Higgsfield Genjutsu Prompt: How to Write Prompts That Rebuild Reality

A Higgsfield Genjutsu prompt tells the engine what your rebuilt video should become: the new character, the new location, the new wardrobe, the new grade. Motion and camera stay exactly as filmed, so the prompt only has to describe the new reality. Genjutsu Lite handles clips up to 30 seconds. Plans start at $19/mo (checked 4 Nov 2026).

What a Genjutsu prompt actually controls

Genjutsu is Higgsfield’s reality-manipulation engine: upload a video, and it rebuilds the scene with a new cast, location, and products, or swaps specific objects inside the frame. The official page is explicit about what the prompt does not need to do: the movement and camera stay exactly as filmed. Your prompt never has to describe motion. It describes identity, place, clothing, light, and look.

This is the mental shift that makes Genjutsu prompts work. Text-to-video prompts are mostly about action: who does what, how the camera moves. Genjutsu prompts are mostly about replacement: who is doing it instead, where they are instead, what they wear instead. Write the prompt as if the shoot already happened and you are recasting it.

The anatomy of a working Genjutsu prompt

Strong Genjutsu prompts name five things. First, the subject: age, build, look, and expression of the new performer. Second, the setting: the exact location and its details, not just a city. Third, wardrobe and props: the garments, colors, and objects that must appear. Fourth, light and grade: golden hour, neon night, soft studio, cinematic teal-and-orange. Fifth, the constraint line: what must not change, such as keep the camera path and timing identical.

Weak prompts fail on specifics. A cool dancer in a city gives the engine nothing to hold onto, so it invents everything and the result looks generic. A woman in her 30s with short red hair, wearing a black leather jacket and silver hoop earrings, dancing on a rain-soaked Tokyo street at night, neon reflections on wet asphalt, cinematic grade, keep the camera path and timing identical gives the engine a complete casting brief. The difference in output quality is immediate.

Prompt examples for Motion Transfer

Motion Transfer takes the movement from your video and performs it with a new cast in a new location. The prompt should therefore describe a full new world, because nearly everything visible will be replaced. A dance clip filmed in a bedroom becomes: a professional dancer in her 20s, athletic build, long braided hair, wearing an emerald silk dress that flows with each spin, performing on the main stage of a packed concert hall, dramatic spotlights, confetti in the air, keep the original choreography and camera timing.

Notice what the prompt never mentions: the steps. The engine already has them. It also never mentions the bedroom. Motion Transfer discards the original setting, so describing the old location is wasted words. Spend those words on the new one instead: architecture, crowd, light, weather, texture. For identity-critical rebuilds, attach reference images of the face or costume and name them in the prompt.

Prompt examples for Object Swap

Object Swap keeps your cast and location and changes specific elements inside the frame. Here the prompt must be surgical: name exactly what changes and exactly what stays. A product demo becomes: replace the white ceramic mug on the table with a matte black travel tumbler with the uploaded logo, keep the woman, the kitchen, the lighting, and the hand movement identical.

The official use cases show why this matters: one ad, every market means the same shoot with different products or styling for each audience. The prompt is doing localization work, so it should read like a localization brief: product name, color, material, label detail, and the constraint that everything else stays. Vague swap prompts change things you wanted to keep; named swap prompts do not.

Reference images: the prompt’s best friend

Genjutsu supports References: uploaded images of the exact character, product, or outfit the rebuild must use. This is the single biggest quality lever in Genjutsu prompting. Text can describe a sneaker; a reference image shows it. For brand work, product shots, and face identity, references turn the prompt from a description into a specification.

Use references whenever the output must match something in the real world: your product’s label, your client’s packaging, your own face. Our Higgsfield prompting guide covers the general prompting habits that apply here too, including how specific nouns beat adjectives for consistent results.

How to write and run a Genjutsu prompt, step by step

1. Watch the source clip twice (0 credits). First for motion, second for what must change. Note the camera: is it locked off, tracking, handheld? Note the light direction, because your new grade should match it.
2. Pick the technique (0 credits). Motion Transfer for a new cast and world; Object Swap for the same scene with changed elements. Write this down before the prompt, because it decides which half of the scene you describe.
3. Draft the five-part prompt (0 credits). Subject, setting, wardrobe and props, light and grade, constraint line. Aim for two to four sentences; every noun should earn its place.
4. Attach references (0 credits). Upload the product, outfit, or face images the rebuild must match, and name them in the prompt. This step is optional for experiments and mandatory for client work.
5. Generate (free generations or credits). Starter includes 5 free Genjutsu generations per month; Plus and Ultra include 3 at 720p plus 5 Genjutsu Lite generations at 1080p. Genjutsu Lite handles clips up to 30 seconds for everyday trends.
6. Judge the rebuild against the source (0 credits). Check identity first: does the face, product, or outfit match? Then check the constraint: did the camera and timing survive? Fix the failing half with a targeted re-prompt, not a full rewrite.
7. Iterate on one variable (free generations or credits). Change the wardrobe or the setting or the light, never all three at once. Single-variable iteration is how you learn what each line of the prompt actually controls.

Common Genjutsu prompt mistakes

The most common mistake is describing motion: she spins faster does nothing, because the spin is already filmed. The second is prompt-stuffing: twenty adjectives for a dress the camera barely shows, while the location gets one word. The third is forgetting the constraint line, which is how backgrounds and camera moves drift. The fourth is swapping techniques mid-prompt: asking for a new cast and a new product in one generation splits the engine’s attention. One technique, one prompt, one variable at a time.

Who should NOT use it

Skip Genjutsu prompting if you do not have source footage worth rebuilding: the engine needs real motion to work with, and prompting from a blank slate is what text to video is for. Also skip it if you need deterministic output, such as exact legal disclaimers rendered on screen. Prompts steer the rebuild; they do not guarantee pixel-exact text or logos, which is why reference images exist for the details that must be right.

Frequently asked questions

What is a Higgsfield Genjutsu prompt?

It is the text brief you give Genjutsu describing what the rebuilt video should become: new character, location, wardrobe, and grade. The motion and camera stay exactly as filmed, so the prompt only covers the new reality, not the action.

How long should a Genjutsu prompt be?

Two to four sentences is the sweet spot: subject, setting, wardrobe and props, light and grade, plus one constraint line. Every noun should earn its place. Overstuffed prompts dilute the details that matter, so edit instead of adding.

Do I need reference images for Genjutsu?

Not for experiments, but yes for anything identity-critical: products, faces, brand packaging. References turn the prompt from a description into a specification, and they are the most reliable way to keep a logo or face consistent through the rebuild.

Can Genjutsu prompts change the camera movement?

No. The movement and camera stay exactly as filmed; that is the core design of the feature. If you want different camera work, shoot or choose different source footage, or use a text-to-video model with camera controls instead.

How many free prompt attempts do I get?

Starter includes 5 free Genjutsu generations per month; Plus and Ultra include 3 at 720p plus 5 Genjutsu Lite generations at 1080p. Each generation runs one prompt, so iterate one variable at a time to make every attempt count.

Why does my Genjutsu output look generic?

Generic output comes from generic prompts: a cool dancer in a city gives the engine nothing specific to render. Name the age, garments, architecture, lighting, and grade, add the constraint line, and attach references for anything that must match reality.

Sources checked

• Higgsfield official Genjutsu page (reality manipulation, motion transfer, object swap, references, Genjutsu Lite specs) – https://higgsfield.ai/genjutsu – checked 4 Nov 2026
• Higgsfield official pricing page (free Genjutsu generations by plan) – https://higgsfield.ai/pricing – checked 4 Nov 2026