FLUX is a family of Text-to-image generation diffusion models released in August 2024 by Black Forest Labs, a company founded by researchers who had previously created Stable Diffusion and related latent diffusion research, including Robin Rombach. Black Forest Labs was formed in 2024 after its founders left Stability AI, and FLUX was the company's first major public release.
Model variants
FLUX was released in multiple variants aimed at different use cases and licensing needs: FLUX.1 [pro], a closed, higher-performance model accessible via API; FLUX.1 [dev], an open-weights model available for non-commercial use and research; and FLUX.1 [schnell], a distilled, faster open-weights model released under a permissive Apache 2.0 license suitable for commercial use. This tiered structure let Black Forest Labs offer both a fully open, commercially usable option and a higher-quality gated option, a middle path between the fully closed approach of Midjourney and the fully open original release of Stable Diffusion.
Technical approach
FLUX uses a diffusion transformer architecture, incorporating flow-matching training techniques that had shown efficiency advantages over earlier denoising diffusion approaches in contemporary research, and scaled up model size substantially relative to earlier open diffusion models. On release, FLUX models were widely reported to outperform existing open alternatives, including various versions of Stable Diffusion, on measures of prompt adherence, image detail, and rendering of text within images, an area that had historically been a weakness of diffusion-based image generators.
Adoption and integration
Because FLUX's open-weights variants could be downloaded and run independently, the model was rapidly adopted by the same community ecosystem that had grown up around Stable Diffusion, including support for LoRA fine-tuning and ControlNet-style conditioning tools, and it became a common base model on platforms like Hugging Face and community fine-tuning hubs. FLUX also gained a significant commercial partnership when xAI integrated FLUX as the image-generation backend for its Grok chatbot, giving the model a large-scale consumer deployment beyond its open-source developer community.
Reception
FLUX was widely regarded by reviewers and the open-source AI community as setting a new state of the art among open or openly licensed image-generation models at the time of its release, closing much of the gap that had previously separated open models from leading closed systems such as DALL-E and Midjourney. Its release illustrated a broader pattern in generative AI in which specialized teams, often composed of researchers who had previously built a first-generation breakthrough model at one company, left to found focused startups building the next generation of the same technology, echoing dynamics also seen among departures from OpenAI and other frontier labs. Black Forest Labs positioned FLUX's continued open releases as central to its identity and business strategy, aiming to sustain an open ecosystem for image generation even as much of the wider generative AI industry, including video generation efforts from Google's Veo, trended toward closed, API-only products.