Price an AI video before you submit: Sume plan calls and catalog

Sume lets you read a price before you spend anything, in two places. The timeline plan call prices an edit, and the model catalogs list per-model prices for generation. Neither reserves credits.

Timeline plan: exact for the edit

POST /v1/timeline-1.0/plan compiles the request without creating a job. Per the timeline docs it returns duration_seconds, segment_count, billable_minutes, estimated_cost_usd_micros and a filtergraph summary. It also needs no Idempotency-Key. One limit: the docs say a plan cannot predict short-source pad or loop warnings.

Model catalogs: per-model prices

The video docs point to GET /v1/videos/models for each model's capabilities and prices, and the image docs do the same with GET /v1/images/models, whose pricing lines carry a billable unit and cost_usd. Image lines already include Sume's margin, so a request for n images costs cost_usd times n. For video, the amount held at submit is the provider list price times 1.25.

A routine that avoids surprises

• Plan the edit and note estimated_cost_usd_micros.
• Read the per-second price of the generation model and multiply by total seconds.
• Add the flat jobs, such as captions at $0.20.
• Compare the total with GET /v1/balance before you submit.

What the previews cannot tell you

A price read is not a promise about outcome. The plan call cannot predict pad and loop warnings for short sources, and a catalog price is per unit, not per project. Neither tells you how many takes you will need. The reserve at submit can also differ from the final capture: for jobs priced by actual compute, the docs say the capture is never more than the hold.

Use the previews to set an upper bound, then compare the final usage row with it. Over a few projects you will learn your own ratio of estimate to spend.

Catalog numbers can change, which is why the docs keep telling you to read the live endpoint. A blog post is a snapshot dated 2026-10-04.