Auto-extract transcribes first when no transcript exists

Narration-gating now works regardless of button order: Auto-extract runs Whisper
first if needed, surfaces the transcript in the box + state (reused, not recomputed),
and never clobbers a transcript the user may have edited.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

app.py

@@ -172,22 +172,38 @@ def on_capture(session: str, frames: list[dict], data_url: str, current_time: fl
 
 
 def on_auto(session: str, frames: list[dict], video_path: str, transcript_obj, progress=gr.Progress()):
+    # Outputs: gallery, frames_state, transcript_box, transcript_state, status
     if not video_path:
-        return _gallery_value(frames), frames, "Upload a video first."
-    progress(0.1, "Detecting scenes…")
+        return _gallery_value(frames), frames, gr.update(), gr.update(), "Upload a video first."
+
+    # Transcribe first if we don't have a transcript yet, so frames can be gated
+    # to the narration (otherwise recorder intro/idle screens get captured).
+    auto_transcribed = False
+    if not (transcript_obj and getattr(transcript_obj, "segments", None)):
+        progress(0.0, "No transcript yet — transcribing first…")
+        transcript_obj = _run_transcription(session, video_path, progress)
+        auto_transcribed = True
+
     spoken = (
         [(s.start, s.end) for s in transcript_obj.segments]
         if transcript_obj and transcript_obj.segments
         else None
     )
+    progress(0.85, "Detecting scenes…")
     recs = extract_auto_frames(video_path, config.session_dir(session), spoken_intervals=spoken)
     merged = frames + [asdict(r) for r in recs]
     progress(1.0, "Done.")
-    tip = "" if spoken else "  ·  tip: transcribe first for narration-aligned frames"
+
+    # If we just transcribed, surface it in the box + state so it's reused (and
+    # don't clobber a transcript the user may have already edited).
+    box_out = transcript_obj.to_timestamped_text() if auto_transcribed else gr.update()
+    note = " (auto-transcribed first)" if auto_transcribed else ""
     return (
         _gallery_value(merged),
         merged,
-        f"Auto-extracted {len(recs)} frames ({len(merged)} total).{tip}",
+        box_out,
+        transcript_obj,
+        f"Auto-extracted {len(recs)} frames{note} ({len(merged)} total).",
     )
 
 
@@ -213,14 +229,19 @@ def on_clear():
     return [], [], None, "Cleared all frames."
 
 
-def on_transcribe(session: str, video_path: str, progress=gr.Progress()):
-    if not video_path:
-        return "", None, "Upload a video first."
+def _run_transcription(session: str, video_path: str, progress):
+    """Extract audio and transcribe — shared by Transcribe and Auto-extract."""
     sdir = config.session_dir(session)
     progress(0.05, "Extracting audio…")
     wav = video.extract_audio(video_path, sdir / "audio.wav")
     progress(0.1, "Loading Whisper…")
-    tr = transcribe_lib.transcribe(wav, progress=progress)
+    return transcribe_lib.transcribe(wav, progress=progress)
+
+
+def on_transcribe(session: str, video_path: str, progress=gr.Progress()):
+    if not video_path:
+        return "", None, "Upload a video first."
+    tr = _run_transcription(session, video_path, progress)
     return (
         tr.to_timestamped_text(),
         tr,
@@ -342,8 +363,8 @@ def build_ui() -> gr.Blocks:
             with gr.Column(scale=2):
                 gr.Markdown(
                     "### Captured frames\n"
-                    "_Tip: **Transcribe** (step 2) before **Auto-extract** — frames then "
-                    "snap to the narration and skip recorder intro/idle screens._"
+                    "_**Auto-extract** transcribes first automatically, so frames snap to "
+                    "the narration and skip recorder intro/idle screens._"
                 )
                 gallery = gr.Gallery(
                     label="Frames pool — click an image to enlarge / select it",
@@ -394,7 +415,7 @@ def build_ui() -> gr.Blocks:
         auto_btn.click(
             on_auto,
             [session_state, frames_state, video_state, transcript_state],
-            [gallery, frames_state, status],
+            [gallery, frames_state, transcript_box, transcript_state, status],
         )
         gallery.select(on_select_frame, None, [selected_state, status])
         delete_btn.click(