Video Router: what a request with no resolution or duration costs

If you send a Video Router request without resolution or duration, Sume still has to price it, and the default is not always the cheap one. On Wan 3.0 an empty request is priced as a 1080p, 5-second clip: $1.25. The same empty request on MiniMax H3 is a 768p, 5-second clip: $0.38. The defaults below come from the pricing code in the Sume repository, and the rates come from the catalog the Video Router serves.

Sume bills provider list price times 1.25, rounded up to the cent, on every model. That is the same rule the video generation docs give for the reserve placed on submit. So a default is not a free choice: it decides how much of your balance is held before the first frame renders.

What does each model assume when fields are missing?

The estimator fills the gaps with a fixed duration and resolution per model family. Wan 3.0 uses fal's own default of 1080p and 5 seconds. MiniMax H3 and H3 Max use 768p and 5 seconds, which the code describes as Sume's product default, native 768p rather than an upscale tier. Gemini Omni Flash 1.1 uses 720p and 8 seconds. Kling 3 has a single per-second rate that depends only on audio, and the estimator assumes generate_audio is true and 5 seconds when you omit both.

Why does the Wan default matter most?

Wan 3.0 is the only model in the table whose default sits at the most expensive tier. At 480p the same 5 seconds costs $0.32, at 720p $0.63, and at 1080p $1.25. A caller who copies a minimal example and never sets resolution pays the 1080p price every time.

Kling 3 is the other trap, in a different direction. Audio is on unless you send generate_audio: false, and sound-on costs $0.21 per second against $0.14 with it off. A silent 5-second clip is $0.70 instead of $1.05.

How do I stop paying for a default?

Set the two fields explicitly in every request, even when the default would have been fine. A pinned request also prices the same way on every retry. The sample below sends a deliberately small Wan clip.

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: wan-draft-001" \
  -d '{
    "model": "wan-3.0",
    "prompt": "A paper boat crossing a puddle, overhead shot",
    "resolution": "480p",
    "duration": 5,
    "aspect_ratio": "16:9"
  }'

What does the reserve look like on a longer clip?

Defaults only cover missing fields, so the cost of a clip still scales with the duration you send. Wan 3.0 accepts 2 to 30 seconds, and each second at 1080p is $0.25 on Sume. A 30-second Wan clip at 1080p therefore holds $7.50, while the same length at 480p holds $1.88.

Kling 3 accepts 4 to 15 seconds in the catalog, so the longest Kling clip with sound is $3.15. Gemini Omni Flash 1.1 tops out at 10 seconds, which is $1.25 at its default 720p. The docs describe the reserve as placed on submit at list times 1.25, so these figures are what a clip holds, not a separate fee.

A practical habit: when you build a request template for a team, put resolution and duration in it with the cheapest tier that passes your review, and let people raise them deliberately. That keeps a forgotten field from becoming the most expensive setting in the catalog.

Which defaults can I trust to stay put?

Treat the table as a snapshot, not a contract. The catalog at GET /v1/video-router/models carries each model's capabilities and pricing block, and the docs say to read capabilities from it rather than assuming one envelope. The Wan constraint line in the catalog notes its list prices are dated 2026-08-24 and subject to change. Re-read the endpoint before you budget a batch, and set both fields in the request so a change in a default cannot move your bill.