How many images can I get per AI image request? Read n, don't guess

It depends on the model, and the catalog tells you. Sume's Image API docs say n accepts 1 to 10 per call at the request level, then add that per-model ceilings are lower and that you should read the n range descriptor from the catalog. The Grok image row is capped at one image. Most other edit models you will use allow up to four.

Stay inside the row's range rather than testing its edge: read it once at start-up and clamp your own requests.

Read it from the catalog

The list endpoint returns each model with a supported_parameters object. Where a model supports several images, that object has an n entry of type range with min and max. A model without an n entry is not advertising multi-image output, so send one image per call and loop.

import os, requests

key = os.environ.get("SUME_API_KEY")
if not key:
    raise SystemExit("set SUME_API_KEY")
r = requests.get(
    "https://api.sume.com/v1/images/models",
    headers={"Authorization": f"Bearer {key}"},
    timeout=30,
)
r.raise_for_status()
for m in r.json()["data"]:
    n = m.get("supported_parameters", {}).get("n")
    print(f"{m['id']:45} {n if n else 'no n descriptor'}")

What the docs fix and what they leave to the catalog

Why not just ask for ten

Large n is one of the settings the docs name as most likely to push a sync request past its 30-second wait and return a 202 job envelope instead of images. If you send n: 4 with high quality and 4K, plan for the envelope and poll the status_url.

Cost scales with images too. A completed generation is billed in full, so a batch you only half need is money you pay anyway. Ask for two, pick one, and ask again if neither works.

A note on other vendors

Do not assume another vendor's count carries over. Google's Gemini image docs (read 2026-10-06) describe their own reference-slot limits per model tier, and those are separate from anything on Sume's rows. The Sume catalog is the only source for what a Sume row accepts.