Seedance has two off-target defaults for UGC: it layers a music track under the voiceover, and it burns hallucinated subtitles into the pixels. Both kill authentic creator vibe. Both go away with two short prompt lines.
The two lines
Add these to every R2V prompt with quoted speech (audio scope + visual scope, separate sentences):
No music score. No subtitles or on-screen text.That's it. Don't expand. Don't list ambient sounds you want to keep ("street noise, café chatter, dish clatter, espresso machine, footsteps") — Seedance produces natural ambient on its own; enumerating it makes the model treat your list as a script and the audio comes out feeling staged. Trust the model.
Skip these lines only when
- User explicitly asks for music ("with a beat", "cinematic score", "müzikli")
- User explicitly asks for burned captions ("with subtitles", "altyazılı", "TikTok-style text overlay")
Subtitle pin for text-bearing reference images
If the reference image contains text (signage, product packaging, mural with words), Seedance sometimes redraws extracted text as a caption-style overlay. Add one explicit pin in addition to the standard line:
Only the text that exists in @Image1 itself — no additional text overlays.How to apply
- Default: every R2V prompt with
--generate-audioAND quoted speech gets the two short lines. Not the long version, not the ambient enumeration. - Multi-segment: include both lines in EVERY segment's prompt — voice continuity (
--reference-audio) doesn't carry the music/subtitle suppression. - Plan summary: surface "music: off" and "subtitles: off" as flags the user can flip.