One model for 1:1, 4:5, 9:16, 16:9 and 21:9: 6 Sume rows do it

Six of the ten Sume image rows checked list all five of 1:1, 4:5, 9:16, 16:9 and 21:9: Nano Banana 2.1, Nano Banana Pro, FLUX.2 pro, FLUX.2 flex, Recraft V4 and Qwen Image Max. The cheapest is FLUX.2 pro at $0.0375 per image, so one image per placement costs $0.1875 in total. GPT Image 2.5, Seedream 5 Lite, Grok Image and Imagen 4 Fast each miss at least one of the five. Recraft V4 and Qwen Image Max only generate from text and do not edit images, and Recraft V4 returns WebP only, so check that before you pick either for an edit job.

A campaign that needs a feed square, an Instagram portrait, a story, a landscape video cover and an ultrawide header can then use one model for the whole set. That keeps the look consistent and means a single price to put in the budget.

Which rows list which ratios

The ratio lists below are the values each model's catalog descriptor advertises through GET /v1/images/models. A model accepts only the values it lists, and any other value returns 400 unsupported_parameter. Prices are Sume's per image at the default option as of 2026-10-09. For Nano Banana 2.1 and Pro the default is the 1K tier.

Why the misses matter

GPT Image 2.5 lists 1:1, 4:5, 9:16 and 16:9 but not 21:9, so an ultrawide header needs another model or a crop. Seedream 5 Lite has the same gap. Grok Image lists 20:9 and 19.5:9 but no 4:5 and no 21:9. Imagen 4 Fast lists only five ratios and has neither 4:5 nor 21:9.

Cropping is the other route. A 16:9 render cropped to 21:9 loses about 24 percent of its height, since 16/9 divided by 21/9 is 0.762 and so the 21:9 frame keeps 76.2 percent of the rows. For a hero banner where the subject sits in the middle that is often fine, but it is a different image from one generated at 21:9.

Check your own set in code

The script keeps the five required ratios and the lists you copied from the catalog. It prints the models that cover all of them. Replace the lists with live data from GET /v1/images/models/{model_id}/endpoints before you rely on it.

REQUIRED = {"1:1", "4:5", "9:16", "16:9", "21:9"}

catalog = {
    "flux-2-pro": {"1:1", "16:9", "9:16", "4:3", "3:4", "4:5",
                   "5:4", "3:2", "2:3", "21:9", "9:21", "1:2", "2:1"},
    "gpt-image-2.5": {"1:1", "16:9", "9:16", "4:3", "3:4", "5:4", "9:8", "4:5"},
    "imagen-4-fast": {"1:1", "16:9", "9:16", "4:3", "3:4"},
}

for model, ratios in catalog.items():
    missing = sorted(REQUIRED - ratios)
    print(model, "ok" if not missing else f"missing {missing}")

Budget note

One image per placement is the minimum. If you want three options per placement, send n: 3 and multiply: three options at five placements on FLUX.2 pro is 15 images, or $0.5625. The Image API docs list the limits on n. Read usage.cost on the first response to confirm the billed amount.