FLUX.1 Kontext [pro] was Black Forest Labs' first entry in the Kontext suite, released May 29, 2025 as, in the maker's own words, "a pioneer for fast, iterative image editing." Where the base Kontext [pro] endpoint edits one image against a text instruction, this multi-image variant accepts several reference images at once — combining an element from one photo with the scene of another, for instance putting an object from image A into image B — in a single generation pass rather than a chain of separate edits.
Where it shines
Speed is the defining trait carried over from the single-image endpoint: Black Forest Labs built Kontext [pro] for rapid, iterative workflows where a user (or an automated pipeline) is cycling through variations quickly rather than waiting on one expensive high-fidelity pass. Extended to multiple images, that makes it a reasonable fit for workflows that need to try several different source-image combinations fast — swapping which reference supplies the background, the subject, or an accessory — before committing to a final composite. Character and object consistency across scenes is a suite-wide strength Black Forest Labs highlights: a reference face or object tends to stay recognizable across edits rather than drifting.
Where it falls short
This is the faster tier, not the highest-fidelity one — Black Forest Labs positions [max] above it specifically for stronger prompt adherence and typography, so dense on-image text and precise instruction-following are more likely to need a second pass or a move up to [max]. As with any multi-image compositing task, ambiguity about which reference supplies which element is the main failure mode: a prompt that doesn't explicitly assign a role to each input image leaves the model guessing, and with several images in play that guess is more likely to be wrong than in a single-image edit. This is also an explicitly experimental multi-image configuration per Black Forest Labs' own description, so treat outputs as needing review rather than production-final on the first try.
Prompting notes
Name each source image's role directly in the prompt rather than relying on upload order — "use the outfit from the first image on the person in the second image" performs more reliably than an instruction that assumes the model infers which image is which. Because the tier is built for iteration, short prompt-and-regenerate cycles tend to converge faster here than trying to get one long, fully-specified prompt exactly right on the first attempt.
When to pick it
Choose the [pro] multi-image tier when the job is iterative — compositing tests, quick creative exploration across multiple reference shots, high-volume workflows where a slightly lower fidelity ceiling is an acceptable trade for speed. Move up to the [max] multi-image tier when a specific output needs to be right the first time, especially where on-image text or tight prompt adherence matters.
Rights and licensing
[pro] is available only through Black Forest Labs' API and commercial partners; it is not an open-weight model. The open-weight sibling in the family, FLUX.1 Kontext [dev], ships under the FLUX.1 Non-Commercial License and does not include this multi-image handling capability.
Alternatives worth comparing
For unified single-image generation and editing without a multi-reference need, gpt-image-1 offers generation, inpainting, and editing from one endpoint. gemini-2.5-flash-image is a comparably fast, iteration-friendly alternative from a different vendor for teams weighing ecosystem fit.
Last technical review: 2026-09-14 — Tokonomix.ai