FLUX is a family of AI image-generation models — and, in its newest version, video and motion-prediction models too — built by the German lab Black Forest Labs. Type a description into one of its tools and it turns that text into a picture, an edited version of an existing image, or a short video clip. It’s one of the most widely used engines behind AI-generated visuals online, even when the product someone is using doesn’t put the FLUX name on the label.
What FLUX actually is
FLUX is a diffusion model: it starts from random visual noise and, guided by a text prompt, gradually refines that noise into a finished image over a series of steps. Specifically, it uses a ‘rectified flow transformer’ architecture scaled to billions of parameters — a design choice that helps FLUX hold detail at high resolution and follow long, specific prompts more reliably than earlier AI image generation tools.
Who builds it
Black Forest Labs was founded in 2024 by Robin Rombach, Andreas Blattmann, and Patrick Esser — the researchers who built Stable Diffusion at Stability AI before striking out on their own. They set up in Freiburg im Breisgau, Germany — the ‘black forest’ of the name — and shipped the first FLUX models within months of founding the company. Backed by investors including Andreessen Horowitz, the lab raised a $300 million Series B round in late 2025 at a $3.25 billion valuation, and has licensed FLUX to companies including Meta and xAI, whose Grok chatbot uses it to generate images.
The FLUX lineup
FLUX isn’t one model but a tiered family, and the tier matters because it determines both quality and how you’re allowed to use it:
- Schnell (German for ‘fast’) and its newer counterpart Klein are small, quick, open-weight models released under an Apache 2.0 license — free to download and run, including commercially.
- Dev models are also downloadable, but under a non-commercial license aimed at hobbyists and researchers.
- Pro, Max, and Flex are the highest-quality models, available only through Black Forest Labs’ paid API — there’s no version of these you can download and run yourself.
- Kontext is built for editing: you feed it an existing image plus an instruction (change the background to a beach, for example) instead of generating from a blank page.
The newest release, FLUX 3, launched in early access in July 2026 and is the first to extend beyond still images — see ‘In the news’ below.
How to actually use it
There are three practical routes in:
- The Playground, a browser tool at playground.bfl.ai where anyone can generate and edit images with no coding, paying per image as they go.
- The API, for developers who want to build FLUX into their own app — Adobe Photoshop and Mistral’s Le Chat both license FLUX rather than building their own image model.
- Running it yourself, for the open-weight Schnell, Klein, and Dev models — download the weights from Hugging Face and run them locally if you have a capable GPU, with no per-image fee and no internet connection required.
There’s no free tier for the hosted Playground or API: pricing is pay-per-image, from roughly $0.014 for the cheapest FLUX.2 Klein image up to $0.08 for a FLUX.1 Kontext Max edit, as of July 2026 per Black Forest Labs’ pricing. Cost scales with model quality and output resolution.
How it compares
FLUX competes most directly with Midjourney, but the two suit different kinds of user. Midjourney is a closed, subscription-only service reachable through its own app and website. FLUX is licensed out — you’ll often encounter it running quietly inside someone else’s product, alongside the option to use it directly through the Playground or download parts of it and run them for free.
In the news
Black Forest Labs pushed FLUX in a new direction this month: FLUX 3 launched in early access as the company’s first model trained to generate images, video with synchronized audio, and predicted robot motion within a single system, instead of stitching together separate tools for each.