FLUX.1 Kontext

What it is

FLUX.1 Kontext is a family of image models from Black Forest Labs (BFL). Its defining feature is in-context generation and editing: the model accepts text and image inputs together, so a user can supply a picture and an instruction, and the model produces a new or changed image. This differs from a pure text-to-image model, which starts only from a written description.

Origins

BFL announced FLUX.1 Kontext on 29 May 2025, and launched the browser-based BFL Playground at the same time, where people could try the models.^[1] The open-weight variant, Kontext [dev], was first available in a private beta. Its weights were released on 26 June 2025.^[1][2]

The three variants

Kontext [pro] is aimed at fast, iterative editing, where a user makes a series of quick changes to one picture.^[1] Kontext [max] is labelled experimental, and BFL described it as having better prompt adherence and better typography, meaning the model follows the written instructions more closely and renders text in images more accurately.^[1]

Kontext [dev] is the open-weight model. It is a 12B diffusion transformer, a neural network with 12 billion parameters that builds images by gradually removing noise, and it is compatible with the inference code used for the earlier FLUX.1 [dev].^[1]

Capabilities

BFL listed five capabilities: text-to-image generation, character consistency (keeping a person or character recognisable across edits), local editing (changing one part of an image while leaving the rest alone), style reference (applying the look of a reference picture to new content) and multi-turn editing (making successive changes through a conversation).^[1]

Speed and benchmarks

BFL claimed that its text-to-image inference could run up to 8 times faster than leading models such as GPT-Image.^[1] The company also built KontextBench from crowd-sourced, real-world editing tasks. On that benchmark, the [pro] variant scored highest in text editing and character preservation.^[1] These are BFL’s own results, so they should be read as the company’s claims rather than independent verification.

Partners and wider use

Kontext [pro] and [max] launched with a group of partners, including Krea, Freepik, Lightricks, OpenArt, Leonardo, fal, Replicate, Runware, DataCrunch, Together AI and ComfyOrg.^[1] In September 2025, Adobe added FLUX.1 Kontext Pro to the generative fill feature in Photoshop, in beta.^[2]

Limits

BFL itself noted some limits: images can show artifacts after many rounds of editing, and the model has limited world knowledge.^[1]