A multi-character Seedance storyboard has two jobs: establish who each person is and explain what happens from shot to shot. Separate character sheets can describe faces, clothes, and proportions. A storyboard can describe framing, positions, and sequence. Keeping those responsibilities explicit gives you a clearer brief and a more useful way to diagnose a result when characters switch places or an important reaction disappears.
This article explains a creator-reported input arrangement and proposes an original planning example. The example is an editorial experiment, not a workflow tested by Seeora, and six panels are not an official requirement.
AI-generated concept illustration made with Codex imagegen. It is not a Seedance output, interface screenshot, or validated reference pack.
What the public example actually says
In an October 11, 2026 workflow reply, creator AI_swwww describes making three character sheets and one six-panel storyboard. Each panel represents a shot, with no text beyond panel numbers. The creator then uses the three sheets and board as reference images. Their example concerns a short confession scene with three people occupying different story roles.
The post also describes a 480p trial with Seedance 2.0 Fast and a 720p final with Seedance 2.5. That is a cross-model workflow reported by one creator, not the same thing as an official Seedance 2.5 Draft Mode process. Their credit figures belong to their service and settings; they are not a universal price.
A separate October 12 example from ashiwata100 assigns the first two images to different characters and the third to a venue. It demonstrates how explicit reference roles can organize a complex action brief. The creator says they tried Seedance 2.5 and warns that results can break. Neither account establishes a measured success rate.
Prepare identity separately from staging
For an initial experiment, use fictional adult characters you own or have permission to depict. Give each a recognizable silhouette and a small set of stable features. A mustard jacket, teal apron, and plum sweater are easier to distinguish in a busy scene than three similar gray outfits. Keep a person's face, hair, and clothing internally consistent across their sheet.
Do not make the identity sheet responsible for every event. If one sheet shows a character seated while the storyboard shows them standing, explain that the sheet defines appearance and the panel defines the current pose. Otherwise, your brief contains competing instructions before generation begins.
Check your chosen service's actual upload controls. Reference labels and accepted image counts vary by interface. The creator's four-image arrangement does not prove that every Seedance frontend accepts four references or reads a composite board the same way.
Build a board around one readable event
Here is an original six-shot plan: an adult customer returns a dropped bouquet to a florist while a friend realizes what happened. Keep the doorway, counter, and character positions readable. The goal is a small interaction with a visible beginning and consequence, rather than six unrelated attractive images.
| Panel | Main action | Continuity to preserve |
|---|---|---|
| 1 | Customer enters holding the bouquet | Customer in mustard jacket; doorway behind them |
| 2 | Florist notices the flowers | Florist in teal apron behind the counter |
| 3 | Customer extends the bouquet | Bouquet stays in the customer's hand until the transfer |
| 4 | Florist takes the bouquet | One bouquet; clear handoff between the same people |
| 5 | Friend reacts from beside the doorway | Friend in plum sweater; no identity swap |
| 6 | All three share a brief smile | Florist still holds the bouquet; same shop layout |
Start with panel numbers and simple compositions. Avoid dialogue balloons if you do not want those words rendered into the video. A close-up can hide a handoff, so give that action enough visual space. If a panel contains multiple simultaneous events, simplify it before adding more instructions.
State what each reference controls
Use your interface's real reference syntax. The following is an original brief to adapt, not a claim that these labels activate special API features:
Reference A defines the adult customer's face, hair, mustard jacket, and proportions. Reference B defines the adult florist's face, hair, and teal apron. Reference C defines the adult friend's appearance and plum sweater. Reference D defines the six-shot flower-shop sequence, framing, and relative positions. Character-sheet poses do not override the storyboard actions.
Follow panels 1 through 6 as a short, readable interaction. The customer returns one bouquet to the florist. The friend watches from beside the doorway. Keep identities distinct and the counter and doorway in consistent positions. Show the bouquet changing hands once, then remaining with the florist. Use simple camera movement, natural reactions, and no visible panel borders, numbers, captions, or dialogue balloons.Set duration and aspect ratio through the service's supported controls. Panel numbering communicates sequence; it does not guarantee exact cut timing. If six shots feel rushed within your available duration, reduce the board to fewer shots or generate separate clips for an edit.
Review causes, not just attractive frames
Watch the complete result once, then inspect the handoff and each cut. An individual frame can look convincing while the sequence changes who holds the flowers or silently replaces a character.
Record failures against their input responsibility. A changed face points to identity continuity; a skipped reaction points to shot coverage; a reversed doorway position points to staging. First simplify the responsible input. Repeating every instruction more forcefully can obscure the actual conflict.
When a composite board is not being interpreted usefully, an alternative experiment is to generate individual shots with consistent character references and assemble them afterward. That changes the workflow and requires editing, but lets you assess one event at a time. It is a proposed fallback, not a tested guarantee.
Keep previews informative
A preview is valuable only if it tests the behavior you care about. If you change both model and resolution between preview and final, the final may interpret the scene differently. Keep the accepted references and brief, but review identity, shot order, and prop state again after the switch.
For scenes where moving camera geometry is the difficult part, read the 3D previs reference explanation. For preview and final-resolution terminology, see the Seedance 2.5 Draft Mode article. A static board specifies selected views; moving previs adds information between those views.
Your next step is to prepare one small board, assign each reference a single responsibility, and write an acceptance checklist before generating. Keep the first result and the revised input together so you can tell which correction helped. This article does not announce a new storyboard upload feature on Seeora.