Assign each reference
Choose the image for visual identity, the video for motion, and the audio for timing only when each input improves the shot.
ByteDance video model
A flexible video route for multi-input generation, longer clips, and projects that need reference control.
Where it fits
Text, image, video, or audio to video. Listed tier: 480p price tier. Compare duration, resolution, reference support, and audio before choosing a model.
Working method
Seedance 2.5 is the flexible route for jobs that begin with more than a text prompt. It can fit campaigns that use image, video, or audio references and need several related outputs built from the same creative source.
That flexibility is most useful when the references have clear jobs. Decide which asset controls the subject, which controls movement, and which controls timing. Test the 480p tier first, then keep the same reference set for the selected final direction.
Choose the image for visual identity, the video for motion, and the audio for timing only when each input improves the shot.
Run a short 480p draft to make sure the model follows the most important reference before expanding the batch.
Hold the reference set steady while changing the hook, crop, or ending for each channel.
Before generation
Common controls
Price basis checked November 5, 2026
LONG AI marketplace pricing →