The video generator starts from the task: text-to-video, animate a still, morph, extend, or build from an audio track. The picker then shows only the models that can do it.
Pick a task On the Generate tab under Video, the action chips define the input rather than the model. Text needs only a prompt. Animate and Extend start from an image or a clip. Morph takes two frames, Scene takes reference images, From audio starts from a music or voice track, and Lip Sync takes a face plus a voice. Further along the row sit the tasks that change a clip you already have; the guide on transforming a clip covers those.
Choose a model Model cards show quality and speed ratings, capability tags and a starting token price. The live price updates as you change duration or resolution.
Seed it (for image-based tasks) Animate can take any library image or the current frame of a clip. Morph takes a start and an end frame. Extend continues an existing clip using its last seconds as context.
Generate variants Run up to 4 variants of the same prompt. They render in the background, and the nav pill tracks progress while you keep editing.
Review and place The review picker compares variants side by side; apply the winner straight to the timeline (replace a clip, insert after, or add to the library).
Tips
Describe motion as well as content: "slow dolly toward the window, dusk light" beats "a room".
Generation history (the History tab) keeps every run, so you can reuse a prompt or re-apply an old result at any time.
Failed runs refund their tokens automatically.
A still portrait plus a voice track is its own task (Lip Sync). You do not need to animate the photo into a clip first.