Reference to video generates footage guided by images you upload. Instead of describing your subject in words and hoping the model gets it right, you show it — and the output stays consistent with the reference across frames and takes.

Image to video animates a single still frame — your image is the first frame. Reference to video uses images as identity and style anchors: the model composes new scenes and motion while keeping the subject faithful to your references.

Clear, well-lit shots of the subject from a representative angle. For characters, a clean face shot works best; for products, a hero image on a simple background. You can add multiple references to lock identity further.

On AirCube, reference-guided generation runs on Seedance, Kling, Wan, Hailuo and more — and the model picker shows exactly what each engine supports so you can compare results side by side.

Start with one clean, representative shot — that's enough for most reference to video generations. Add two or three more angles when identity drifts: a front-facing portrait plus a profile locks a character much harder than a single photo.

Yes — the reference anchors who or what appears, and your text prompt directs the scene: motion, camera, lighting and mood. That combination is what makes reference to video work for storyboards and campaigns.

Yes — generated videos can be used in commercial projects subject to each model provider's terms. Make sure you hold the rights to any reference images you upload.