Short answer: Write a Seedance 2.5 storyboard prompt as a small shot plan: define the subject and setting once, assign each shot a time range, describe one visible action per shot, and state what must remain consistent. This is easier to review than asking for an entire sequence as one paragraph.
What a storyboard prompt should decide
A storyboard is not a list of camera adjectives. It answers five questions before generation:
- What is the subject, and which details must remain stable?
- Where does the scene begin and end?
- Which visible action belongs to each time range?
- How does the camera move or change between shots?
- Which transition is intentional, and which changes should not happen?
The current creator exposes 4–30 second duration presets, six aspect ratios, 480p, 720p, and 1080p output, and a prompt field of up to 2,500 characters. That makes a bounded shot plan more useful than trying to describe every frame.
A practical three-shot structure
For a short product or social clip, start with three shots:
| Time | Shot job | What to specify |
|---|---|---|
| 0–3s | Establish | subject, location, starting composition, light |
| 3–7s | Demonstrate | one action, camera direction, key visual cue |
| 7–10s | Resolve | final position, hold, and ending composition |
Keep the subject, location, and visual style outside the time-coded list. Put only changes inside each shot. This reduces contradictions such as “locked camera” followed by a push-in or “one continuous take” followed by several cuts.
Prompt template
Use this as a starting structure, then replace the bracketed details:
Create a [duration]-second video in [aspect ratio]. The main subject is [subject]
in [setting]. Keep [identity, product shape, clothing, color, or other details]
consistent from start to finish. Use [lighting and visual style].
Shot 1, [time range]: [starting composition]. [Small opening action].
Shot 2, [time range]: [one visible action]. The camera [specific movement or
framing change] while [continuity condition].
Shot 3, [time range]: [final action]. End with [final composition] and hold for
[short hold].
Use [reference label] only for [its defined role]. No extra subjects, products,
scene changes, unrequested text, or additional cuts.
This template is illustrative, not a tested result. Attach the same reference assets again for each new generation, and replace labels such as @Image1 with the asset labels shown in the interface.
Worked example: a product reveal
The following is an illustrative prompt for a travel mug. It is not presented as a measured product result:
Create a 10-second 16:9 product video of the travel mug in @Image1 on a light
wooden table beside a window. Keep the mug's shape, lid, handle, color, and
visible markings consistent with @Image1. Use soft morning light and a clean,
quiet commercial style.
Shot 1, 0–3s: Begin with a stable medium shot; the mug is already centered and
still. A narrow patch of sunlight moves across the table.
Shot 2, 3–7s: Make one slow left-to-right camera move while the mug remains
still and fully visible. Do not introduce hands or another product.
Shot 3, 7–10s: Ease to a stop, keep the mug centered, and hold the final frame
for one second. Preserve any markings already visible in @Image1, but add no new
text or logos. No zoom, orbit, camera cut, or scene change.
The example deliberately uses one action per shot. If the goal is to show the lid opening, replace the camera move with that action instead of adding both to the same time range.
References: assign one job to each asset
The creator supports multimodal references, but more references do not automatically make a prompt clearer. Give each asset one role:
- an image for product or character appearance;
- a video for motion or camera inspiration;
- an audio file for pacing or atmosphere;
- a second image for a separate location or design constraint.
Name the role in the prompt. Do not ask one reference to control appearance, camera movement, lighting, and editing at the same time. If a detail is essential, describe it in text as well as attaching the reference.
Review the output shot by shot
Before rewriting the whole prompt, check the result against the plan:
- Does each shot begin and end in the intended time range?
- Is the subject still identifiable after the camera changes?
- Did the model add a cut, person, prop, logo, or text that was not requested?
- Does the final frame hold long enough to use in an edit?
- Is the chosen aspect ratio suitable for the destination, such as 16:9 for a landscape player or 9:16 for a vertical feed?
If only one shot fails, change that shot's action or timing first. If the subject changes across shots, simplify the action and restate the continuity condition. If two separately generated clips need to meet, use the Seedance 2.5 transitions guide for join planning rather than adding more instructions here.
When to split a storyboard into separate generations
Use one generation when the shots share a subject, setting, and visual direction and the sequence fits the selected duration. Split the plan when the location changes, the subject needs a new reference, the action becomes too dense, or a specific shot must be regenerated without changing the others. In that case, keep the last frame of one clip and the first frame of the next visually compatible, then align the clips in an external editor.
For the current generation workflow, see Text to Video, and review the Seedance 2.5 model overview before choosing settings and credits.
FAQ
Is a storyboard the same as a prompt?
No. A storyboard is the plan of shots and timing; a prompt is the instruction sent to the generator. Combining the two gives the model both sequence structure and visual detail.
How many shots should I put in one short video?
Start with two or three bounded shots. Add another only when it has a distinct visual job. Too many tiny beats make it harder to tell which action should happen first.
Should every shot use a different reference image?
No. Reuse the same reference when the subject should stay consistent. Add a new asset only when it has a clear role, such as a new location, object, or motion reference.
Can a storyboard guarantee exact timing?
No. Time ranges make the request easier to interpret and review, but generated motion can still drift. For exact alignment, trim or reposition the final clips in an external editor.
