Genjutsu 30 reference images vs Sume's 1 to 8

Higgsfield's Genjutsu takes a 4 to 30 second video and up to 30 reference images per generation, with output up to 1080p. Sume's higgsfield-genjutsu is narrower: one source video, 1 to 8 reference images, 480p or 720p, and 4 to 30 seconds. If you plan more than 8 references or a 1080p output, the Sume limits are the ones to design around.

Limits side by side

The Higgsfield column is from the Genjutsu announcement; the Sume column is from the Video Router and video generation docs.

Staying inside the Sume limit

Pick the 8 references that matter most: the ones that carry the subject's identity or the object you want swapped in. Fewer, clearer references are easier to review than a crowd. The model is called through the Video Router route with video_url plus reference_image_urls, and it accepts video references but not audio.

Because the model is listed only when its provider is configured, read GET /v1/video-router/models before you build around it.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: genjutsu-001" \
  -d '{
    "model": "higgsfield-genjutsu",
    "prompt": "Swap the mug for the ceramic teapot in the references",
    "video_url": "https://example.com/source.mp4",
    "reference_image_urls": [
      "https://example.com/teapot-front.png",
      "https://example.com/teapot-side.png"
    ],
    "resolution": "720p",
    "mode": "async"
  }'

If you need more than 8

• Split the work into two passes, each with its own set of up to 8 references.
• Merge several views of one object into a single contact-sheet image to carry more views per slot.
• For 1080p, plan an upscale as a separate step, since the model tops out at 720p on Sume.