FLUX.1 Kontext [max] is the top tier of Black Forest Labs' Kontext suite, first released May 29, 2025, built to unify text-to-image generation with instruction-based editing in a single model rather than two separate tools. This multi-image variant extends that further: instead of editing one image against a text prompt, it accepts several reference images at once and composites or edits across all of them in a single pass — the documented example is placing an element from one image onto a subject in another.
Where it shines
Black Forest Labs positions [max] specifically as the tier with improved prompt adherence and typography generation compared to [pro], plus higher consistency for editing — the trade the suite makes for [max] is somewhat more latency in exchange for fidelity. Extended to multiple images, that means more reliable results when a task involves combining a specific object, outfit, or background element from one reference photo into a different scene, or maintaining a character's appearance while placing them in fresh contexts pulled from other references.
Where it falls short
Multi-image compositing is inherently harder to control than single-image editing: with several references in play, ambiguity about which element comes from which image increases, and prompts that don't clearly anchor each source image to a role in the output tend to produce results where the model guesses. Typography and fine text — a stated strength of [max] over [pro] — still degrades on small or dense text the way it does across the whole FLUX.1 lineage. This is also an explicitly experimental configuration of the model, per Black Forest Labs' own framing of the multi-image handling capability, so behavior is less battle-tested than the single-image Kontext endpoints.
Prompting notes
Multi-image prompts work best when each input image is given a clear role in the instruction — which image supplies the subject, which supplies the background or the object being placed — rather than a general instruction that assumes the model will infer intent from image order alone. Character and object consistency, the feature Black Forest Labs highlights for the suite, holds up best across two or three images; asking the model to reconcile many references at once increases the odds of an inconsistent result.
When to pick it
Reach for the [max] multi-image tier over the [pro] multi-image tier when the job specifically needs better prompt adherence or cleaner typography and the extra latency is acceptable — product photography composites, marketing creative that pulls specific elements from multiple source shots, or any edit where "close enough" isn't the bar. For faster, more iterative multi-image editing where you're cycling through variants quickly, the [pro] multi-image tier is the better fit.
Rights and licensing
[max], like [pro], is available only through Black Forest Labs' API and commercial partners — it is not an open-weight release. Black Forest Labs has published an open-weight sibling, FLUX.1 Kontext [dev], under the FLUX.1 Non-Commercial License, but that license does not extend to [max] or [pro], and [dev] does not include the multi-image handling this variant offers.
Alternatives worth comparing
For general single-image editing without the multi-reference compositing need, gpt-image-1 covers generation, editing, and inpainting through one unified endpoint. For teams comparing across ecosystems, gemini-3-pro-image-preview is Google's current top-tier image model and a reasonable point of comparison on general compositional editing quality.
Last technical review: 2026-09-14 — Tokonomix.ai