Select models and write motion prompts
Use current capability discovery to choose a model and supported settings, then write focused motion instructions and approve the resulting quote.
On this page
Discover before choosing
Studio lists currently available models for the selected input mode. API callers use the model catalog; MCP callers use list_models. Documentation examples are not a live availability or pricing table.
Choose the mode first, then a model, resolution, supported duration, and ratio. Recheck the quote after each meaningful change. A bigger image or longer clip does not guarantee a better action for your use case.
| Returned capability | What to check |
|---|---|
modes and support flags |
Whether the requested input workflow is allowed |
durations_by_resolution |
Which durations apply to the selected resolution |
ratios_by_mode |
Ratios allowed for this input mode |
max_reference_images |
Total reference count including the first image |
reference_image_duration_seconds |
Whether reference mode fixes duration |
requires_prompt |
Whether a nonempty prompt is required |
Use exactly the returned model ID. Do not reconstruct IDs from a marketing name. Omitted optional settings may use service defaults; store explicit supported values when repeatability matters.
Describe one action clearly
A useful prompt names the subject, action, camera, and important constraints. For a sprite, a starting point could be:
The same character runs in place, side view, fixed camera.
Keep the entire body and equipment inside the frame.
Preserve the costume and proportions.
For an endpoint transition, describe how to arrive at the end pose. For reference images, state the desired action instead of treating the image order as timing instructions.
These are requests to the model, not guarantees. Inspect the preview before exporting. If the action is cropped, address the source framing or input canvas, not only the wording.
Use optional controls only when supported
Negative prompts, seed, watermark control, last-frame output, and reference images differ between models. An unsupported optional field can fail validation instead of being ignored.
A fixed seed is not a promise of identical pixels across providers, model revisions, or other changed parameters. Keep the full request and returned identifiers if you need to compare runs.
Avoid combining several unrelated actions or conflicting camera instructions in one short clip. Change one major factor at a time so you can understand what changed; each new generation still requires its own quote and approval.
Keep model exploration safe for AI clients
Model discovery does not spend generation credits. A quote is also not approval to generate. An AI should explain the chosen model, input set, settings, output formats, and proposed credit amount before calling a paid tool.
If a preferred model is unavailable, report the difference and ask for approval of a materially different replacement. Never silently increase duration, resolution, or a spending cap to get around a failure. See MCP safeguards and generation troubleshooting.
Was this page helpful?
No search queries, code, or free text are collected. This switch controls documentation interactions only; general site analytics follow the privacy policy. Privacy policy