Who makes RefVid, and how we test it

RefVid is built and run by Louis Chi from Ireland; the legal notice names Louis Chi as its operator. Louis Chi writes the guides on this site, makes the example clip for every template and answers the mail at hello@refvid.com.

Every guide shows the date it was last updated and the name of its author, linked to this page.

You save a person, a pet, a mascot or a product once, from 1 to 5 photos. Then you describe a scene or pick a template, and RefVid makes a 3–30 second video in which that subject keeps the same face, outfit, label or shape, clip after clip.

Videos run on Kling 4 by default, with Kling 3.0 Omni as the lower-cost option, out of 8 models in the studio. Up to 3 subjects can share one shot, a clip can continue into the next one, and no video carries a watermark.

Every template has an example clip, and every example is a real generation, not a mock-up. It is made with the same request the studio sends when you press Generate, on the model the template runs on, from reference photos of an AI-made subject. Its caption names the model and the number of photos.

A template's prompt is the prompt that made its example, word for word, and a check in our build fails if the two drift apart without a stated reason. Clips are shown as the model made them, uncut and not retouched, only compressed for the web, and each keeps a record of its model, date and request.

Before a new model goes into the studio, we run it on our own reference sets, such as the AERO 1 shoe and the character Mia, and check whether the subject holds. A model's lengths, resolutions and prices come from its maker's documentation, dated when we checked them, and a comparison says so whenever two models were not run on the same references.

Write to hello@refvid.com about an account, a payment, a refund, a bug or working together. The postal address is in the legal notice.