sume/auto for an image series: why to pin a model id instead

For a series that has to look alike, pin one catalog id instead of sume/auto. The Sume Image API page says Auto picks the family for you and never discloses which one ran: it is not listed in GET /v1/images/models, and job.model stays sume/auto. With a pinned id such as google/nano-banana-2, the model in the result is the one you asked for, and the per-image price you read from the catalog is the price you pay.

Everything here is from the Sume Image API page, read 2026-10-02.

What does sume/auto tell me back?

Little by design. The docs say model in the response echoes the id you requested, so sume/auto stays sume/auto, and upstream provider identity is not disclosed. cost is the USD amount billed to your wallet. Because the family is not named, you cannot tell from the response which model produced a given frame.

What does pinning give me?

Three things the docs spell out: the id you send is the id echoed back; GET /v1/images/models/{model_id}/endpoints shows that model's supported parameters and pricing lines; and a request that sets a parameter the model does not list returns 400 unsupported_parameter instead of being silently dropped.

When is Auto still the right choice?

When you do not care which family renders and would rather not track the catalog: one-off drafts, internal mockups, quick tests of a prompt. For a campaign set, a comic, or product shots that sit next to each other, the docs give you no promise about which family renders each call, so the safe move is to choose.

How do I pick the id?

Call GET /v1/images/models, read supported_parameters for the features you need (references, aspect_ratio values, quality), then read the pricing line on the endpoints record. Send that id for every frame in the series, and keep the same reference image and prompt block across calls. Sume also accepts legacy bare ids such as nano-banana-2 as aliases for their org/slug equivalents, but a new integration should send the full id.