Flow image edits keep the original in History: the Sume edit API

When you edit an image in Google Flow you do not lose the original: the History panel lists every previous version, and the standard prompt box defaults to Nano Banana Pro (read 2026-10-02). Sume's image API behaves differently: each edit call returns a new image URL and your reference URL is left as it was. History is something you keep yourself, using job ids.

What Flow documents

Flow's image help page says editing preserves the original and documents changes, with all previous versions and their prompts in a History panel. It also names Nano Banana Pro as the default model and lets you set aspect ratio and the number of outputs (Flow Help, read 2026-10-02).

The Sume edit call

On Sume you edit with POST /v1/images, a catalog model such as google/nano-banana-pro, a prompt, and input_references holding public HTTPS image URLs, up to 10 on that model (Image API). The response carries data[].url pointing at Sume-hosted media, plus the billed cost.

For edits, send aspect_ratio: "auto" to match the reference; leaving the field out is not the same as auto. A model only accepts the parameters its catalog descriptors list, and anything else returns 400 unsupported_parameter.

Rebuilding History on Sume

Sume has no History panel in the API. A workable pattern is to store, per edit, the job id, the prompt and the input URL in your own table. Jobs are durable, so the id lets you read the result again later (Jobs and results).

Because every edit gives a new URL, a chain of edits is just a list: image 0, then the edit that used image 0 as its reference, and so on. Roll back by using an earlier URL as the next reference.

Which model to pin

Flow picks Nano Banana Pro unless told otherwise. On Sume, name the model you want so a series stays consistent; sume/auto hides which family ran, which is a poor fit for a matched series.