Seedance 2.0 is most useful when an AI video needs more than a striking first frame. A usable shot needs a clear subject, a controlled action, a camera decision, and enough detail to make the output consistent from the beginning of the clip to the end.
This guide explains how to use the Seedance models available in Cinelyo, how to structure prompts, and when to use references or higher output resolution. Cinelyo is an independent AI video service; model names describe the generation options available in the product and do not imply official affiliation.
Start with the production decision
Choose the model after deciding what the clip must do. The best choice is usually about duration, iteration speed, and final delivery requirements rather than a single universal quality ranking.
| Need | Recommended model | Why |
|---|---|---|
| A polished short commercial or cinematic hero shot | Seedance 2.0 | Supports 480p, 720p, 1080p, and 4K output for final-quality passes. |
| Several prompt or camera variations | Seedance 2.0 Fast | Built for efficient exploration at 480p and 720p. |
| Budget-conscious early drafts | Seedance 2.0 Mini | Useful for testing the subject, setting, and shot structure before refinement. |
| A longer single sequence | Seedance 2.5 | Supports selectable output from 4 through 30 seconds at 480p and 720p. |
For a closer duration and resolution comparison, see Seedance 2.5 vs Seedance 2.0.
Write prompts as shot plans
A reliable video prompt is not a list of visual adjectives. It is a compact shot plan. Start with the one action that must remain readable, then add only details that help the model preserve that action.
Use this order:
- Subject and setting: identify the subject, wardrobe or material, and a specific environment.
- Action: describe one short action with a clear beginning and end.
- Camera: choose a lens feeling, distance, and movement such as a slow dolly, lateral track, or held close-up.
- Light and material: add the lighting direction and the surfaces that need to read correctly.
- Constraints: say what must stay stable, such as identity, object geometry, full-body framing, or a single subject.
For example, a product prompt can begin with a clear product silhouette, then add one camera move and one material change:
An unbranded glass bottle on dark stone. The camera makes a slow three-quarter orbit while one warm reflection travels across the surface. Keep the bottle geometry stable and finish on a centered hero frame.
That structure is more useful than asking for a "beautiful luxury commercial" without defining what happens in the shot. The same approach works for fashion, character, architecture, and action work.
Keep each clip focused on one motion idea
Most weak generations try to do too much: a subject turns, runs, changes wardrobe, enters a new setting, and triggers several effects in one short request. The result is often hard to evaluate because no single decision has enough time to read.
Use one primary motion idea per generation. A food commercial might use a single sauce pour. A fashion clip might use a three-step walk and one turn. An architecture reveal might use a slow move from the entry toward a window. When the first pass works, change one variable at a time: the setting, a material, a light source, or the final frame.
The prompt library includes reusable shot plans organized by product, fashion, beauty, character, architecture, cinematic, and previsualization work. Each prompt can be copied or opened directly in the generator.
Use references for decisions that words cannot carry
Text prompts are excellent for shot direction, but reference images, video, and audio are useful when the project depends on a particular look or continuity. Use a reference when you need to protect a specific subject, product shape, layout, or visual atmosphere.
Before uploading a reference, decide what it is meant to control:
- Subject reference: identity, wardrobe, hair, or a repeated character.
- Product reference: silhouette, packaging, material, and fine geometry.
- Style reference: lighting, set design, color restraint, and composition.
- Motion reference: timing, framing, or the shape of an action.
Do not ask one reference to carry every decision. A strong workflow names the visual priority in the prompt, then uses the reference to reinforce that priority. For example, use a product image to protect the packaging and text to direct the camera, lighting, and reveal.
Choose resolution after the direction is approved
Higher resolution is most valuable after the shot itself is working. Start with a compact test at 480p or 720p when you are checking subject stability, camera movement, and the basic action. Once the direction is approved, use Seedance 2.0 at 1080p or 4K when the final destination needs the additional detail.
This sequence protects the budget and improves review quality. A clean low-resolution test tells you more than an expensive final pass built around an unclear prompt.
For longer clips, Seedance 2.5 supports 480p and 720p with a maximum generation length of 30 seconds. For a short 4K hero frame, the full Seedance 2.0 model is the stronger fit. The generator shows the credit requirement before a task is submitted, so you can compare settings before committing to a run.
A practical iteration loop
Use the same loop whether you are creating a social ad or a cinematic proof of concept:
- Write one shot plan with one main action.
- Generate a short draft at an appropriate review resolution.
- Review only the most important variable first: subject, motion, camera, or material.
- Change one variable in the prompt and generate again.
- Add references when a visual decision must remain consistent.
- Move to final resolution only after the composition and motion are approved.
Keep the version that best expresses the intended shot, not simply the version with the most visual effects. A clear, repeatable prompt becomes an asset the next time the team needs a related scene.
Common prompt fixes
The subject changes during the clip
Reduce the number of actions, state that there is one subject only, and define the wardrobe or key physical feature early in the prompt. If identity is critical, use a subject reference and keep the camera move modest.
The camera feels random
Name one camera move and one framing change. For example: "camera tracks backward at chest height, then settles into a medium portrait." Avoid combining a crane, orbit, zoom, handheld movement, and focus pull in five seconds.
The product shape drifts
Use an unambiguous product description, keep the composition uncluttered, and state that geometry must remain stable. A reference image is especially useful for packaging, watches, vehicles, and other precision objects.
The result looks busy
Remove secondary effects before adding more style language. A single lighting change, material detail, or final reveal is often enough to make a short video feel deliberate.
Where to go next
Start with the AI video generator when you have a shot idea ready. Use the prompt library when you need a proven structure to adapt, or compare Seedance 2.0 Fast and Seedance 2.0 Mini when the priority is faster or lower-cost experimentation.