Text change, region change or background swap: which Sume image route

Choose by what you are changing. Words in the image: pin ideogram/ideogram-v4.5 and name the old and new text. A shape or region you can outline: use ChatGPT Image 2.5 with mask_url. A new background or scene around a product: a reference edit with a "keep the product identical" instruction. This list is a working rule drawn from Sume's docs, not a ranking of models.

The routes

A quick test for each route

Before you commit a catalog to one route, run one image through each and compare. For a text change, check the letters. For a region change with mask_url, check the edges of the mask and whether anything leaked outside it. For a background swap, check the product's edges and any text printed on the product.

The tests take three requests at low cost. They tell you more than any summary, including this one, about what your images need.

Mask details to get right

For a mask_url edit, the mask is a public HTTPS URL. The Image API docs name it for ChatGPT Image 2.5 edits. Keep the mask the same size as the image and use one edit per region, so you can see which mask caused which result. Several stored posts on this site cover mask sizes and alpha channels in depth.

Why the split

Masks and words are different problems. A mask says where to change and lets the model fill. Words say what to change, and the model must find them. Ideogram's launch post (read 2026-10-05) pitches 4.5 on precision and multi-turn editing, so a text-only change fits it. Sume's docs list mask_url for ChatGPT Image 2.5 edits, so a region change belongs on that route. If you send mask_url to a model that does not list it, the docs say Sume returns 400 unsupported_parameter rather than ignoring it.

Chaining routes

Real jobs often need two. Mask-edit a region on one model, then run a text pass on another: two billed images and two jobs, each started from the previous result URL. Keep each pass to one change and read the output before you spend on the next.

Whatever you pick, pin the model id for anything you plan to repeat. Auto is fine for exploration, but the docs are clear that it will not tell you which family ran.

If you are unsure, start with the Image API on a pinned model for the edit you need and keep Auto for requests where you genuinely do not mind which family answers. Auto is convenient, but it hides the family, so it makes a comparison across runs harder. For production workloads, a pinned id and a recorded prompt make results easier to reproduce. Keep the scope of this advice in view. It rests on the Sume docs and the vendor pages named in the sources, read on 2026-10-05, and on nothing measured by Sume. Where a behavior depends on your own images, such as how a model redraws a certain typeface, run a small pilot at the low quality tier and judge the result yourself before you plan a batch. Write down the prompt, the model id and the quality tier you used, so the run can be repeated. When the catalog or the docs change, re-read them; the live catalog is the contract, and a post is only a snapshot of it.