Black Forest Labs unveils FLUX.2 image generation models with NVIDIA optimisations

Black Forest Labs has announced the release of FLUX.2, its latest family of visual generative AI models designed to push the boundaries of image fidelity. Developed in collaboration with NVIDIA and ComfyUI, the models launch with FP8 quantisation and RTX GPU performance optimisations, reducing VRAM requirements by up to 40 percent.

FLUX.2 is built for photorealism, producing images at resolutions up to 4 megapixels with realistic lighting and physics to avoid the artificial “AI look”. The models introduce direct pose control, allowing users to specify subject positioning, and generate clean, legible text across infographics, user interfaces, and multilingual content.

Despite their capabilities, the models are resource-intensive, requiring up to 90GB VRAM to fully load. To make them more accessible, NVIDIA and Black Forest Labs have streamlined performance for GeForce RTX GPUs. Improvements to ComfyUI’s weight streaming feature further enhance usability, enabling broader adoption among PC users.

FLUX.2 is available now through ComfyUI templates and on Black Forest Labs’ Hugging Face page. For more information on the Nvidia-specific optimisations and what sort of results you can expect from this, Nvidia has released a dedicated AI Garage blog post covering Flux.2 extensively.

KitGuru Says: GenAI is advancing very quickly. When you look at the amount of money being pumped into AI companies right now, it is easy to see why. 

Become a Patron!