Seedance 2.5 reference limit: BytePlus says 50, Sume varies by model

BytePlus allows up to 50 multimodal references for one Seedance 2.5 clip. Sume documents limits per model, so read the catalog for the id you send. Examples.

4 min readSume
All posts

BytePlus says Dreamina Seedance 2.5 allows up to 50 multimodal references for a single clip, covering images, video, audio, and text. Sume does not have one global reference number: limits are documented per model, so read the catalog entry for the id you actually send.

The 50 figure is from BytePlus's Seedance 2.5 announcement (read 2026-09-30). For the exact per-type split and the Sume request cap for seedance-2.5, see Seedance 2.5 reference limits. This post covers why the number changes by model.

How do reference limits differ across Sume models?

The same word, "reference", means different caps on each route.

Reference inputs as documented by Sume, read 2026-09-30: https://docs.sume.com/models/video-router and https://docs.sume.com/models/video
Route or modelReference imagesReference videos
Gemini Omni Flash 1.1 (Video Router)Up to 10Up to 3, each 3 s or less
Video 1.0 (reference_*_urls)1 to 9 image URLs1 to 3 video URLs
BytePlus Seedance 2.5 (vendor)Part of 50 totalPart of 50 total

What does Sume say about seedance-2.5 itself?

The Video Router docs state that seedance-2.5 accepts 4 to 30 seconds at 480p, 720p and 1080p. Its catalog row lists reference images, reference videos and reference audio as supported. Counts per type are not repeated on that page, so confirm them with GET /v1/video-router/models/seedance-2.5 before you build around a number.

Does the vendor's 50 apply to my Sume request?

Treat it as BytePlus's limit on BytePlus. A request on Sume is validated against Sume's catalog, so a count that is valid on the vendor can be rejected here. Check, send fewer, and split scenes if you need more assets across a video; see Seedance 2.5 @ tags for addressing the ones you do send.

What is the practical plan?

Prioritize the two or three references that carry the look, generate one shot at a time, and join the clips afterward. More references give the model more to follow, not a guarantee that each one is used.

Sources

Related posts

More in Models

All Models posts

Written by Sume