Seedance 2.5 Flash character consistency: keeping one face across a series

Quick answer: Seedance 2.5 Flash keeps a character consistent the way the flagship does: from reference images, not from prompt wording. In SceneMixer every character is drawn once as a reference sheet when the script is read, and each Flash segment is sent the sheets of the people in it — so the lead has the same face in episode 1 and episode 40, whether the segment is a Flash draft or a Pro final.
Still from “Rust City Knockout”, an AI short drama generated end-to-end by SceneMixer (6 characters, 1 episode)
SampleRust City KnockoutPainterly Anime1 episodeProduced end-to-end by SceneMixer from the scriptWatch the episodes

Why wording is not enough

A description like “a woman in her thirties with dark hair” fits a thousand faces, and a video model picks a new one each time. Faces drift for three reasons: text gets diluted, a reference carries lighting baked into it, and shots have no anchor between them. Each has its own fix, and none of them is a better adjective.

Layer one: the sheet

Each character gets one image on a plain background with neutral, directional light, identity features first. That image goes to the video model with every segment the character appears in. Flash accepts up to ten images a request, which covers the people in a scene, the room and the props. Character consistency in three layers

Layer two: places and props

A location is generated once as an empty set and reused for every scene there; a prop that matters is an entry with its own image. The kitchen in episode 3 is the kitchen from episode 1, and the letter is the same letter.

Layer three: continuity between shots

Each segment opens where the last one ended, and the storyboard names only the people who are in the frame. Naming someone off-screen is the commonest cause of a face swap, because the model tries to put them in the picture.

Outfits and states

A character wears what their entry says. When the story needs a change — a wedding dress, a bandaged arm — you make a state for it and choose the segments that use it, so the change is deliberate rather than drift. Costume changes and scene states

Drafting on Flash does not change the cast

The tier is a setting for rendering, not for casting. A segment drafted on Seedance 2.5 Flash · 480P and finished on Seedance 2.5 Pro is sent the same sheets both times, so the face is the same; what changes is resolution and the take itself.

A quick check

  1. Render five segments with the same character: a close-up, a wide, a profile, a walking shot, a two-shot.
  2. Put a still from each side by side.
  3. If one is off, fix the sheet or the shot line, not the tier.

FAQ

Can Seedance 2.5 Flash keep a character consistent?

Yes, from reference images. SceneMixer sends each character's sheet with every segment they appear in.

Do I need to describe the face in every prompt?

No. The face comes from the sheet; the shot line says what the character does.

Does switching from Flash to Pro change the face?

No. Both tiers get the same sheets; resolution and the take change.

Why did a face change mid-scene?

Usually someone off-screen was named in the shot, or the sheet carries lighting or a background.

How many characters can one Flash segment hold?

Up to ten images go with a segment, shared between people, the location and props.

Can a character change clothes?

Yes, as a state you assign to the segments that need it.

Approve the cast once

Paste a script, check the sheets, and render every episode from the same faces.

Start a project

Updated November 16, 2026