How are you handling audio on AI-generated clips right now?

Launching here tomorrow, and the thing I keep going back and forth on is how much the audio step actually costs people.

Most video models hand back a silent MP4. The workflow I see most often is: render, then go find music, then cut sound effects, then if there is dialogue, dub it and hope the lips line up. Three tools and a timeline for something that started as one prompt.

MiniMax H3 generates the audio in the same forward pass as the picture - 32 kHz stereo, dialogue stable in eleven languages - so there is no silent version to score afterwards, and no mute toggle either, because the sound was never a separate layer. I built a browser front-end for it at https://minimax-h3ai.video (independent third party, not affiliated with MiniMax).

But I am genuinely unsure that matches how people actually work, so:

1. Do you want the audio baked in, or do you specifically want a silent render because you keep control in the edit?

2. If you do add sound afterwards, which step eats the most time - finding music, syncing effects, or lip-sync?

3. Has anyone hit a case where model-generated dialogue was good enough to ship without a human pass?

If the honest answer is "I always want it silent", that changes what I work on next.