AI Video With Your Own Face: The Models That Accept a Real Photo (November 2026)

Live figures last changed · 5 min read · The GenMagic team

Diese Seite auf Deutsch

To make an AI video with your own face, use a model that accepts a photo of a real person: Wan 3.0 puts you into a new scene from one or more photos, Kling 3.0 and Veo 3.1 bring a photo to life as the opening frame, and HeyGen Avatar IV makes your photo say a script, lip-synced. ByteDance's Seedance turns real faces down on GenMagic and in most apps.

This guide covers which models take a real photo, putting yourself in a new scene, bringing a photo to life, making your photo talk, the photo to use, what it costs, and questions.

A note on who wrote this: GenMagic runs every model named here from one balance. The examples below were made in its studio and through its API in November 2026 from a photo of a person made with an AI image model, so nobody real is shown; with your own photo the models work the same way.

Table of contents

  1. Which models take a real photo
  2. Put yourself in a new scene
  3. Bring a photo to life
  4. Make your photo talk
  5. The photo to use
  6. What it costs
  7. Questions

Which models take a real photo

ModelHow it uses your photoClip lengthSound
Wan 3.0As a reference: you appear in a new scene the prompt describes; several photos at once (you, a second person, a place)2 to 30 sYes
Kling 3.0As the opening frame: the clip starts on your photo and moves from there3 to 15 sYes
Veo 3.1As the opening frame, and you can say a short line with sound4 to 8 sYes
HeyGen Avatar IVYour photo says a script word for word, lip-synced, in a voice you pickAs long as the scriptSpeech
Seedance (2.0, 2.5)Turns photos of real people down

Seedance's refusal is a rule of ByteDance's, the same in most apps that run it (a few that buy it from ByteDance offer a likeness agreement instead; the "may contain real person" error, explained). Seedance still takes drawn, painted and 3D characters.

Put yourself in a new scene

Pass your photo to Wan 3.0 as a reference and describe the scene: where you are, what you do, how the camera moves. The model builds the scene around you and keeps your face, hair and clothes from the photo.

The woman from the reference photo, in the same beige T-shirt, walks along a seaside promenade at golden hour, then stops, turns to the camera and smiles. Keep her face, freckles and hair exactly as in the photo. Handheld camera, warm natural light, the sound of waves and gulls.

  • Say "the person from the reference photo" and what must stay the same (face, hair, clothes). Change the clothes in the prompt if the scene needs it.
  • Two people: attach one photo of each and name them ("the man in the first photo, the woman in the second"). One person per photo works better than a group photo.
  • Long takes: Wan 3.0 makes up to 30 seconds in one take; for a longer video, continue from the last frame and cut the clips together.

This is what people used Sora's cameos for. Sora closed on April 26, 2026 ( OpenAI's help center); every replacement, by use: Sora alternatives.

Bring a photo to life

Pass the photo as the opening frame on Kling 3.0, Wan 3.0 or Veo 3.1 and describe a small, natural motion. The clip starts exactly on your photo, so it suits portraits, family photos and old photographs.

More on old photos: animate old photos.

Make your photo talk

For a talking head that says your exact words, HeyGen Avatar IV animates the face in your photo to speak a script, lip-synced, in one of its voices, for as long as the script runs. You pay for the seconds of speech.

For a short line inside a scene with the room's sound, Veo 3.1 can also speak from an opening photo: put the line in quotes in the prompt. Talking photos step by step: talking photo.

The photo to use

  • One person, face clearly visible, looking roughly at the camera, in even light.
  • No sunglasses, hats low over the eyes, or heavy filters: the model keeps what it can see.
  • Sharp and reasonably large: a phone photo is fine; a tiny crop from a group photo is not.
  • Your own photo, or one you have the right to use, of adults who agreed to it. Do not make a real person appear to say or do something they did not.

What it costs

Every clip draws its own price from your GenMagic balance, shown before you generate. Today:

ClipPrice today
Wan 3.0, 6 s at 480p with sound (the example above)$0.27
Wan 3.0, 20 s at 480p$0.90
Kling 3.0, 5 s at 720p with soundno longer listed
Veo 3.1 Lite, 8 s at 720p with sound$0.60
HeyGen Avatar IV, per second of speech$0.075

A new account gets $0.50 of free credit once its email is verified, top-ups start at $5 and never expire, and nothing renews unless you choose a monthly plan. Try the example on Wan 3.0, then swap in your own photo.

Questions

Which AI video generator works with my own face? Wan 3.0 (as a reference, in any scene), Kling 3.0 and Veo 3.1 (as the opening frame) and HeyGen Avatar IV (a talking photo) all accept a photo of a real person. Seedance does not.

Can I put myself into an AI video from one selfie? Yes, with Wan 3.0: attach the selfie as a reference, describe the scene and say what must stay the same. One clear photo is enough; a second angle helps.

Can I make an AI video of myself talking? Yes: HeyGen Avatar IV makes your photo say a script word for word, lip-synced, in a voice you pick. For one short line inside a scene, Veo 3.1 can speak from an opening photo.

Why does Seedance refuse my photo? ByteDance's Seedance models turn down any picture that looks like a real person, your own face included, on GenMagic and in most apps. Use Wan 3.0 or Kling 3.0 for real photos ( why).

What replaced Sora's cameos? Sora closed on April 26, 2026. For putting yourself into a generated scene, Wan 3.0 with your photo as a reference does the same job.

Can I use a photo of someone else? Only with their agreement, and never to make them appear to say or do something they did not.

Is there a free way to try it? A new GenMagic account gets $0.50 of credit once its email is verified, about 11 seconds of Wan 3.0 at 480p with sound.