Estimate AI video generation cost before running a Sume job

To estimate what an AI video costs on Sume before you pay, read the published rate on API pricing or in GET /v1/catalog, then, where Sume offers one, check the exact request without submitting it: the unbilled Timeline plan and Video filter check return an estimate, and a paid hosted MCP tool such as generate_video called with dry_run=true previews admission and cost. At submit, Sume reserves the estimated amount and captures it when the job completes.

The methods below come from Sume's Timeline 1.0, Generation admission, MCP tools and gates, and API reference docs pages, read on 2026-09-26.

Which ways can I see the cost before paying?

Six places give a price or an estimate without submitting anything. The submit response then shows the estimate Sume reserved:

What do the catalog estimates mean?

GET /v1/catalog is one of the few public routes that need no API key. Paid capabilities carry pricing metadata where available: model_pricing entries with estimated_usd_cents, minimum_usd_cents, maximum_usd_cents, and a pricing_basis description. Called with your API key, an entry can also carry applied_* cents: the estimate your workspace pays, with the model amount and the agent fee as separate lines.

These are rounded, per-model reference figures; the reservation for your own request is stored in exact USD micros at submit. Video models also list pricing_skus on GET /v1/videos/models. Per-second rates multiply out by hand, as AI avatar video API pricing does for avatar video.

Can I check a Timeline render's cost without paying?

Yes. POST /v1/timeline-1.0/plan takes the same body as a render and returns billable_minutes and estimated_cost_usd_micros, with duration_seconds, segment_count, and a filtergraph_summary. It creates no job, reserves no credits, downloads no media, and needs no Idempotency-Key. The render rate is $0.10 per output minute, and the reserve is ceil(audio.duration_seconds / 60) minutes.

The plan's estimated_cost_usd_micros is billable_minutes × that rate, with no agent fee added; the render's reservation adds a 5.5% agent fee by default. Validate a timeline before rendering covers what the plan checks.

Video filter has the same kind of check: POST /v1/video-filter/check returns an estimate when the program is valid, without creating a job or reserving credits.

curl -X POST https://api.sume.com/v1/timeline-1.0/plan \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "audio": {
      "url": "https://media.sume.com/artifacts/artf_demo/voice.wav",
      "duration_seconds": 24
    },
    "video": [
      {
        "source_url": "https://media.sume.com/artifacts/artf_demo/intro.mp4",
        "start": 0,
        "duration": 8
      },
      {
        "source_url": "https://media.sume.com/artifacts/artf_demo/talk.mp4",
        "start": 8,
        "duration": 16,
        "transition": { "type": "fade", "duration": 0.25 }
      }
    ]
  }'

How do I preview a paid MCP call without paying?

If your agent works through Sume's hosted MCP server, the MCP docs name two previews to prefer before expensive bursts: the generation_admission_preview tool, and a paid tool such as generate_video called with dry_run=true, which is an admission and cost preview that does not submit the job. Ordinary single creates do not need them.

After either one, confirm the estimate, the balance, and the queue behavior, then call the paid tool again with dry_run omitted or false to submit.

What happens to the estimate when I submit?

Paid generation uses public Sume USD estimates. When Sume accepts a submit, it reserves the estimated amount; if the balance cannot cover it, the submit fails with 402 insufficient_credits before provider work starts. Job submit responses, such as a Timeline render or an avatar video, carry a usage block with Sume's USD estimate for the request; quote billable_amount_usd.

Successful completion captures the reserved usage, and a failed job releases or refunds the reservation; Do failed AI video generations cost credits? walks through it. For unattended work, a spend cap bounds what a run can spend whatever the estimate said, as Spend caps for unattended AI agents explains.