AI video generation
Prepare short scenes from prompts and supported references. Video Studio helps you compare a model's available duration, resolution and billing unit before submission.
Open Video StudioKeep the scene focused
Describe who is in the frame, what changes during the shot, the camera movement and the intended sound. A focused scene is easier to review than several unrelated actions squeezed into one request.
Plan longer edits as separate scenes within the chosen model's output limits. Character identity, facial expression, spoken language and lip synchronization are not guaranteed by a reference upload alone.
Reference and output limits are model specific
Supported inputs may include images, video and audio, but availability and limits differ. Check individual file limits, combined counts, media durations, aspect ratios and output resolution in the current model details.
Video Studio can filter models by token, second or task billing. For token models, the amount shown before submission is an estimate; final settlement uses actual measured tokens. A longer reference can also affect a model's billing conditions.
Generation is an asynchronous task
After a request is accepted, wait for its task status rather than repeatedly submitting it. API clients use the task ID to check progress and download the completed file through the authenticated content endpoint.
A submission timeout can leave acceptance uncertain. Keep any returned request or task ID and check its status before considering another billable request.
Download results you need to keep
Review movement, shot continuity, audio and mouth movement before publishing. Download completed videos promptly.
Some video models have a server-only retention policy of no more than 24 hours after completion and are not saved to Cloud Works. The model notice identifies when this applies; other models should not be assumed to have the same policy.