What can I create with Genjutsu Motion Transfer?

Genjutsu Motion Transfer lets video creators build short motion-transfer clips from existing footage and reference images. Supply a source video at least four seconds long, then add images and optional text instructions. Longer source videos are trimmed to thirty seconds before the output is generated.

What output does this motion-transfer model provide?

Higgsfield returns a generated video at either 480p or 720p resolution, with 720p as the default setting. The output duration follows the prepared source clip, capped at thirty seconds. Genjutsu Motion Transfer accepts up to eight reference images rather than requiring a text prompt alone.

inference · 467.0s

Example inputhover

prompt: "Replace the two dancers in the source video with the people from the reference images. @Image 1 replaces the woman on frame right. @Image 2 replaces the man on frame left. Keep each person's face, hair, skin tone, body and clothing exactly as in their reference image. Both faces must stay clearly visible and recognizable in every frame, including during fast moves and the half-turn around 6 seconds. Keep everything else from the source video unchanged: the static camera, single shot, framing, graffiti wall, golden-hour light, boombox, the full 10-second duration, and the original audio. Transfer the choreography exactly: every step, bounce, arm move, turn and the final pose, with the same timing and perfect sync between the two dancers. The original dancers must not remain visible in any frame. No identity drift, no blended faces, no swapped positions, no camera movement, no added text, logos or watermarks."
video_url
image_urls
resolution: "720p"

Genjutsu Motion Transfer

Image Gen·higgsfield·by Higgsfield

Genjutsu Motion Transfer creates videos from source motion and reference images. Guide short-form video edits with optional prompts and 480p or 720p output.

Try it now →

API reference

Runtime (p50): -
Estimated price: $0.159–$0.3405 per second, depending on resolution

Call the API

prediction.sh

curl -X POST \
  -H "Authorization: Bearer $EACHLABS_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
    "model": "genjutsu-motion-transfer",
    "input": {
        "prompt": "Replace the two dancers in the source video with the people from the reference images.\n\n@Image 1 replaces the woman on frame right.\n@Image 2\n  replaces the man on frame left.\n\nKeep each person's face, hair, skin tone, body and clothing exactly as in their reference image. Both faces must stay clearly\n  visible and recognizable in every frame, including during fast moves and the half-turn around 6 seconds.\n\nKeep everything else from the source video\n  unchanged: the static camera, single shot, framing, graffiti wall, golden-hour light, boombox, the full 10-second duration, and the original audio. Transfer the\n  choreography exactly: every step, bounce, arm move, turn and the final pose, with the same timing and perfect sync between the two dancers.\n\nThe original\n  dancers must not remain visible in any frame.\n\nNo identity drift, no blended faces, no swapped positions, no camera movement, no added text, logos or\n  watermarks.",
        "video_url": "https://cdn-us.eachlabs.ai/defaults/27d176a28e944232a2f6f6406d795b76.mp4",
        "image_urls": [
            "https://cdn-us.eachlabs.ai/defaults/507b1b74be064da3a51f0cc9d9134f8e.png",
            "https://cdn-us.eachlabs.ai/defaults/77a8e0ba7bec4785b386e729bcdb3a64.png"
        ],
        "resolution": "720p"
    },
    "webhook_url": ""
}' \
  https://api.eachlabs.ai/v1/prediction/

Related models

4 models

Hover Preview SpritesUnified AI API for hundreds models. Access video, image, and audio generation from leading providers through one integration with eachlabs.

Clip SegmenterUnified AI API for hundreds models. Access video, image, and audio generation from leading providers through one integration with eachlabs.

ABR Ladder PackagerUnified AI API for hundreds models. Access video, image, and audio generation from leading providers through one integration with eachlabs.

Minimax H3 Max RecastMinimax

* FAQ

About Genjutsu Motion Transfer

01 / 03

What is Genjutsu Motion Transfer?

Genjutsu Motion Transfer is a Higgsfield video model that combines motion from a source clip with one to eight reference images. An optional text prompt guides the generation. It supports reference-driven video editing on each::labs, with a video file as the final output.