FLUX.2 is Black Forest Labs' new family of models for AI image generation and editing. It reaches 4-megapixel resolutions, maintains character and style consistency, and is designed for professional production.
Black Forest Labs has announced the launch of FLUX.2, its new generation of artificial intelligence models designed for professional creative workflows. The FLUX.2 family unifies image generation and editing in a single model, with four variants ranging from open source to commercial production services. According to the company, the model significantly reduces generation costs and times compared to previous alternatives.
The model allows working with up to 10 reference images simultaneously to maintain coherence in characters, products, or visual styles. The maximum resolution reaches 4 megapixels, aimed at projects requiring high definition such as product photography or architectural visualization.
Technical improvements include higher levels of detail in generated images, sharper textures, and more stable lighting. Complex typography rendering, infographics, and interface mockups work more reliably than in previous versions. The model also incorporates better tracking of structured instructions with multiple elements and compositional constraints, along with more grounded knowledge about real-world lighting and spatial logic in generated scenes.
FLUX.2 is offered in four versions. FLUX.2 Pro is the highest performance option, available via API for professional production. FLUX.2 Flex allows adjusting parameters such as the number of processing steps to balance quality and generation speed. FLUX.2 Dev is a 32-billion parameter model available on Hugging Face for local execution. FLUX.2 Klein, still in beta, will be an open-source version under Apache 2.0 license.
The technical architecture combines the Mistral-3 vision-language model with 24 billion parameters with a rectified flow transformer. According to Black Forest Labs, this combination provides contextual knowledge of the real world while processing spatial relationships and material properties. All models in the family use FLUX.2 VAE, a new image representation component available under Apache 2.0 license.
FLUX.2 Dev is available through platforms such as FAL, Replicate, Runware, TogetherAI, Cloudflare, and DeepInfra. Black Forest Labs maintains its "open core" strategy, which combines downloadable and locally executable models for the developer community with commercial services for teams requiring scalability.
Black Forest Labs develops artificial intelligence models for image and video generation based on advanced research. The company implements accessible technologies with a focus on efficiency and ...
Flux is an AI model for creating and editing photorealistic high-resolution images. It maintains visual consistency through multiple simultaneous references and generates legible text with complex ...
16/07/2026
SpaceXAI has introduced Grok 4.5, its most advanced model to date, trained alongside Cursor and built for coding, agentic tasks and knowledge work, ...
15/07/2026
Thinking Machines Lab launches Inkling, its first open-weight AI model: it natively processes text, images and audio, and is designed for other ...
13/07/2026
The Stanford Digital Economy Lab has presented the manifesto We Must Act Now, backed by 16 Nobel laureates and more than 200 economists and ...
09/07/2026
The GPT-5.6 family, with the Sol, Terra and Luna models, is now available in ChatGPT, Codex and the API after its limited preview, with better ...