Seedance 2.5 concert prompt uses 18 images: fitting Sume's 9

Seed's Seedance 2.5 concert example uses 18 reference images. Sume's Video Router takes at most 9 for seedance-2.5, so merge groups into sheets.

5 min readSume
All posts

Seed's own Seedance 2.5 concert prompt uses 18 reference images, and Sume's Video Router rejects more than 9 reference images on seedance-2.5 with the message "reference_image_urls must not exceed 9 files." To reproduce the idea, give each principal performer one image and merge each group (orchestra, choir, audience) into one contact sheet, which brings the 18 down to 8.

The prompt is in ByteDance's Seedance 2.5 launch post (read 2026-10-02); the cap is from Sume's request validation for the Video Router and the OpenAPI schema in the API reference.

What does Seed's 18-image prompt reference?

It is a 30-second 16:9 concert. @Image 1 is the venue, @Image 2 the pianist, @Image 3 the cello, @Image 4 the violin, @Image 5 the lead vocalist, @Images 6 to 10 the rest of the orchestra, @Images 11 to 14 the choir and @Images 15 to 18 the audience seating. Seed says Seedance 2.5 takes up to 30 images, 10 video clips and 10 audio clips as references in one pass, so 18 images is well inside its own limit.

Reference groups in Seed's concert prompt (read 2026-10-02) and one way to fit Sume's 9-image cap on seedance-2.5.
GroupImages in Seed's promptAs one sheet each
Venue11
Pianist11
Cello11
Violin11
Lead vocalist11
Rest of the orchestra51 sheet
Choir41 sheet
Audience seating41 sheet
Total188

What are Sume's limits for seedance-2.5 references?

On the Video Router, reference_image_urls may hold up to 9 images, reference_video_urls up to 3 and reference_audio_urls up to 3, with no more than 12 reference files in total. The OpenAPI schema lists higher array ceilings (10, 5 and 5) because they are shared with other models, but the per-model check for Seedance is tighter. On POST /v1/videos the check for seedance-2.5 is the 12-file total across input_references.

Read the limit from the endpoint you call, and expect a 400 rather than a silent drop.

How do I merge a group into one reference?

Put the group on one image: a grid of the orchestra members on a plain background, or one photo of the choir, uploaded to a public HTTPS URL. Then name the sheet in the prompt ("@Image 6 shows the orchestra; place them on both sides of the stage"). Whether one sheet holds identity as well as several separate images is something to test; this post does not claim it does.

Keep the list order and the numbers in the prompt in step. If you drop from 18 to 8 images, renumber the tags, or the prompt will point at the wrong picture. How @image tags work has the tag syntax.

What does the request look like?

Eight references on the Video Router, with the principal performers first. Nothing here is more than the 9-image cap.

{
  "model": "seedance-2.5",
  "prompt": "30-second concert in 16:9, cinematic realism. @Image 1 is the venue. @Image 2 pianist, @Image 3 cello, @Image 4 violin, @Image 5 lead vocalist. @Image 6 is the orchestra, @Image 7 the choir, @Image 8 the audience. Open on a high-angle wide shot of the hall.",
  "reference_image_urls": [
    "https://example.com/venue.jpg",
    "https://example.com/pianist.jpg",
    "https://example.com/cello.jpg",
    "https://example.com/violin.jpg",
    "https://example.com/vocalist.jpg",
    "https://example.com/orchestra-sheet.jpg",
    "https://example.com/choir-sheet.jpg",
    "https://example.com/audience-sheet.jpg"
  ],
  "duration": 30,
  "resolution": "720p",
  "aspect_ratio": "16:9"
}

Sources

Related posts

More in Models

All Models posts

Written by Sume