How to turn a photo into a video character

Step 1: consent and likeness rights come first

If the photo shows a real person, you need their permission before you make video of them, and written permission if the video will be published or used commercially. A person's face and voice are protected by right-of-publicity and privacy laws in many places.

In the US these are mostly state laws; Tennessee's ELVIS Act of 2024, for example, explicitly covers voice and likeness. A federal bill, the NO FAKES Act, passed the Senate Judiciary Committee on June 18, 2026, according to its sponsors, but is not law as of October 2026. In the EU, the AI Act's transparency rules in Article 50 have applied since August 2, 2026 and require deepfakes, realistic AI content showing real people, to be disclosed as AI-generated.

In practice: get a signed release from the person that names the use (social posts, ads, a series), the duration, and whether you can edit or alter them. Do not make video of public figures, celebrities or anyone who has not agreed, even as a joke.

Never use photos of minors. Never create sexual, defamatory or deceptive content of a real person. This page is general guidance, not legal advice; for commercial campaigns, ask a lawyer in your market.

Your own photo is the simplest case. Even then, keep in mind that a realistic video of you can be reshared without context.

Step 2: prepare the source photo

Pick the clearest photo you have: the face large in the frame, sharp eyes, even light, neutral or slight smile, no sunglasses, nothing covering the face. Crop to head and shoulders with some space around the hair, and keep it at least 300 pixels on the short side.

Remove filters if you can find the original file. A photo taken for this purpose, against a plain wall in window light, works better than any social media picture.

Step 3: expand one photo into angles

A single front photo gives the model one view; it must invent every other angle, and it can invent them differently each time. The fix is to build a set of three to five matching images before you make video. RefVid takes up to five photos per character.

1. Best option: photograph the person again in one session, same clothes and light, from the front, both three-quarter sides and the profile, plus one full-body shot. Five minutes with a phone is enough.

2. If that is not possible, draw the missing angles from the one photo. RefVid offers this on the character's page: it draws a three-quarter view, a side profile and a full-body view from your photo with Kling's image model, and you pay only for the views that come out. Any image model's edit mode can do the same, one angle per request.

3. Check every drawn angle against the source: ears, jawline, nose shape and hairline should match. Delete any view where they changed and replace it. Keep the source photo first in the set.

Step 4: test, save and reuse

Run four short test clips with the full set, 3 seconds each: a talking close-up, a walk toward camera, a turn of the head to the side, and a shot in different light. RefVid can show a free still of the character in each scene first, so you only pay for clips whose framing already works.

Compare each clip to the source photo. If the person is recognizable in all four, the set works. If the side view fails, the profile reference is weak; replace it.

Save the set under the person's name and reuse exactly that set for every video. Scene prompts should describe the setting, action and camera, and refer to the person by that name, never by their features.

Step 5: label realistic results

If the video is realistic and shows a real person, say that it is AI-generated where you publish it. The EU requires it for deepfakes, and TikTok, YouTube and Meta have their own disclosure rules for realistic synthetic content; use the platform's AI label when posting. If the person appears in an ad, they should know exactly which ads use their likeness.

Common mistakes and checklist

Using a friend's photo because they said yes once in a chat. Relying on one photo for a series. Accepting drawn angles where the ears or nose subtly change. Describing the person's face in the prompt. Posting realistic clips without a label.

Checklist: written consent covering the planned use; no minors and no public figures; a clear source photo; three to five matching angles; four test clips passed; set saved under one name; AI label applied when publishing.

RefVid saves the set once as a character and reuses it for every clip. When you upload, you confirm that the people in the photos are you or adults who agreed; public figures and minors are not allowed.