FLUX.1 Kontext

FLUX.1 Kontext, announced by Black Forest Labs on 29 May 2025, is a suite of models built around a different premise from the rest of the family: the input is not only a prompt but an image, and the model edits in context rather than regenerating from scratch. You can retouch, restyle and iteratively refine an existing picture without finetuning and without a mask-and-layer workflow. The capability that mattered commercially is consistency. Kontext extracts visual concepts from a reference image and keeps a character or a style recognisable across multiple scenes and environments, while applying localised edits to individual elements and preserving the relationships in the rest of the scene. It also handles typography inside images. BFL claims inference up to eight times faster than leading alternatives, and reports that the [pro] variant leads on all six in-context generation tasks it measures, with its strongest results in text editing and character preservation. The line has three members. Kontext [max] is the premium tier, aimed at prompt adherence, typography and rendering quality without giving up speed. Kontext [pro] is the fast iterative editor and the volume workhorse. Kontext [dev] is the open-weights variant — 12 billion parameters, published on GitHub, licensed either under non-commercial terms or a separate self-hosted commercial licence, and designed to deliver proprietary-level editing on consumer hardware. Access runs through the BFL API, the browser-based Playground that shipped with the launch, and self-hosting for the dev weights. In the catalogue Kontext sits beside the numbered generations FLUX.1, FLUX1.1 [pro], FLUX 2 and FLUX 3 — it is a parallel editing branch of the family, not a successor to any of them.

▸Videos

No videos yet.

▸Reviews