What is Genjutsu: one clip in, a rebuilt world out

What Genjutsu is, in plain words

Genjutsu is a video-to-video AI model for restaging reality: you bring footage that already exists, and it rebuilds the world inside the shot - the cast, wardrobe, location and props - while the performance, camera and timing carry over untouched.

You will also see it described as "reality manipulation" or "reality restaging". Both mean the same thing: instead of generating a video from a text prompt, you rebuild one that was already shot. The edit changes what the video shows; the motion underneath stays real.

Where the name comes from

Genjutsu is the Japanese term for illusion techniques - attacks that rewrite what a character perceives while reality keeps moving underneath. The model borrowed the name for the same reason: the video you see is transformed, but the motion it is built on is real footage.

So two meanings coexist whenever someone searches for genjutsu: the fictional technique, and the AI tool named after it. This page is about the tool - what it does, and what it refuses to fake.

The two ways people use it

• Motion Transfer - a full restage. The whole scene is rebuilt from your references while the choreography, camera and timing stay frame-anchored.
• Object Swap - a point edit. One element is replaced - a jacket, a bottle, a sign - and the rest of the original frame survives.

What stays, and what gets rebuilt

• Stays from your clip: the performance, the camera moves, the cuts, the pacing, contact points like footsteps and punches, and the overall timing.
• Rebuilt from your references: the people in the shot, their wardrobe, the location, products and props, and the grade and atmosphere.
• Decided by your prompt: which of those elements change in this particular run, and which stay as they are.

What Genjutsu is not

• Not a filter or an effect preset - the footage is regenerated, not decorated.
• Not a text-to-video generator - the motion comes from a real clip, not from a prompt.
• Not a face-swap app - it rebuilds the shot around your references instead of pasting one face onto another.
• Not a lip-sync tool - performance timing transfers, but it is not built to dub dialogue word for word.

Why clip-first matters

Generated video struggles with the things audiences notice first: hands, physics, weight, continuity. Genjutsu sidesteps most of that by starting from footage where those things are already true - a real person really moves - and only replacing what the shot shows.

That is why music videos, ads and character-driven series lean on it: one great take becomes dozens of variants without reshooting, and every version still moves like the original.

How to try it

• Open the generator on this site and sign in with Google - nothing to install.
• Upload a clip of 3 to 30 seconds and attach 3 to 6 reference images per element you want to change.
• Pick Motion Transfer or Object Swap, write a short three-part prompt, check the credit cost and run.
• The Genjutsu free guide covers exactly what costs nothing and the cheapest way to go further.