FLUX is Black Forest Labs' suite of open-weight image generation and editing models, distributed through a compact official inference repository with over 25,000 GitHub stars. Built by some of the original creators of latent diffusion, FLUX has become a default starting point for developers who want state-of-the-art text-to-image quality they can run and fine-tune themselves rather than calling a closed API.

A Full Family of Open-Weight Models

Rather than a single checkpoint, FLUX ships as a family covering different needs. FLUX.1 [schnell] is released under Apache-2.0 for fast, permissively licensed text-to-image generation, while FLUX.1 [dev] targets higher quality under a non-commercial license. The lineup extends to specialized variants for in/out-painting (Fill), structural conditioning (Canny and Depth, with LoRA options), image variation (Redux), and a Krea-tuned aesthetic model — so teams can pick the exact capability they need.

Editing with FLUX.1 Kontext

FLUX.1 Kontext brings instruction-based image editing into the open-weight lineup, letting users transform existing images from natural-language prompts — for example, replacing a logo with specified text. This moves FLUX beyond pure generation toward a practical editing tool, narrowing the gap with proprietary editing systems while keeping weights downloadable for local experimentation.

Minimal, Hackable Inference Code

The repository deliberately keeps the inference code minimal, making it easy to read, modify, and integrate. Installation is a standard Python virtual-environment setup, with optional NVIDIA TensorRT support for accelerated inference. The autoencoder weights are released under Apache-2.0, and each model links to its HuggingFace repo and license, making the licensing terms explicit for every component.

Ecosystem and Commercial Path

FLUX is widely supported across the open generative-AI ecosystem, including tools like ComfyUI, which makes it straightforward to slot into existing pipelines. Black Forest Labs also offers an API for its full model range, including Pro-tier non-open weights, plus a commercial licensing program with built-in usage tracking for teams that deploy the open models in production.

Considerations

Licensing is the main nuance: only [schnell] and the autoencoder are Apache-2.0, while [dev], Kontext, and the Tools models carry a non-commercial license that requires a paid arrangement for commercial use. High-resolution generation also demands capable GPUs. For developers who want top-tier open image generation and editing with a clean, auditable codebase, though, FLUX remains one of the most influential releases in the space.