lucid.page Using reference clips and audio to direct AI video
Text size
Read time1 min

# Using reference clips and audio to direct AI video

Text prompts are a blunt instrument for describing movement. Saying “a graceful camera move” can mean a dozen different things. Reference material is more precise: a short clip that shows the camera motion you want, or an audio track that sets the rhythm.

Contents

Multimodal generators accept several kinds of reference at once, such as images, video and audio. The skill is deciding what each reference is for. Use an image for identity and look, a clip for motion, and audio for timing. Say so explicitly in the prompt so the model does not mix up their roles.

Keep references clean. A reference clip with a hard cut inside it may produce a hard cut in the result. An audio file with long silences may stretch the action.

Tools such as https://seedance2ai.im/seedance-2-fast accept this mixed input, which makes it easier to be specific about what you want. Start with one reference type, confirm it behaves as expected, and add the next only after that.

End