FLUX (image generation model)

FLUX is a family of text-to-image generative AI models developed by Black Forest Labs, a company founded in 2024 by researchers formerly associated with Stability AI.^[1] The models generate photorealistic images from text prompts and are known for their strong text rendering, prompt adherence, and support for image editing through reference images.^[2] FLUX models are built on a rectified flow transformer architecture and are offered in both open-weight and proprietary commercial variants, making them a widely used alternative to systems such as Stable Diffusion, Midjourney, and DALL-E.^[1][4]

History

Black Forest Labs was founded in 2024 by Robin Rombach and other researchers who had previously worked on the Stable Diffusion series at Stability AI.^[1] The company released the first FLUX.1 models in August 2024 in three variants: FLUX.1 Schnell, an Apache 2.0 licensed open-weight model optimised for speed; FLUX.1 Dev, a higher-quality open-weight model released under a non-commercial licence; and FLUX.1 Pro, a commercial model available through an API.^[1] The open-weight releases were widely adopted by the open-source community and quickly became the default choice for local image generation.

In November 2025 the company released the FLUX.2 family, anchored by FLUX.2 Dev, a 32-billion-parameter model released under the FLUX.2-dev non-commercial licence, capable of both text-to-image generation and multi-reference image editing.^[2][4] FLUX.2 introduced multi-reference support combining up to ten images into a novel output, output resolution up to 4 megapixels, pose control, and substantially improved typography.^[3][5] In January 2026 the company released the FLUX.2 Klein family, its fastest models, capable of sub-second generation on consumer GPUs.^[2]

Key Concepts

FLUX models are rectified flow transformers: they use the flow matching framework, in which a neural network learns a vector field that transports noise to data along nearly straight paths, allowing high-quality generation in a small number of sampling steps.^[1][4] This architecture descends directly from the research behind Stable Diffusion 3 and is shared with other frontier generative systems.

A distinctive feature of the family is the range of variants optimised for different trade-offs. The FLUX.2 lineup spans Klein (fast, sub-second inference for prototyping), Flex (fine-grained control with a configurable number of steps), Pro (production at scale), and Max (highest quality, with web search grounding for real-time information).^[5] The models are released with responsible-development safeguards; Black Forest Labs states that prior to release it evaluated checkpoints and hosted services for risks including unlawful content, and implemented pre-release mitigations.^[4]

Applications and Impact

FLUX is used for creative and commercial image generation, including marketing assets, product visualisation, game art, and design mockups, with its reliable text rendering making it well suited for infographics, user interfaces, and multilingual content.^[3] Open-weight variants have spawned a large ecosystem of fine-tunes, LoRA adapters, and third-party hosting platforms, and the models are available on major model-serving services and through NVIDIA-optimised deployments that support FP8 quantisation for reduced memory use.^[6] The FLUX.2 release was noted for setting a new standard among open-weight image models across text-to-image, single-reference editing, and multi-reference editing benchmarks.^[5]

>See Also

• Stable Diffusion
• Diffusion Model
• Flow Matching
• Midjourney
• DALL-E

References