RefVid

Your mascot grows a new nose in every AI clip, while the label on your product scrambles a little more each time. RefVid keeps a person or product looking the same across videos. You save it once as a subject with one to five reference photos, and the first serves as the main view. Each new clip starts from the subject plus a written scene and is kept on the subject's page.

Several saved subjects can share one shot, and a scene photo can set the location. A finished clip can be continued into the next shot of a longer story. Characters speak lines written in quotes, with one of six preset voices or a voice made from a short sample you provide. Prompts in other languages are translated first, and templates start from ads, product shots, pets or holidays.

The video itself comes from Kling and Seedance models made by other companies, which RefVid runs through the fal.ai service, the terms say. Kling 4 is the default, and saved subjects carry over when you switch model in the studio settings.

Seedance 2.5 is meant for products and refuses photos of real-looking people. The terms ban clips that impersonate real people, public figures included, and anyone in an uploaded photo must be you or someone who agreed.