Genjutsu AI video generator: real motion in, any reality out

What the Genjutsu AI video generator actually makes

Most AI video generators invent a clip from a sentence. Genjutsu works the other way: you bring footage that already exists - a dance take, a product shot, a walk - and the model generates a new version of the scene on top of the real motion. The movement, the camera work and the timing are kept frame for frame; the character, wardrobe, location and props are generated from your reference images.

That order matters for one practical reason: generated motion rarely feels real, while real motion always does. By keeping the motion you filmed, every video that comes out of the generator inherits the same weight and rhythm as the original take.

Three kinds of videos you can generate from one clip

• Full scene rebuild (motion transfer) - keep the motion, replace the person, the outfit and the location in a single pass.
• Object swap - keep the whole shot intact and replace one product, prop or item of clothing.
• Character continuity - generate many videos with the same character from one set of reference images.

Generator, editor, or text-to-video: which one do you need?

Text-to-video tools are for clips that never existed: you describe a scene and the model imagines both the look and the motion. Genjutsu is for clips where the motion is the expensive part - the dancer, the stunt, the product turning in a real hand. The performance stays; the world around it is generated.

A classic editor adjusts what is already there. This generator replaces it. If your shot already performs, that difference is the whole point.

How to start generating, step by step

• Open higgsfieldgenjutsu.lol and sign in with Google - no install and no card to try it.
• Upload a 3-30 second single-shot clip with clear, readable motion.
• Add 3-6 reference images for every element you want generated: character, outfit, location, product.
• Write a three-part prompt: what changes, which reference it uses, and what must stay. One big change per render.
• Run the render, review the result, then download - clean, no watermark on an active plan.
• Your first renders are free; after that, credits are used per render by plan.

Prompts that get the best results from the generator

Prompting here is closer to directing than to writing: you are describing changes against a motion that is already fixed. The three-part formula - change, reference, anchor - survives the most renders.

What you get out of it

• Videos built from 3-30 second source footage, one shot per render.
• Vertical or horizontal output, ready for TikTok, Reels and Shorts.
• Up to 40 reference images per project for characters, products, wardrobe and locations.
• Clean downloads without watermarks on an active plan.

Who gets the most out of a generator like this

• Creators riding a format: one dance, many realities, posted as a series.
• Brands: turn an ad that already converts into fresh variants with new talent, product or location - without a reshoot.
• AI influencers: keep one consistent character across dozens of generated scenes.
• Localisation: the same edit with local streets and faces, for every market.