Applies to: Users creating multi-clip videos or using reference images (where supported).
Use these four techniques to keep the same character looking identical across multiple generated video clips.
Reuse identical descriptions verbatim. Copy the exact same character, wardrobe, and setting wording into every video prompt in the sequence.
Use reference images if the video model supports them. Multiple angles (front, three-quarter, profile, full body) of the same character work best.
Shorter clips drift less. If a character's face changes mid-clip, generate shorter video clips.
Avoid video prompts that imply transformation ("she becomes...", "turns into...") unless intended.
On Seedance 2.5, assign a reference image the explicit job of defining identity (@Image1 defines the subject) and keep clothing, position, and screen direction consistent across clips. See Seedance 2.5: advanced reference & editing controls.