FLUX 3 Image combines 10 references: Sume's limits per model

BFL's FLUX 3 Image documentation says you can "combine up to ten references" in one request, and it describes edits by bounding box on a 0 to 1000 grid in [y0, x0, y1, x1] order. If you are planning an image-to-image workflow around that, the practical question is how many references each model you can actually call accepts.

Sume does not list a FLUX 3 id. The FLUX family in the Sume image catalog is black-forest-labs/flux.2-pro and black-forest-labs/flux.2-flex. The Sume docs describe input_references as an array of image_url entries and say the catalog is the only source of truth for the ceiling.

Reference ceilings in the Sume catalog

In the code that builds the catalog descriptors, the shared ceiling is 10 references for any model that can edit. Two models override it: both ChatGPT Image 2.5 variants take 16, and Ideogram 4.5 takes 5 (one source image to edit plus four references). Models that cannot edit report a range of 0 to 0.

Read it live instead of trusting a table

Descriptors can change between releases, so read them before you pin a number in your code. This call prints the ceiling for every model:

import os, requests
headers = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}
r = requests.get("https://api.sume.com/v1/images/models", headers=headers, timeout=30)
r.raise_for_status()
for m in r.json()["data"]:
    ref = m["supported_parameters"].get("input_references", {})
    print(m["id"], ref.get("max", 0))

What this means for a FLUX 3 plan

A ten-reference FLUX 3 workflow maps to FLUX.2 Pro, Nano Banana 2 or any other default-ceiling model on Sume without trimming. If you need more than ten, ChatGPT Image 2.5 takes 16. A request over the ceiling is rejected with a 400 that names the maximum, so you find out before you pay for a generation. The Sume Image API page lists the request shape.

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.