Wan 3.0 references on Sume: 10 images, 5 videos, 5 audios

A Sume job on wan-3.0 accepts up to 10 reference images, 5 reference videos and 5 reference audios, which is more than Seedance 2.5's 9, 3 and 3. The API returns an error that names the limit if you exceed it.

The Sume docs say the Seedance 2.x models, Wan 3.0 and the MiniMax H3 models accept audio and video references, while Gemini Omni Flash 1.1 accepts video but not audio.

Limits table

Sending references

On /v1/videos you pass input_references entries of type image_url, video_url or audio_url. On the Video Router path the fields are reference_image_urls, reference_video_urls and reference_audio_urls. Every URL must be public HTTPS.

• Do not mix first/last frame fields with reference fields in one Video Router call.
• Keep file counts well under the cap while you tune a prompt.
• If a job fails, check that each URL downloads without a login.

When the larger cap helps

Ten image references is enough for a full product line-up, or a cast of four with two angles each plus a location. Five video references lets you pass several short motion examples, and five audio references covers voice, music and effects beds.

Common reference failures

Reference URLs must be public HTTPS and available when the provider fetches them; check each with a plain curl -I before you submit.

Over-cap requests are rejected up front with a message naming the field, such as model wan-3.0 accepts at most 5 reference_video_urls. Audio references also need at least one image or video reference.