Ratio check before a batch: 4:5, 9:16 and 8:1 on three Sume rows
Imagen 4 lists 5 ratios, GPT Image 2.5 lists 9 and Nano Banana 2.1 lists 15. Which of 4:5, 9:16 and 8:1 each Sume row accepts, and the fallback.
Before you run a batch, check the ratio. On Sume, Imagen 4 lists five aspect ratios (no 4:5, no 8:1), ChatGPT Image 2.5 lists nine (4:5 and 9:16 yes, 8:1 no) and Nano Banana 2.1 lists fifteen, including 4:5, 9:16, 4:1 and 8:1.
Lists from the repo
These come from the aspect-ratio catalogs in Sume's API contract package.
| Model | Ratios listed | 4:5 | 9:16 | 8:1 |
|---|---|---|---|---|
| google/imagen-4-fast, imagen-4-ultra | 1:1, 16:9, 9:16, 4:3, 3:4 | No | Yes | No |
| openai/gpt-image-2.5 | auto, 1:1, 16:9, 9:16, 4:3, 3:4, 5:4, 9:8, 4:5 | Yes | Yes | No |
| google/nano-banana-2.1 | auto, 21:9, 16:9, 3:2, 4:3, 5:4, 1:1, 4:5, 3:4, 2:3, 9:16, 4:1, 1:4, 8:1, 1:8 | Yes | Yes | Yes |
What happens on a miss
A model accepts only the values its descriptors list, so an unlisted ratio gives a 400, not a silent change. Read supported_parameters.aspect_ratio from the catalog and compare it with the ratio you want before you queue the job.
Fallbacks
For a missing ratio, render a nearby ratio and crop. Imagen has no 4:5, so generate 3:4 and crop the height. GPT Image 2.5 stops at 3:1 for custom sizes, so an 8:1 banner needs Nano Banana 2.1 or a wide render that you crop. Crop in code and keep the subject away from the cut edge in the prompt.
Costs of the switch
Ratio does not change the per-image price on these rows, so choose by capability. Nano Banana 2.1 is $0.10 at 1K, Imagen 4 Fast $0.025 and Imagen 4 Ultra $0.075.
Automating the check
A short script can read the three descriptors from GET /v1/images/models and print which of your target ratios each row supports. Run it in CI so that a catalog change fails a build, not a customer job.
Target ratios from common placements: 1:1 for a feed tile, 4:5 for a portrait feed post, 9:16 for a story or reel cover, 16:9 for a thumbnail, 4:1 for a wide header and 8:1 for a leaderboard strip. Only Nano Banana 2.1 covers all six on Sume.
When you crop, render at a higher tier than you ship. A 3:4 render cropped to 4:5 loses about 6% of its height, so ask for the smallest crop that works.
The ratio lists come from Sume's API contract package, which the catalog reads. They are correct for the repo state on 2026-10-08. A fresh GET /v1/images/models is the answer for your account today. If the two disagree, trust the API.
Note that auto is in the lists for GPT Image 2.5 and Nano Banana 2.1 but not Imagen. On edits auto matches the reference image, and omitting the field is not the same as sending auto, as the docs state.
One more point on cost: ratio does not change the per-image price on the Imagen, GPT Image 2.5 or Nano Banana 2.1 rows, but resolution does on Nano Banana 2.1, and quality and size do on GPT Image 2.5. A 4:1 banner at 1K on Nano Banana 2.1 is $0.10, the same as a 1:1 at 1K. Choose the ratio first, then the tier, then read the quote, and keep a note of which of the three decisions changed the bill.
Sources
Related posts
More in Developers
- Read back a Sume TTS job to keep the next line in the same voice
A completed Sume text_to_speech job records model_id, voice, language, output_format, generation_config and speed. Reuse them so line two matches line one.
- Worker crashed mid-poll: list Sume jobs and join on idempotency_key
After a restart, GET /v1/jobs?status=processing lists your in-flight video jobs. Each row carries the idempotency_key you sent, so match your records on it.
- Recast prompt limit is 2,000 characters: what to write, what it costs
The H3 Max Recast prompt is optional and capped at 2,000 characters. See what to put in it for a person swap and why the length never changes the price on Sume.
- Refresh 1,200 SKU video ads before Black Friday: waves by plan
On an empty workspace, 1,200 jobs take 300 submission waves on Free, 67 on Pro and 14 on Scale, using Sume's wave_size_hint of 75% of accepted capacity.
Written by Sume