Seedance 2.5 Flash prompt guide: writing shots that render well on a fast model

Quick answer: Seedance 2.5 Flash follows the same shot language as the rest of the Seedance 2.5 family, so write it the same way: one action and one camera move per shot, lines paced at about three English words a second, a sound cue for the room, and on-screen text only where a shot really needs it. In SceneMixer the storyboard is written for you from your script; prompting is editing those shot lines before you render, which costs nothing.
Still from “The Veil at Dawn”, an AI short drama generated end-to-end by SceneMixer (6 characters, 1 episode)
SampleThe Veil at DawnDigital HD1 episodeProduced end-to-end by SceneMixer from the scriptWatch the episodes

Where the prompt lives

You do not type one long prompt. SceneMixer turns your script into a storyboard: each segment is a run of shots, and each shot is a line like this one.

[Shot 1] 6s: Medium close-up, slow push-in. @Mara reads the letter at the window, then folds it. @Mara (quiet): He knew. (voice: 🔊Mara) Ambient: rain on the glass.

The @ names point at the character sheets, so the model is given the faces as images rather than as words. Editing a shot line is free; rendering it is what costs. Edit the storyboard with AI

One action, one move per shot

A fast model rewards plain shots. Give each shot a framing, one camera move and one thing that happens. “Tracking shot alongside, she crosses to the door” renders; “she crosses, turns, the camera circles, we see the street” asks for four things in one breath. When the beat changes, start a new shot — the seconds of the shots add up to the segment.

ByteDance lists instruction following and physical motion among Flash’s strengths, and an early test found quick cuts between wide and close framings holding together. Use that: a short chain of clear shots is a good fit for a 30-second take.

Pace the lines to the seconds

About three English words a second is a natural pace, so a ten-word line needs three or four seconds before anyone moves. Fill a take with dialogue and nobody has time to act. Two people trading a line each and then a beat of silence reads as a scene.

Sound cues

Seedance 2.5 Flash makes the sound with the picture. A short Ambient: cue — rain, a crowded corridor, a kettle — gives the room a sound of its own. Delivery notes in brackets set how a line is said: (low, not turning). Lines, voice-over and captions

Keep lettering off the frame

Small text is where fast models slip; an early Flash test saw signs garble in quick shots. Only write on-screen text where the story needs it, as a caption line, and let subtitles come from the editor rather than from the picture. Seedance 2.5 Flash and on-screen text

End where the next take can start

Each segment begins where the last one ended. Land the final shot on a settled pose and a clear framing, not mid-gesture, and the next take has nothing to repair. The Seedance 2.5 prompt guide covers long takes in more depth.

FAQ

Do I write prompts myself for Seedance 2.5 Flash?

In SceneMixer the storyboard is written for you from your script; you edit the shot lines you want to change before rendering.

Is prompting different from the full Seedance 2.5?

No. Same shot language; plain shots with one move each suit the fast model best.

How many lines fit in a 30-second take?

Pace at about three English words a second and leave time for action; a few short exchanges, not a monologue.

Can Flash render text in the picture?

It can, but small lettering is where fast models slip. Use a caption only where the story needs it; subtitles come from the editor.

How do I keep a character's face the same?

Name them with @ so the shot uses their reference sheet; the face comes from the image, not the words.

Does editing the storyboard cost credits?

No. Editing text is free; rendering a segment is charged per second.

Write one scene, render it on Flash

Paste a script and edit the shot lines the storyboard gives you.

Start a project

Updated November 16, 2026