Which Sume image models cannot edit a photo? Five reject references

This fall's image launches all sell editing. Midjourney's update feed says its V8.2 edit model can edit with text instructions, take up to 4 image references and inpaint or outpaint (Midjourney updates feed, read 2026-10-04). OpenAI says ChatGPT Images 2.5, released September 8, 2026, lets users upload photos for editing (OpenAI Developer Community, read 2026-10-04). Before you promise a customer an edit button, check that the model you plan to call accepts a reference at all.

What the Sume catalog says

The Image API docs say that a model whose input_references descriptor is {"min": 0, "max": 0} is text-to-image only and rejects references. In the Sume catalog source, five rows have image editing switched off. The other fourteen accept references.

List them yourself before you build

Do not hard-code this table. The catalog is the allowlist, so read it at start-up. The script below prints every model that does not advertise a reference slot.

import os
import requests

r = requests.get(
    "https://api.sume.com/v1/images/models",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    timeout=30,
)
r.raise_for_status()
for m in r.json()["data"]:
    refs = m["supported_parameters"].get("input_references")
    if not refs or refs.get("max", 0) == 0:
        print(m["id"], "is text-to-image only")

What to do with the answer

Reference URLs must be public HTTPS. For the other limits (16 references on ChatGPT Image 2.5, 5 on Ideogram 4.5), read the per-model descriptor, not a blog table, because rows change when the catalog does.

• Show the edit control only for models that list a reference slot.
• If a user picks a text-only model, route them to an editing model instead of sending a reference and waiting for a rejection.
• If you do not need to choose, omit the model on the MCP generate_image tool; the docs say it then routes to sume/auto.