Fix: transcode uploaded video to H.264/AAC on upload for browser preview

app.py

@@ -156,6 +156,31 @@ def strip_audio_from_video(video_path: str, output_path: str) -> None:
         overwrite_output=True, quiet=True
     )
 
+def _transcode_for_browser(video_path: str) -> str:
+    """Re-encode *video_path* to H.264 baseline + AAC so all browsers can preview it.
+
+    Returns the path to the transcoded file (written alongside the original).
+    If the input is already browser-safe the re-encode is still cheap (~1s for
+    short clips) and ensures consistent behaviour across all upload sources.
+    """
+    if video_path is None:
+        return video_path
+    out_path = video_path.rsplit(".", 1)[0] + "_browser.mp4"
+    (
+        ffmpeg
+        .input(video_path)
+        .output(
+            out_path,
+            vcodec="libx264", preset="fast", crf=18,
+            profile="baseline", level="3.0",
+            pix_fmt="yuv420p",
+            acodec="aac", audio_bitrate="128k",
+            movflags="+faststart",
+        )
+        .run(overwrite_output=True, quiet=True)
+    )
+    return out_path
+
 # ------------------------------------------------------------------ #
 # Temp directory registry — tracks dirs for cleanup on new generation #
 # ------------------------------------------------------------------ #
@@ -2719,6 +2744,13 @@ with gr.Blocks(title="Generate Audio for Video", css=_SLOT_CSS, js=_GLOBAL_JS) a
                 hf_slot_vids, hf_slot_waves,
             )
 
+    # ---- Browser-safe transcode on upload ----
+    # Re-encodes uploaded video to H.264 baseline so all browsers can preview it.
+    # Each tab transcodes independently; cross-tab sync propagates the result.
+    taro_video.upload(fn=_transcode_for_browser, inputs=[taro_video], outputs=[taro_video])
+    mma_video.upload(fn=_transcode_for_browser,  inputs=[mma_video],  outputs=[mma_video])
+    hf_video.upload(fn=_transcode_for_browser,   inputs=[hf_video],   outputs=[hf_video])
+
     # ---- Cross-tab video sync ----
     _sync = lambda v: (gr.update(value=v), gr.update(value=v))
     taro_video.change(fn=_sync, inputs=[taro_video], outputs=[mma_video, hf_video])