FLUX.1 Kontext pro vs FLUX.2 for edits: what to use on Sume

For a new edit workflow, use FLUX.2. BFL's Kontext page calls FLUX.1 Kontext [pro] a previous-generation model and says FLUX.2 is now its recommended model for generation and editing. On Sume, send black-forest-labs/flux.2-pro with input_references, and I found no Kontext id in the Sume image catalog.

Vendor statements are from the BFL page, read 2026-10-01. Sume rules are from the Image API docs.

What does BFL recommend?

The page leads with a note that FLUX.2 offers multi-reference support for up to 10 images, improved text editing and output up to 4MP. It describes Kontext [pro] as combining text-to-image and image editing, with prices of $0.04 per image for [pro] and $0.08 for [max] on that page.

How do references work on Sume?

input_references is an optional array of reference images for image-to-image. Reference URLs must be public HTTPS; localhost, private-network and non-HTTPS URLs are rejected before submission. Models whose input_references descriptor is min 0 and max 0 are text-to-image only and reject references, so read the descriptor on the catalog row before sending.

For a worked request, see FLUX.2 multi-reference editing.

Is the reference cap the same on Sume?

Do not assume it. The docs say per-model ceilings are published as capability descriptors on the catalog, and the examples there show a max of 10 for some rows. Read the input_references max for the exact model you call.

What should I do with an existing Kontext prompt?

Keep the instruction text, move your source images into input_references, and switch the model id to black-forest-labs/flux.2-pro. Test a handful of edits before cutting over, because prompts tuned for one model may need rewording on another.