FLUX.2 [max] is a flagship image generation model developed by Black Forest Labs, released in December 2025 as the most capable variant in the second-generation FLUX suite. Built on a 32-billion parameter rectified flow matching transformer architecture, the model integrates a Mistral-3 24B vision-language model (VLM) to achieve advanced contextual understanding and spatial reasoning. It is designed for professional creative workflows requiring high fidelity and precise adherence to complex instructions.
Grounded Generation and Multi-Reference Control
A defining feature of the model is grounded generation, which allows it to perform real-time web searches to incorporate current events, trending products, and up-to-date visual information into its outputs. Additionally, the model supports multi-reference conditioning, enabling users to provide up to 10 reference images simultaneously. This capability ensures visual consistency for characters, products, and styles across multiple generations without the need for manual fine-tuning or LoRA training.
Professional Output and Control
The model generates high-resolution images up to 4 megapixels (4MP) natively, featuring enhanced texture synthesis for materials such as fabric, wood, and skin. It provides superior typography performance, capable of rendering complex text, infographics, and user interface mockups with high legibility. For granular control, FLUX.2 [max] supports JSON-structured prompting, allowing for precise specification of scene elements, and hex-code steering for exact color matching across outputs.