Restore YouTube URL input with yt-dlp

app.py

@@ -55,12 +55,32 @@ def to_h264(src: str, dst: str):
     )
 
 
+def download_youtube(url: str, out_path: str):
+    import yt_dlp
+    raw = out_path.replace(".mp4", "_raw.mp4")
+    ydl_opts = {
+        "outtmpl": raw,
+        "format": "best[height<=480]/best",
+        "http_headers": {
+            "User-Agent": (
+                "Mozilla/5.0 (Windows NT 10.0; Win64; x64) "
+                "AppleWebKit/537.36 Chrome/120.0.0.0 Safari/537.36"
+            )
+        },
+    }
+    with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+        ydl.download([url])
+    to_h264(raw, out_path)
+    if os.path.exists(raw):
+        os.remove(raw)
+
+
 # ── Core processing ──────────────────────────────────────────────────
-def process(face_image, video_file, trim_seconds, progress=gr.Progress(track_tqdm=True)):
+def process(face_image, video_file, youtube_url, trim_seconds, progress=gr.Progress(track_tqdm=True)):
     if face_image is None:
         return None, "Please upload a source face image."
-    if video_file is None:
-        return None, "Please upload a video file."
+    if video_file is None and not (youtube_url or "").strip():
+        return None, "Please upload a video file or paste a YouTube URL."
 
     try:
         progress(0.0, desc="Detecting source face...")
@@ -82,12 +102,16 @@ def process(face_image, video_file, trim_seconds, progress=gr.Progress(track_tqd
         source_face.embedding /= np.linalg.norm(source_face.embedding)
 
         # Prepare video
-        progress(0.05, desc="Preparing video...")
-        raw_video   = f"{WORK_DIR}/temp/input.mp4"
-        converted   = f"{WORK_DIR}/temp/input_h264.mp4"
+        raw_video = f"{WORK_DIR}/temp/input.mp4"
+        converted = f"{WORK_DIR}/temp/input_h264.mp4"
 
-        shutil.copy(video_file, raw_video)
-        to_h264(raw_video, converted)
+        if (youtube_url or "").strip():
+            progress(0.05, desc="Downloading YouTube video...")
+            download_youtube(youtube_url.strip(), converted)
+        else:
+            progress(0.05, desc="Preparing video...")
+            shutil.copy(video_file, raw_video)
+            to_h264(raw_video, converted)
 
         # Verify codec
         cap_check = cv2.VideoCapture(converted)
@@ -176,6 +200,11 @@ Swap any face into a video using **InsightFace + inswapper_128**.
                 type="filepath",
                 height=220,
             )
+            youtube_input = gr.Textbox(
+                label="YouTube URL (optional)",
+                placeholder="https://www.youtube.com/watch?v=...",
+            )
+            gr.Markdown("**— or —**")
             video_input = gr.Video(label="Upload Video File")
             trim_input = gr.Slider(
                 label="Trim to first N seconds (0 = full video)",
@@ -197,7 +226,7 @@ Swap any face into a video using **InsightFace + inswapper_128**.
 
     run_btn.click(
         fn=process,
-        inputs=[face_input, video_input, trim_input],
+        inputs=[face_input, video_input, youtube_input, trim_input],
         outputs=[video_out, status_box],
     )
 

requirements.txt

@@ -2,4 +2,5 @@ insightface==0.7.3
 onnxruntime
 opencv-python-headless
 huggingface_hub
+yt-dlp
 gradio