AI motion transfer & face swap

Genjutsu: make anyone dance with AI motion transfer

Upload one photo and pick a dance. Genjutsu turns it into a video of your character doing the moves, with the same face, outfit and style from start to finish. No camera, no motion capture, no 3D rig.

  • 1 photo in
  • 3–30 s clips
  • up to 1080p
  • Photo, 3D, anime
Made from this one photo

Examples made with Genjutsu

Each clip started as a single still image and a reference dance. The same hip-hop routine drives a street dancer and a man in a mustard suit; the same K-pop routine drives a 3D cyber ninja and an anime idol.

  • Street dancer × Hip-hop groove

    Photoreal

  • Cyber ninja × K-pop flow

    3D render

  • Anime idol × K-pop flow

    Anime

  • Mustard suit × Hip-hop groove

    Photoreal

What is Genjutsu?

Genjutsu is an AI video generator for motion transfer, face swap and character swap. You give it a picture of who should appear in the video and a clip of how they should move, and it renders a new video that combines the two.

The name comes from the Japanese word for the art of illusion. That is the idea: a convincing performance by someone who was never filmed doing it, whether that is you, a brand mascot, a game character or an anime idol.

What is AI motion transfer?

AI motion transfer, also called pose transfer or video-driven animation, separates appearance from movement. The engine reads a reference performance as a sequence of body, hand and face poses, then renders a completely new video where your character performs that same sequence.

Because every frame is generated rather than filtered, the result keeps the lighting, texture and art style of your source image. It is the fastest way to make an AI dance video from a photo.

How Genjutsu works

Three steps from a still image to a finished video.

  1. 01

    Upload a photo

    One clear, full-body photo of a person or character. Photos, 3D renders, illustrations and anime art all work.

  2. 02

    Pick a dance

    Choose a routine from the dance library or upload your own 3–30 second reference clip. Genjutsu tracks the body, hands and head in every frame.

  3. 03

    Generate

    Describe the scene if you like, then render. Your character performs the motion with the same face, outfit and style in every frame.

Try it

Pick a character and a dance to preview the result, then make your own.

Character
Dance
Generate video (opens in a new tab)

Motion transfer vs. face swap vs. character swap

Three techniques that are often confused. Here is what each one does, and when to use it.

AttributeMotion transferFace swapCharacter swap
What changesThe whole character performs new motionOnly the faceThe whole performer
What staysIdentity, outfit and style from your photoBody, clothes, scene and motionBackground, camera and lighting
Input1 photo + reference motion1 portrait + target video1 photo + scene video
Best forDance trends, avatars, mascotsPersonalized clips, cameosRe-casting, previs, localization

Motion transfer generates a new video of your character, head to toe, performing the reference motion.

Face swap keeps an existing clip intact and replaces only the face, keeping the original expressions, blinks and lip movement.

Character swap replaces the whole performer but leaves the room, camera move and lighting untouched.

Features

Motion transfer, face swap and character swap share one identity-preserving engine.

  • Full-body motion transfer

    Footwork, spins, freezes and hand gestures from any reference clip, applied to a character from one photo.

  • Video face swap

    Replace a face in a video while keeping expressions, lip movement and head turns.

  • Character swap

    Swap the whole performer and keep the background, camera move, lighting and timing.

  • Identity lock

    Face, hairstyle and wardrobe stay consistent across every frame, so the character never drifts.

  • Any art style

    Photoreal stays photoreal, 3D stays 3D and anime keeps its line art and cel shading.

  • Prompt-directed scenes

    Change the location, lighting or camera with a sentence without touching the choreography.

  • Dance library

    Start from ready-made routines such as hip-hop and K-pop, cleaned up for motion transfer.

  • Original audio

    Keep the soundtrack from the reference video, in sync with the moves.

  • Up to 1080p

    Clips of 3–30 seconds, delivered as MP4 files that play everywhere.

Who uses AI motion transfer

From solo creators to agencies, it replaces the shoot, the mocap stage and the animation queue with one upload.

  • Creators and short-form video

    Post your avatar, mascot or alter ego doing a dance trend the day it goes viral on TikTok, Reels and Shorts, without learning the choreography.

  • Music artists and labels

    Turn a press photo into a performance clip for a release, a lyric video or a fan challenge, without booking a shoot.

  • Brands and advertising

    Put a mascot or spokesperson into a routine and adapt it for every market. Try new creative in hours instead of weeks.

  • Games, VTubers and virtual idols

    Animate 2D and 3D characters straight from concept art to prototype emotes, trailers and idol performances.

  • Fashion and e-commerce

    Animate lookbook photos into catwalk and dance clips so shoppers see how clothes move, not just a static pose.

  • Film, previs and agencies

    Block out choreography with stand-ins and deliver previs and pitch films from reference footage in an afternoon.

Tips for the best results

Most problems in a motion transfer video come from the inputs. These habits fix the majority of them.

  • Use a full-body photo. Stand the character in a relaxed pose with the feet visible.
  • Keep the reference camera still. Locked-off shots with one performer transfer most cleanly.
  • Match the framing. A full-body photo pairs with a full-body dance; a portrait pairs with a talking head.
  • Avoid heavy occlusion. Props or people crossing in front of the dancer confuse pose tracking.
  • Direct with the prompt. Lighting, location and camera moves are promptable; the choreography comes from the clip.
  • Going long?. Chain several shots with the same identity to cover a full 3–5 minute song.

Use it responsibly

Motion transfer and face swap are creative tools, not a way to put moves or words in someone’s body without asking.

  • Consent first

    Only animate yourself, people who have agreed to it, or characters you own.

  • No public figures or minors

    Don’t upload celebrities, politicians or children.

  • Label it as AI

    Say a video is AI-generated when you post it, and follow each platform’s rules.

  • Don’t deceive

    Don’t use motion transfer or face swap to mislead, harass or impersonate anyone.

FAQ

Genjutsu FAQ

Answers about AI motion transfer, face swap and how Genjutsu works.

What is AI motion transfer?

AI motion transfer takes the movement from one video — a dance, a workout, a gesture — and applies it to a different person or character from a single still image. Genjutsu estimates the full-body pose, hands and head orientation in every frame of your reference clip, then synthesizes new frames of your character performing that exact motion while keeping their identity, clothing and art style consistent.

How is motion transfer different from face swap?

Face swap replaces only the face in an existing video; the original body, clothes and movement stay the same. Motion transfer generates an entirely new video of your character, head to toe, performing the reference motion. Genjutsu does both, plus a third mode — character swap — that replaces the whole performer while keeping the original background, camera move and lighting.

What do I need to upload?

One clear, full-body photo of the character (JPG or PNG, at least 340px on the short side) and a reference video between 3 and 30 seconds (MP4 or MOV, up to 100MB). For face swap, a front-facing portrait is enough. You can also skip the video upload and pick a move from the built-in dance library.

Does it work with anime, 3D and illustrated characters?

Yes. Genjutsu is style-preserving: photoreal photos stay photoreal, 3D renders stay 3D, and anime or cartoon art keeps its line work and cel shading. The showcase above includes a photoreal dancer, a 3D cyber ninja and an anime idol, all driven by the same reference choreography.

How long can a generated video be?

Each render can be up to 30 seconds when the character follows the orientation of the reference video, or up to 10 seconds when the character keeps the orientation of the source photo. Longer performances — full songs of three to five minutes — are produced by chaining shots with a shared identity lock so the character never drifts between segments.

What resolution and frame rate do I get?

Standard renders are 720p; Pro renders go up to 1080p. Output follows the frame rate of your reference clip (typically 24–30 fps) and is delivered as an H.264 MP4 that plays everywhere, with an option to keep the original audio track.

Is it legal and safe to swap faces?

Only animate people who have given you permission — yourself, your talent, or characters you own. Genjutsu blocks uploads of public figures, adds an invisible watermark and C2PA content credentials to every render, and never uses your uploads to train models. See the Responsible AI section for the full policy.

Is there an API?

Yes. Every mode in the studio is available through a REST API with webhooks, so you can batch-render hundreds of personalized dance videos for campaigns, games or apps. Request early access to get API keys.

Your first Genjutsu video is one photo away.

Upload a photo, pick a dance and see your character move.

Start creating (opens in a new tab)