTalker source
+    if not os.path.exists(SADTALKER):
+        subprocess.run(
+            ["git", "clone", "--depth=1",
+             "https://github.com/OpenTalker/SadTalker.git", SADTALKER],
+            check=True,
+        )
+
+    os.makedirs(CKPT, exist_ok=True)
+
+    # Download SadTalker checkpoints (no GFPGAN — we skip the enhancer)
+    from huggingface_hub import snapshot_download
+    snapshot_download(
+        repo_id="vinthony/SadTalker-V002rc",
+        local_dir=CKPT,
+        ignore_patterns=["*.md", "*.txt"],
+    )
+
+    # Install SadTalker deps that aren't in requirements.txt
+    subprocess.run(
+        [sys.executable, "-m", "pip", "install", "-q",
+         "face-alignment", "imageio", "imageio-ffmpeg",
+         "pydub", "scipy", "kornia==0.6.8", "basicsr"],
+        check=False,
+    )
 
 _setup()
 
+if SADTALKER not in sys.path:
+    sys.path.insert(0, SADTALKER)
+
 from src.gradio_demo import SadTalker as _ST  # noqa: E402
 
 _sad: _ST | None = None
@@ -43,14 +57,14 @@ def _load():
 
 @spaces.GPU
 def infer(source_image: str, driven_audio: str):
-    """image filepath + audio filepath → video filepath"""
+    """Portrait image + WAV audio → talking head MP4."""
     st = _load()
     result = st.test(
         source_image=source_image,
         driven_audio=driven_audio,
         preprocess="crop",
-        still_mode=True,
-        use_enhancer=False,
+        still_mode=True,       # minimal head sway — portrait mode
+        use_enhancer=False,    # no GFPGAN needed
         batch_size=1,
         size=256,
         pose_style=0,
@@ -65,13 +79,17 @@ def infer(source_image: str, driven_audio: str):
     return result
 
 
-with gr.Blocks(title="AIBRUH/ditto — Amanda Engine") as demo:
-    gr.Markdown("## AIBRUH/ditto · Amanda Talking Head Engine")
+with gr.Blocks(title="AIBRUH/ditto · Amanda Engine") as demo:
+    gr.Markdown("## AIBRUH/ditto · Amanda Talking Head Engine\n*SadTalker ZeroGPU — portrait image + audio → talking head video*")
     with gr.Row():
-        img_in = gr.Image(type="filepath", label="Source Portrait")
+        img_in = gr.Image(type="filepath", label="Source Portrait (JPG/PNG)")
         aud_in = gr.Audio(type="filepath", label="Driving Audio (WAV)")
-    btn = gr.Button("Generate", variant="primary")
+    btn = gr.Button("Generate Talking Head", variant="primary")
     vid_out = gr.Video(label="Amanda Speaking")
     btn.click(fn=infer, inputs=[img_in, aud_in], outputs=vid_out, api_name="infer")
+    gr.Examples(
+        examples=[],
+        inputs=[img_in, aud_in],
+    )
 
 demo.launch()