By Higgsfield Genjutsu editorial teamPublished Oct 21, 2026Updated Oct 30, 2026
What is a Genjutsu dance video?
A Genjutsu dance video starts with choreography you already recorded and rebuilds the dancer, outfit or world around it. The source clip keeps the timing, body motion and camera, which makes the output useful for social edits, performance concepts and character-driven dance content.
This is not a template library where one image automatically invents a dance. Your source performance is the choreography. The references decide who performs it and what the finished reality looks like.
Choose the workflow by your starting asset.| You already have | Best starting workflow | Why |
|---|
| Your own dance footage | Genjutsu Motion Transfer | The original choreography, camera and timing stay as the performance anchor |
|---|
| A character image plus a stock motion template | Viggle or another template-driven dance tool | Those products are built to animate a still character with supplied motion |
|---|
| Only a text description | A text-to-video model | Genjutsu needs real reference footage rather than inventing a dance from a prompt |
|---|
How is this different from Viggle and dance generators?
Viggle is built around controllable character animation and motion sources, including a dedicated character dance generator. Genjutsu’s advantage is the production clip: a real take, with its choreography, spacing and camera movement, can be recast as a different performer or scene.
Choose Genjutsu when the source dance is already right. Choose a template-driven tool when you want to apply a stock dance move to a character image without shooting footage first.
- Genjutsu: existing choreography + reference images → rebuilt dancer and world.
- Viggle: character image + motion source or template → animated character.
- Template dance tools: stock choreography is the starting asset.
- Text-to-video tools: the performance is generated rather than preserved.
What dance footage works best?
Use one continuous 3 to 30 second take with the dancer fully visible. The strongest clips have stable lighting, a readable silhouette and one dominant movement direction. Full-body framing is more useful than a tight portrait because the references and prompt have to preserve the whole performance.
Trim before a cut, a camera whip or a crowded section. A short clean chorus is a better master than a full routine with multiple angles. Longer dances are recast shot by shot and assembled in the editor.
- Keep feet, hands and head visible for most of the clip.
- Use an even exposure that does not hide limbs in shadow.
- Avoid overlapping backup dancers unless you intentionally want a multi-character recast.
- Match the source aspect ratio to TikTok, Reels, Shorts or the final campaign placement.
What references keep the dancer consistent?
Give the model a complete identity set instead of a single face photo. Use a front view, profile, full-body view and outfit details. If the character has a signature prop, mask or hairstyle, include an image where it reads clearly.
Keep the identity references internally consistent. Mixing different hairstyles, body proportions or costume versions forces the model to guess what should survive through the dance.
How do you write a dance recast prompt?
Protect the choreography first. Name the source motion, timing, camera and footwork before describing the new character. Then introduce the reference for the performer and the scene or outfit as separate decisions.
One render should change one main idea. If the dancer, costume and entire stage all change, that is a full Motion Transfer recast. If only the jacket changes, Object Swap may be cleaner.
Character dance recastRecast the dancer as the character in reference set A. Preserve the exact choreography, hand positions, footwork, camera movement and timing. Keep the original stage and lighting.
New stageKeep the same performer, dance, camera and timing. Rebuild only the environment as the neon rain stage in the reference, with the ground plane and light direction matching the source.
Outfit changeChange only the dancer’s outfit to the lime technical costume in the reference. Preserve identity, body proportions, choreography, camera and background.
How do you keep complex choreography readable?
Keep the first test simple. Choose a section with one dancer, one camera angle and one clear direction of movement. Then expand to quicker footwork, floor work or overlapping performers after identity and timing survive.
Beat matching happens after generation. Let the source clip establish the action, then cut the result to the track in your editor. Asking the model to invent a new edit rhythm creates a second, competing timing system.
Can Genjutsu recast two or more dancers?
Yes, when the dancers are clearly separated and each person has a distinct reference set. The simple performance should still be one continuous take. Crossing paths, lifts and hand-offs reduce reliability.
Label the left and right performer in the prompt, keep reference order consistent and protect the gap between bodies. For complex contact, split the clip into shorter sections and join them in post.
What goes wrong in an AI dance recast?
The usual failures are changing hand ownership, unstable feet, costume drift and identity loss during fast turns. These come from low-information frames and ambiguous references, not from the idea of dance itself.
Fix one variable at a time. Add a clearer profile reference for face turns, a full-body reference for silhouette, and a costume detail for wardrobe continuity. If the source motion is too blurred, choose another section rather than writing a longer prompt.
- Feet slide or merge: trim to a stable grounded section and keep the full body in frame.
- Hands trade owners: use a single dancer test before a group edit.
- Face changes on a spin: add a profile reference captured under similar light.
- Costume drifts: reduce the motion window or rebuild the full character instead.
- Stage and dancer compete: change only one of them in the first pass.
What should you check before publishing?
Watch the output once for identity, once for timing and once for anatomy. Place the source and result side by side to see whether the groove survived. A clean first pose is not enough if the footwork stops belonging to the beat.
Use footage, choreography and music you are licensed to publish. If the clip shows real people, secure likeness permission and follow the platform’s synthetic-media disclosure rules where applicable.
Key takeaways
- Genjutsu recasts existing choreography rather than generating from a still image.
- One dancer, one angle and one clean section is the strongest starting point.
- Use full-body references and protect camera, timing and footwork in the prompt.
- For two dancers, use separate reference sets and explicit left-right ownership.
- Cut to the beat after generation; do not ask the model to redesign the timing.
Higgsfield Genjutsu