A family of image models
FLUX is a family of image generation and editing models developed by Black Forest Labs, used for producing images from text, reworking composition or mood from a reference image, and editing part of an existing image.
There is no single FLUX application. You reach the models through Black Forest Labs’ API and Playground, through third-party image services, through supported software, or on your own machine.
Generations and variants
FLUX.1 and FLUX.2 denote generations, and within each there are models differing in speed, quality, editing ability, and where they can run. Names such as pro, flex, max, and klein indicate different capability, pricing, and terms even within the same family.
- Generate a new image from text
- Combine characteristics from several reference images
- Edit part of an image, such as a product color or the background
- Render text naturally within an image
- Run published models locally on supported hardware
Available features differ by model. To reproduce an example from an article, check the exact model name and settings, not just the service.
Compared with Stable Diffusion and ChatGPT Images
Stable Diffusion is similarly a family of models reachable from many applications and local setups. FLUX works the same way, but differs in supported tooling, hardware requirements, and model licensing.
ChatGPT Images and Midjourney are packaged as services with conversation or a dedicated interface, which makes them easier to start with. FLUX offers more freedom in choosing which model to run from where, at the cost of a less obvious entry point.
Licensing and cautions
Not all of FLUX is released under the same free terms. Even where weights are downloadable, commercial use, redistribution, and inclusion in a generation service may carry conditions. Check both the model’s license and the terms of the service you use.
Generated images can contain artifacts in hands, text, and fine detail, and can resemble existing work unintentionally. Inspect at full size before publishing, and consider rights and the risk of misidentification when real people, brands, or copyrighted works are involved.