AI video generator

The Video studio covers the three ways a clip actually gets made: from a written prompt, from a still you already have, or from an existing video you want restyled. Those are separate job types under the hood — text-to-video, image-to-video and video-to-video — and each has a different set of models that are good at it, so the studio switches the model list when you switch the input.

Model choice is the whole game here. Veo and Sora are strong on prompt adherence and physical plausibility; Kling and Seedance hold character consistency better across a shot; Wan and LTX are faster and cheaper for iteration. You are not locked into one — the form re-renders from each model's own schema, so switching from a model with a duration control to one with an aspect-ratio enum changes the controls rather than silently dropping the parameter.

Generation runs server-side and durably. Once a job is submitted you can close the tab: the work is checkpointed, and the finished file lands in your gallery whether or not the browser that started it is still open.