fix: validate URLs upfront — reject invalid/non-YouTube URLs immediately

ZeroGPU blocks all external DNS except HuggingFace services, so only
YouTube URLs (via proxy) and uploaded files work. Previously:
- Invalid text ("hello") was treated as a file path → FileNotFoundError
- Non-YouTube URLs tried download.download_video → hung on DNS
- YouTube URLs without scheme ("youtube.com/...") → treated as file
- YouTube URLs with invalid video ID → confusing proxy error

Now validates source before the pipeline starts:
- Auto-adds https:// for youtube.com/youtu.be URLs missing scheme
- Rejects non-YouTube URLs with clear explanation
- Validates YouTube video ID format (11 chars from URL)
- Rejects garbage input with helpful guidance
- Simplifies Stage 1 download logic (only YouTube proxy or local file)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

pipeline_runner.py

@@ -322,23 +322,56 @@ def run_pipeline(
     """
     from pathlib import Path
     import soundfile as sf
-
-    # Resolve slug — YouTube DNS is blocked on ZeroGPU so skip yt-dlp for
-    # YouTube URLs entirely (avoids 30-60s of pointless DNS retries).
     import hashlib
-    if _is_youtube_url(source):
+
+    # ── Upfront source validation ──────────────────────────────────────
+    # ZeroGPU blocks all external DNS except HuggingFace services.
+    # Only YouTube URLs (via proxy) and uploaded files work.
+    source = source.strip()
+    is_local = os.path.exists(source)
+    is_url = download.is_url(source)
+    is_yt = _is_youtube_url(source)
+
+    if not is_local and not is_url:
+        # Might be a URL without scheme — try adding https://
+        if "youtube.com" in source or "youtu.be" in source:
+            source = "https://" + source
+            is_url = True
+            is_yt = True
+        else:
+            yield 1, (
+                f"[ERROR] Invalid source: '{source}'\n"
+                "Please provide a YouTube URL (e.g. https://youtube.com/watch?v=...) "
+                "or upload a video/audio file."
+            ), {"error": "invalid source"}
+            return
+
+    if is_url and not is_yt:
+        yield 1, (
+            f"[ERROR] Non-YouTube URLs are not supported on this Space.\n"
+            f"URL: {source}\n"
+            "ZeroGPU blocks external network access. Only YouTube URLs work "
+            "(downloaded via proxy). Please use a YouTube link or upload your file directly."
+        ), {"error": "unsupported URL"}
+        return
+
+    if is_yt:
         match = _YT_RE.search(source)
-        slug = match.group(1) if match else hashlib.md5(source.encode()).hexdigest()[:12]
-    elif download.is_url(source):
-        try:
-            slug = download.resolve_slug(source)[0]
-        except Exception:
-            slug = hashlib.md5(source.encode()).hexdigest()[:12]
-    else:
+        if not match:
+            yield 1, (
+                f"[ERROR] Could not find a valid YouTube video ID in: {source}\n"
+                "Expected format: https://youtube.com/watch?v=VIDEO_ID or https://youtu.be/VIDEO_ID"
+            ), {"error": "invalid YouTube URL"}
+            return
+        slug = match.group(1)
+    elif is_local:
         try:
             slug = download.slug_from_path(source)
         except Exception:
             slug = hashlib.md5(source.encode()).hexdigest()[:12]
+    else:
+        slug = hashlib.md5(source.encode()).hexdigest()[:12]
+
     proj = ProjectPaths(slug, base_dir=BASE_DIR, target_language=target_language)
     proj.ensure_dirs()
     tracker = LLMUsageTracker()
@@ -351,18 +384,17 @@ def run_pipeline(
     yield stage, f"[{STAGE_NAMES[stage - 1]}] Downloading: {source}", {}
 
     try:
-        if download.is_url(source):
-            if _is_youtube_url(source):
-                # YouTube: use proxy Space (ZeroGPU blocks youtube.com DNS)
-                yield stage, f"[{STAGE_NAMES[stage - 1]}] Downloading via YouTube proxy...", {}
-                _download_via_proxy(source, proj.video)
-            else:
-                download.download_video(source, proj.video)
+        if is_yt:
+            yield stage, f"[{STAGE_NAMES[stage - 1]}] Downloading via YouTube proxy...", {}
+            _download_via_proxy(source, proj.video)
             download.extract_audio(proj.video, proj.audio)
-        elif download.is_audio_file(source):
+        elif is_local and download.is_audio_file(source):
             download.ingest_local_audio(source, proj.audio)
-        else:
+        elif is_local:
             download.ingest_local_video(source, proj.video, proj.audio)
+        else:
+            yield stage, f"[{STAGE_NAMES[stage - 1]}] FAILED: unsupported source", {"error": "unsupported"}
+            return
     except Exception as exc:
         yield stage, f"[{STAGE_NAMES[stage - 1]}] FAILED: {exc}", {"error": str(exc)}
         return