Is there one API for Kling, Seedance and Veo with a single bill?

Yes. Several APIs bill many video models to one key and one balance, but their model lists differ, so check that all three are on the list. On their own pages, fal and Replicate list Kling, Seedance and Veo models; Runway's API lists Seedance and Veo, with no Kling row in its video pricing; Higgsfield's API model reference names Kling and Seedance, not Veo, among its video families; Kling's own API bills Kling models only.

Sume's POST /v1/videos is one of these APIs: one API key and one workspace balance across 11 video model ids in 6 families (Kling, Seedance, Wan, Grok Imagine, MiniMax and Gemini Omni Flash), and model: "sume/auto" lets Sume pick the model. It covers Kling and Seedance. Veo is not in Sume's catalog today, so a Veo call still needs another provider.

Sume facts come from the Video generation docs, the Video Router docs, and Sume's pricing code. Facts about fal, Replicate, Runway, Higgsfield, and Kling come from each company's own pages. All were read on 2026-09-29.

Which APIs bill Kling, Seedance and Veo on one account?

The table lists the multi-model APIs in alphabetical order, plus Kling's own API for comparison. Check four things: whether the models you need are listed, the unit each model is priced in, how a request picks the model, and what the account limits are.

Which video models does Sume's one API cover?

Every model below is called on the same route with a bare catalog id, never an org/slug id. GET /v1/videos/models lists them with each model's supported resolutions, aspect ratios, durations, and pricing_skus. Limits differ by model: seedance-2.5 and wan-3.0 run up to 30 seconds, gemini-omni-flash-1.1 takes 3–10 seconds, and the rest stop at 15 seconds or less; AI video length limits by model has the full list.

If you only need one family, calling that vendor directly is the other option; Sume vs Kling and Sume vs Seedance compare the two routes.

How do I switch models without changing code?

Change the model field. The rest of the request, the job, polling, and the download stay the same. An Idempotency-Key makes retries safe only with the same key and the same body, so send a new key with the new body. Check the target model's supported_durations, supported_resolutions, and supported_aspect_ratios first.

The route follows OpenRouter's video API field for field, so a client written for that schema works after changing the base URL and the key; An OpenRouter-compatible video API covers the differences. To skip choosing, send sume/auto: responses echo sume/auto, and Sume does not disclose which family ran. Auto defaults to 720p and 8 seconds, with 3–10 second clips at 16:9 or 9:16.

curl -sS -X POST "https://api.sume.com/v1/videos" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: orbit-001" \
  -d '{
    "model": "seedance-2",
    "prompt": "Cinematic product orbit on a clean table",
    "duration": 5,
    "resolution": "720p",
    "aspect_ratio": "16:9"
  }'

# Same body on another family: change "model"
# (for example "wan-3.0") and send a new Idempotency-Key.

What does a single bill mean on Sume?

One workspace balance pays for every model, and one ledger records it:

• One balance: Sume's pricing FAQ says video generation, Sume Agent, Formats, and the API all draw from one balance at each model's published rate. You add to it with a top-up from the Billing & subscription page in the dashboard; the API reads the balance but has no top-up endpoint.
• One billing rule: every video model is reserved from the workspace USD balance on submit at the provider's list price × 1.25, plus a 5.5% agent fee by default. What is the Sume agent fee? shows where the fee is reported.
• Where rates live: API pricing lists the Video Router as pass-through, per model (provider list rates × Sume margin), with per-model rates in the catalog. For video models, GET /v1/videos/models returns those rates as pricing_skus.
• One ledger: GET /v1/balance and GET /v1/usage read the workspace's balance and usage ledger, GET /v1/usage?job_id=… sums what one job cost, and a finished job's usage.cost is the Sume billable amount. Reserved usage is refunded after a failure, or a cancellation before capture.

Can I call Veo through Sume?

No. Veo is not in Sume's video catalog today, and no Sume model id calls it.

If you need Veo on the same account as other models, three multi-model pages list it: fal's pricing shows Veo 3 at $0.4 per second, Runway's API pricing shows veo3.1 at 20 credits per second without audio and 40 with audio, at $0.01 per credit, and Replicate's text-to-video collection lists google/veo-3.1.

What should I check before choosing a multi-model video API?

One bill does not mean one price or one feature set. Each point below applies to every API in the table:

• The model list for your account. Higgsfield's FAQ says model availability depends on your account; on Sume, GET /v1/videos/models is the list your key can call.
• The markup over the direct price. Sume bills the provider's list price × 1.25, plus a 5.5% agent fee by default; Runway's Model Router bills the standard rate of the model it selects. Compare with the model maker's own price, such as Kling's per-second units, before high volume.
• The price unit. Rates come per second, per video, per 1,000 video tokens, or in credits, and they change with resolution and audio, so price a sample clip on each API.
• Provider-specific options. On Sume, a non-empty provider.options returns 400 unsupported_parameter, no v1 model accepts seed, and size is not supported; send resolution plus aspect_ratio instead.
• Jobs, webhooks, and failures. fal, Higgsfield, and Sume run video as async jobs you poll or receive by webhook. Sume takes an HTTPS callback_url and refunds reserved usage after a failure; Higgsfield says a failed request is not charged, Replicate says the same for a failed run of a public model (private models and deployments are billed for instance time); fal says you are never charged for server errors.
• Limits. Sume's processing concurrency is set by plan, from 1 job on Free to 20 on Scale in the docs' default table, and prepaid top-ups do not raise it; a full queue answers 429 queue_full. Video job concurrency and queueing covers it. fal ties concurrency to credit purchases, and Higgsfield's rate limits depend on the account and model.

Which option fits which need?

Pick by the models you need first, then by unit price and API shape:

• Pick fal or Replicate if you need Veo next to Kling and Seedance on one account: both list all three on their own pages.
• Pick Runway's API if Seedance and Veo cover your needs and you want its Model Router to choose between models; its video pricing lists no Kling model.
• Pick Higgsfield's API if Kling and Seedance are enough and you want a cost estimate before each request.
• Pick Kling's own API if you only use Kling and want to pay Kling's own unit prices.
• Pick Sume if you want Kling, Seedance, and the other 4 families in its catalog behind one route, with sume/auto, per-job usage reads, and one balance. For the wider platform comparison, see fal.ai and Replicate alternatives for video.
• Sume is the wrong pick if you need Veo or another model outside its catalog, a seed, provider-specific options, or one family at that vendor's list price.