Genjutsu prompts: 12 templates you can copy right now

How Genjutsu prompts work

Genjutsu is a reference-driven video model, not a text-to-video model. Your source clip already carries the motion, the camera and the timing, so the prompt never has to describe them. It only has to say which element changes and what that element becomes.

That is why good Genjutsu prompts are short. A long cinematic paragraph leaves the model nowhere to apply your reference images, while a two line instruction about a single element usually lands on the first attempt.

The prompt formula

Every prompt on this page is built from the same three parts, in this order:

• What changes - name the element plainly: "the defender", "her jacket", "the mountains behind them".
• What to use - point at the reference image you attached for that element: "the woman in image 1".
• What must stay - name the one thing you cannot afford to lose: the choreography, the product label, the camera move.

Prompt library: wardrobe, products and props

Prompt library: characters and casting

Prompt library: locations and scenes

Prompt mistakes that waste credits

• Describing the motion or the camera. Your clip already defines both, so restating them fights your own footage.
• Asking for several unrelated changes in one run. A prompt that swaps the cast, the location and the wardrobe at once usually returns none of them cleanly.
• Naming an element without a reference. "the pink jacket" is guesswork; "the jacket in image 2" is not.
• Padding with quality tokens like cinematic, 8k or masterpiece. They do nothing for a video-to-video model and pull the prompt away from your references.
• Forgetting the thing that must stay. If the product label or the punchline has to survive the recast, say so explicitly.

Matching the prompt to the mode

Motion Transfer keeps the motion, the camera and the timing, and rebuilds the scene from your references. Prompts for this mode read like casting and staging decisions.

Object Swap changes one selected element and preserves the rest of the original shot. Prompts for this mode are a single instruction: this thing, replaced by that reference.