Fix: restore transcode-on-upload for browser preview; mux already re-encodes so no conflict

app.py

@@ -156,6 +156,35 @@ 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 uploaded video to H.264/AAC MP4 so the browser preview widget can play it.
+
+    Returns the transcoded path on success, or the original path if transcoding fails.
+    Only called on upload — not during generation.
+    """
+    if video_path is None:
+        return video_path
+    try:
+        probe     = ffmpeg.probe(video_path)
+        has_audio = any(s["codec_type"] == "audio" for s in probe.get("streams", []))
+        out_path  = video_path.rsplit(".", 1)[0] + "_preview.mp4"
+        kwargs = dict(
+            vcodec="libx264", preset="fast", crf=18,
+            pix_fmt="yuv420p", movflags="+faststart",
+        )
+        if has_audio:
+            kwargs["acodec"]        = "aac"
+            kwargs["audio_bitrate"] = "128k"
+        else:
+            kwargs["an"] = None
+        ffmpeg.input(video_path).output(out_path, **kwargs).run(
+            overwrite_output=True, quiet=True
+        )
+        return out_path
+    except Exception as e:
+        print(f"[transcode_for_browser] failed, using original: {e}")
+        return video_path
+
 
 # ------------------------------------------------------------------ #
 # Temp directory registry — tracks dirs for cleanup on new generation #
@@ -2555,7 +2584,7 @@ with gr.Blocks(title="Generate Audio for Video", css=_SLOT_CSS, js=_GLOBAL_JS) a
         with gr.Tab("TARO"):
             with gr.Row():
                 with gr.Column(scale=1):
-                    taro_video   = gr.Video(label="Input Video", format="mp4")
+                    taro_video   = gr.Video(label="Input Video")
                     taro_seed    = gr.Number(label="Seed (-1 = random)", value=-1, precision=0, elem_id="taro_seed")
                     taro_cfg     = gr.Slider(label="CFG Scale", minimum=1, maximum=15, value=8.0, step=0.5, elem_id="taro_cfg")
                     taro_steps   = gr.Slider(label="Sampling Steps", minimum=10, maximum=50, value=25, step=1, elem_id="taro_steps")
@@ -2620,7 +2649,7 @@ with gr.Blocks(title="Generate Audio for Video", css=_SLOT_CSS, js=_GLOBAL_JS) a
         with gr.Tab("MMAudio"):
             with gr.Row():
                 with gr.Column(scale=1):
-                    mma_video    = gr.Video(label="Input Video", format="mp4")
+                    mma_video    = gr.Video(label="Input Video")
                     mma_prompt   = gr.Textbox(label="Prompt", placeholder="e.g. footsteps on gravel", elem_id="mma_prompt")
                     mma_neg      = gr.Textbox(label="Negative Prompt", value="music", placeholder="music, speech", elem_id="mma_neg")
                     mma_seed     = gr.Number(label="Seed (-1 = random)", value=-1, precision=0, elem_id="mma_seed")
@@ -2674,7 +2703,7 @@ with gr.Blocks(title="Generate Audio for Video", css=_SLOT_CSS, js=_GLOBAL_JS) a
         with gr.Tab("HunyuanFoley"):
             with gr.Row():
                 with gr.Column(scale=1):
-                    hf_video    = gr.Video(label="Input Video", format="mp4")
+                    hf_video    = gr.Video(label="Input Video")
                     hf_prompt   = gr.Textbox(label="Prompt", placeholder="e.g. rain hitting a metal roof", elem_id="hf_prompt")
                     hf_neg      = gr.Textbox(label="Negative Prompt", value="noisy, harsh", elem_id="hf_neg")
                     hf_seed     = gr.Number(label="Seed (-1 = random)", value=-1, precision=0, elem_id="hf_seed")
@@ -2723,6 +2752,15 @@ 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 ----
+    # Gradio serves the original uploaded file to the browser preview widget,
+    # so H.265 sources show as blank. We re-encode to H.264 on upload and feed
+    # the result back so the preview plays. mux_video_audio already re-encodes
+    # to H.264 during generation, so no double-conversion conflict.
+    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])