Short answer: Put the subject and non-negotiable continuity rules first, then describe only the actions that change over time. The current creator exposes a prompt field of up to 2,500 characters, so a short structured brief is easier to review than a long list of cinematic adjectives.
What belongs in the prompt
Use four blocks:
- Subject and setting: who or what is visible, where it starts, and the visual style.
- Continuity rules: details that must remain stable, such as a product shape, clothing color, or character identity.
- Timed action: one visible action per time range, including the camera change if it matters.
- Constraints: only the exclusions that prevent a likely failure, such as no extra people, cuts, or new text.
The goal is not to use every available character. The goal is to leave the model with an unambiguous order of operations.
A compact prompt formula
Create a [duration]-second [aspect ratio] video of [subject] in [setting].
Keep [three important identity or continuity details] consistent. Use [lighting
and style].
[Time range 1]: [opening composition and action].
[Time range 2]: [one visible action and camera movement].
[Time range 3]: [ending action and final hold].
Use [reference label] only for [defined role]. No [two or three likely unwanted
additions].
This is an illustrative template, not a tested result. Reattach the same references for each new generation and replace labels such as @Image1 with the labels shown in the interface.
What to cut first when the prompt is too long
Remove repeated adjectives before removing useful instructions. For example, “cinematic, premium, polished, beautiful, visually stunning” can usually become “clean commercial lighting.” Keep the subject, action, timing, and ending.
Next, combine related constraints:
- Instead of “the mug stays the same shape, same color, same lid, same handle,” write “keep the mug’s shape, color, lid, and handle consistent.”
- Instead of listing every unwanted object, exclude the two most likely failures.
- Replace a paragraph of camera mood with one movement and one stopping point.
Do not shorten the prompt by deleting the ending. The final frame is often the part that makes a clip usable in an edit.
Example: a focused product brief
The following example is illustrative. It does not claim a measured generation result:
Create a 10-second 16:9 product video of the travel mug in @Image1 on a light
wooden table beside a window. Keep its shape, lid, handle, color, and visible
markings consistent with @Image1. Use soft morning light and a clean commercial
style.
0–3s: stable medium shot; the mug is centered and still.
3–7s: one slow left-to-right camera move while the mug remains still and fully
visible.
7–10s: ease to a stop, keep the mug centered, and hold the final frame.
Preserve markings already visible in @Image1, but add no new text or logos. No
hands, extra products, zoom, orbit, camera cut, or scene change.
The prompt gives the model one role per input: @Image1 controls appearance, while the text controls timing and camera movement. If the intended action is to open the lid, replace the camera move instead of adding both actions to the same time range.
Match detail to the output settings
Choose settings before you polish the wording. The current creator exposes 4–30 second duration presets, six aspect ratios, 480p, 720p, and 1080p output, and a credit preview before submission. A 9:16 short-form clip needs different framing language from a 16:9 product demo. A 30-second sequence also needs more deliberate time blocks than a four-second shot.
Check the prompt before generating
- Does the opening composition agree with the first camera instruction?
- Does each time range contain one main visible action?
- Are references assigned a clear role?
- Do exclusions contradict a requested action, such as asking for readable text and banning all text?
- Is the final frame described?
- Can the prompt still be understood if the adjectives are removed?
If the result misses one part, revise that part only. Keep a simple test log with the prompt version, settings, and the specific failure. If the subject changes across shots, simplify movement and restate continuity rather than adding more style words.
For the full model workflow, see the Seedance 2.5 model overview. For examples of multimodal timing, read Seedance 2.5 Audio to Video, and for broader prompt structure use Seedance 2.5 prompt examples.
FAQ
Is a longer prompt better?
Not automatically. A longer prompt can add useful constraints, but repeated style language and conflicting instructions make the intended action harder to identify.
Should I put the prompt in one paragraph?
No. Short labeled blocks and time ranges make the sequence easier to inspect and revise. The generator receives the text, but the structure helps you catch contradictions first.
What if I need more than one action in a shot?
Separate the actions by time when possible. If they must happen together, state which action is primary and which is secondary; otherwise the result may emphasize the wrong event.
Does the character limit guarantee exact output timing?
No. A concise prompt reduces ambiguity but does not guarantee exact motion or frame timing. Use an external editor when the final cut needs precise alignment.
