Seedance 2.5, Wan 3.0, Kling 4.0: which model ids Video Router accepts

Video Router lists 12 catalog ids plus three Auto spellings. seedance-2.5 and wan-3.0 are there; Kling 4.0 is not, and kling-3 is the Kling id. Limits per id.

5 min readSume
All posts

Sume's Video Router accepts 12 catalog ids, and seedance-2.5 and wan-3.0 are two of them. Kling 4.0 is not an id you can call on Sume today: the only Kling entry is kling-3 (Kling Video v3 Pro), capped at 4 to 15 seconds at 720p or 1080p with no reference inputs. If a tutorial tells you to send a Kling 4.0 model name, the request will not match anything in the catalog.

The list below is read from the OpenAPI enum and the catalog code. The OpenAPI model enum has 15 values: those 12 ids plus the Auto spellings auto-, sume/auto and auto.

The 12 ids at a glance

Input shape and limits differ a lot between ids, which is why the docs say to read capabilities from GET /v1/video-router/models rather than assume one envelope. higgsfield-genjutsu appears in that listing only when its provider is configured.

Video Router catalog ids, duration and resolutions (read 2026-10-03)
IdStarts fromDurationResolutions
seedance-2.5Text, first/last frame, references4 to 30 s480p, 720p, 1080p
seedance-2, seedance-2-fast, seedance-2-miniText, first/last frame, references4 to 15 s480p, 720p, 1080p
wan-3.0Text, first/last frame, references2 to 30 s480p, 720p, 1080p
kling-3Text, first/last frame4 to 15 s720p, 1080p
minimax-h3Text, first/last frame, references5 to 15 s480p, 768p
minimax-h3-maxText, first/last frame, references5 to 15 s480p, 768p, 1080p
gemini-omni-flash-1.1Text, frames, references, or a source video to edit3 to 10 s360p, 720p, 1080p, 4K
grok-imagine-video-1.5A first-frame image (required)4 to 15 s480p, 720p
h3-max-recastOne source video plus 1 to 4 person photos5 to 30 s768p, 1080p
higgsfield-genjutsuOne source video plus 1 to 8 reference images4 to 30 s480p, 720p

Choosing between the long-clip ids

Four ids reach past 15 seconds: seedance-2.5 and wan-3.0 both stop at 30 for a normal generation, while the person-swap and motion-transfer ids (up to 30 as well) take their length from the source video. Wan 3.0 is the only id with a 2-second floor, so it is the one for very short hooks; Seedance 2.5 starts at 4 seconds.

Both accept reference images, reference videos and reference audio. Wan 3.0 counts them per type (up to 10 images, 5 videos, 5 audio files), while Seedance rejects more than 12 reference files combined. If your job is a long single take with a few photos, either fits; if it needs a dozen mixed references, check the per-type caps before you pick.

What the news maps to on Sume

Seedance 2.5 maps to seedance-2.5, billed on a per-1,000-video-token basis in the catalog, and Wan 3.0 maps to wan-3.0, whose catalog note dated 2026-08-24 lists $0.05 per second at 480p, $0.10 at 720p and $0.20 at 1080p. Sume bills list × 1.25, so a ten-second 720p Wan clip is $1.00 list and $1.25 billed.

The Kling mention in the news is the one without an id. Until a Kling 4.0 entry appears in GET /v1/video-router/models, there is nothing to pin; use kling-3 for Kling-style output or one of the other ids for features Kling 3 lacks, such as reference images, which kling-3 does not take.

What to do with a name that is not in the list

Do not guess an id. Ask the catalog, then pin what it returns.

``bash curl https://api.sume.com/v1/video-router/models \ -H "Authorization: Bearer $SUME_API_KEY" ``

Each entry carries its own capabilities block with the duration range, resolutions and which reference types it takes. A newly launched model, whether Kling 4.0 or another, becomes callable only when it appears there. Until then, seedance-2.5, wan-3.0 and kling-3 are the long-form, short-form and Kling options on Sume, and sume/auto is the choice when you do not want to pick.

Sources

Related posts

More in Models

All Models posts

Written by Sume