What is FLUX? Distinguishing the model family, editing tools and open weights
FLUX is a Black Forest Labs model family. Generation, reference editing and inpainting depend on the version. Learn how hosted APIs differ from open weights and why code and model licenses are separate.
Meet this model
FLUX is a model family developed by Black Forest Labs. Common image tasks include text-to-image generation, reference-based editing and specialized inpainting or outpainting. Input requirements and licenses vary by version. FLUX is neither one uniform model nor the exclusive name of a third-party image website.
Choose a model branch for the image task
A text description calls for text-to-image generation; replacing an object in an existing poster requires an editing model. The FLUX.1 repository lists Kontext editing separately from Fill inpainting, showing that these input controls are not interchangeable merely because they share a family name.
The repository also lists structural-conditioning models such as Canny and Depth, using edge or depth information to guide an image. They serve tasks with an existing composition rather than unrestricted generation. Use them when the silhouette or spatial relationships actually need to be preserved.
Read the current catalog separately from older tutorials
The current documentation includes the FLUX 3 family and keeps FLUX.2 supported for production image generation and editing. Keep the full version name; parameters in a FLUX.1 tutorial are not necessarily universal across the family.
For product backgrounds, start with one reference photo and the appearance details that must remain, then compare candidate versions. Intact product edges can matter more than an attractive isolated image. This is a selection method, not a model run or an independent verification of vendor speed claims.
Open code does not give every weight the same license
The FLUX.1 repository code uses Apache 2.0, while its model table specifies separate weight licenses. For example, schnell is listed under Apache 2.0, whereas dev and several related models use the FLUX.1 dev Non-Commercial License. The repository’s code label does not replace the target weights’ terms.
Before local deployment, read the specific model card and license to check the intended use; commercial licensing has separate arrangements. Read output-use terms in their own context. Do not infer either that all outputs from non-commercial weights are forbidden commercially or that every use is unrestricted.
Local inference and hosted services require different preparation
Open weights require a suitable runtime and enough GPU memory, with actual usage depending on the model and implementation. A hosted API delegates inference to the provider, but the called version still matters. A third-party interface labeled FLUX does not establish identical settings or features.
Proofread generated posters when wording is strict and inspect the file for transparency when needed. This article uses official documentation and the inference repository without installing dependencies or running a model. When upgrading, compare the same reference rather than attribute input differences to model improvements.
Checked sources
These sources support the concepts and methods described here. Examples and test scope are labeled separately; check the version of the tools you use.