The 5 aspect ratios all 19 Sume image rows accept (a portable set)

Five aspect ratios are listed by all 19 rows in Sume's image catalog on 2026-10-10: 1:1, 16:9, 9:16, 4:3 and 3:4. If you compare models, use only those, and no row will answer with a 400 for ratio. Anything outside the five, such as 4:5 or 21:9, is accepted by some rows and rejected by others.

I computed the intersection from the catalog code on origin/main, which builds the descriptors the list route publishes, and checked it against the Image API docs. Run the same check yourself on the live list route, since rows are added over time.

Who drops out beyond the five

The rest of the ratio space is uneven. Imagen 4 rows list only the five. Soul lists seven. Seedream 5.0 Lite and 4.5 list nine, Nano Banana Pro eleven, and several rows 13. Nano Banana 2.1 and Ideogram rows list 15.

Why a portable set helps

A bake-off compares rows on equal terms. If one row cannot accept the ratio, you either drop it or pick a different ratio for it, and the comparison stops being fair. With the five, every row runs the same prompts at the same shapes.

The five also cover the main placements: a square, a landscape and a portrait video frame, and the two classic photo crops. Feed portrait at 4:5 is not among them, which is a good reason to test it separately.

Checking it in code

Fetch the list route, read each row's aspect_ratio.values, and intersect the sets with a few lines of Python: start with the first row's set and loop &= through the rest, skipping auto. If a new row removes one of the five, the check will show it and your test plan can change before a run.

Remember that listing a ratio only means the API will accept it. How a model composes a picture at that shape is something you judge by looking.

• Use the five for cross-model tests.
• Test 4:5 and 21:9 separately and only on rows that list them.
• Re-run the intersection when a row is added.

Extending the set row by row

After the five, add ratios as you confirm each row's support. 4:5 and 5:4 and 3:2 and 2:3 appear on most of the rows with 9 or more ratios, but not all: Grok Imagine lacks 4:5 and 5:4, and Seedream rows have their own nine. Rather than memorize it, store the intersection of the rows you actually use and regenerate it whenever you add one.

If you serve many placements, keep a second table that maps each placement to the ratio and to the set of rows that can serve it. That table is the real asset: it says what you can produce and from where.

A last practical point: when you publish results from a bake-off, say which ratios you used. A comparison at 1:1 says little about a banner at 21:9, and readers who know the catalog will ask. The five-ratio set is honest about what it covers, and it leaves the wider claims to separate tests with narrower row lists.

Also note that auto is not one of the five. Only six rows list it, so a test that sends auto will fail on the other rows with a 400, and a comparison should always pass an explicit ratio.

Keep the intersection result in your repository next to the date, so a reviewer can see what the set was when the test ran.