Video Prompting Tips

Video generation is mostly about good choices, not clever wording — the action you pick and the image you start from matter far more than any text. Here's how to get the best results.

Pick the right action#

The action defines the motion and the scene, so it's your biggest decision. Each action in the library shows a preview thumbnail of what it produces — browse, or use search to jump to what you want, and match the action to the result you're picturing. Actions are split by Realistic / Anime and span general, positions, foreplay, and solo scenes. See Models & actions.

Start from a great image (for I2V)#

For image-to-video actions, the starting image is everything — a bad still makes a bad video, full stop. Perfect the image before you animate it. This is the single highest-leverage tip for I2V; the full explanation is in Text-to-video & image-to-video.

Choose the right model#

• Variety and speed → V1 (biggest action library, ~7.5s clips).
• Longest, highest quality — with sound → V2 (up to 30s, audio).
• A fully custom scene from an image → V3 (Custom).
• Make her talk on camera → V4 (Talking, lip-sync).

Settings that matter#

• Frame rate — 16 FPS generates fastest; 32 FPS is the balanced standard; 64 FPS is the smoothest motion (costs more and takes longer).
• Length — longer clips look better but cost more XOT and take longer to render.
• Upscaling — sharper, higher-resolution output for extra XOT.

Mind cost and render time#

Higher frame rate, longer length, and upscaling all raise both the XOT cost and the render time. The studio shows the exact cost before you generate, so dial the settings to match how much the clip matters. See XOT costs.

Stitch for longer scenes#

Generate a few strong clips and combine them in the video stitch studio to build a longer, sequenced scene rather than one long generation.