Sume image models that do not list the 4:5 Instagram ratio
Imagen 4, Grok Image and Higgsfield Soul do not list 4:5. Which Sume ids do, what 4:5 means in pixels, and the nearest ratio to use when a model lacks it.

A 4:5 portrait is the Instagram feed format: 1080 by 1350 pixels. The Sume Image API page is explicit that 4:5 is Instagram portrait and not 4:3, and that Banana Pro sends it natively. The model lists below come from the aspect-ratio catalog in the repository.
From the catalog data in the repo, the ids that do not list 4:5 are Imagen 4 Fast, Imagen 4 Ultra, Grok Image and Higgsfield Soul. Everything else, including both ChatGPT Image 2.5 variants, Seedream, Qwen, FLUX.2, Ideogram and Recraft, lists it.
Which models do not, and what to use
| Sume id | Lists 4:5? | Nearest listed ratio |
|---|---|---|
| google/imagen-4-fast | No | 3:4 |
| google/imagen-4-ultra | No | 3:4 |
| x-ai/grok-image | No | 3:4 |
| higgsfield/soul | No | 3:4 |
| google/nano-banana-2 | Yes | 4:5 |
| openai/gpt-image-2.5 | Yes | 4:5 |
What a 3:4 fallback costs you
3:4 is slightly taller than 4:5. Cropping a 3:4 image to 4:5 trims a little off the top and bottom, so keep the subject out of the outer edge when you must use these models. Check the model's GET /v1/images/models descriptor before you send 4:5, so the miss shows up in testing rather than in a batch.
import os, requests
headers = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}
r = requests.post("https://api.sume.com/v1/images", headers=headers, timeout=60, json={
"model": "google/nano-banana-2",
"prompt": "Feed post, centered product, soft shadow",
"aspect_ratio": "4:5",
"resolution": "1K",
})
print(r.status_code)
Exact 1080 by 1350
On Nano Banana, a 1080x1350 image size becomes aspect_ratio: "4:5" plus a job target_pixels; the model's native output is about 928 by 1152 at 1K and the exact 1080 by 1350 comes from a documented post-step. Check the dimensions you get back if the exact pixel size matters.
How this was checked
Vendor facts come from the pages listed in the sources, read on 2026-10-05. Sume facts come from the Image API docs and the catalog code on main on the same date. Catalogs and limits change, so read the descriptors from GET /v1/images/models before you pin a number in production code.
Sources
Related posts
More in Models
- Summer 2026 video model features mapped to Sume routes, and the gaps
Seedance 2.5, Wan 3.0, Omni 1.1 Flash and Ray3.2 launched features this summer. Which map to a Sume route, which are vendor-only. Read 2026-10-05.
- Talking video, face swap, Recast or motion control: pick by your input
Four Sume routes make a person on video. Which one fits depends on whether you have a script, a source video, a photo of each person or a motion clip.
- Tavus Human Interaction Model: what Griffin is, and what ships today
Tavus calls Griffin a Human Interaction Model. Only select testers have Griffin-Lite. Here is what it is, and what ships today on Sume for a talking presenter.
- Thai, Vietnamese, Indonesian voiceover: on MAI's list, not Sume's tags
MAI-Voice-2.1 lists th-TH, vi-VN and id-ID; Sume's 16 voice tags do not. What that means, a 1-cent audition, and the Unicode trap in Vietnamese.
Written by Sume