Black Forest Labs releases FLUX.3 Image Klein open weights, quantized builds and API
Article timeline
The articles in this story. Times are ET.
Stable Diffusion Tutorials@SD_TutorialXBlack Forest Labs releases open weights for Flux 3 Image KleinAIBlack Forest Labs has released open weights for Flux 3 Image Klein, a 5.6B-parameter model, according to a post from Stable Diffusion Tutorials. The model supports text-to-image generation and image editing, supports finetuning, and comes in two variants, Klein and Klein base, with Bf16, NVFP4 and FP8 formats. The weights are linked on Hugging Face.
vLLM@vllm_projectOfficialvLLM-Omni adds day-0 support for FLUX 3 Image KleinAIvLLM-Omni now supports FLUX 3 Image Klein image generation from the day of its release, developed with Black Forest Labs and NVIDIA. The integration supports BF16, FP8 and NVFP4 checkpoint formats, with custom kernels optimizing the transformer and text encoder. A Docker image tagged flux-3-image-klein and a deployment recipe are available.
SGLang@sgl_projectOfficialSGLang publishes a cookbook guide for FLUX.3 diffusion modelsAISGLang shares a cookbook link for running FLUX.3 with its diffusion serving framework. The post gives no further details on the guide's contents, setup steps, or performance figures.
- SGLang@sgl_projectOfficialSGLang adds Day-0 support for FLUX 3 Image Klein image model
AISGLang has Day-0 support for FLUX 3 Image Klein from Black Forest Labs, a 5.6B-parameter image model for text-to-image, editing, and multi-reference generation. The model accepts up to 10 reference images per generation and bounding-box layout control, and it is distilled to 16 steps with BF16, FP8, and NVFP4 variants.
Black Forest Labs@bfl_aiOfficialBlack Forest Labs releases FLUX.3 Image Klein weights on Hugging FaceAIBlack Forest Labs says the FLUX.3 Image Klein weights are available for download on Hugging Face. The post links to a model page and a blog post for further details.
- Black Forest Labs@bfl_aiOfficialPickBlack Forest Labs releases open weights, quantized versions, and an API for its image model
AIBlack Forest Labs announces base and distilled open weights for its image model. It also releases FP8, NVFP4, and INT8 quantized versions that run on GPUs and laptops with 8 GB of VRAM. The distilled model is available through the BFL API with native 4K output, bounding boxes, and pixel-perfect editing. The company also lists launch support for AI Toolkit, ComfyUI, Diffusers, SGLang Diffusion, vLLM-Omni, and Nvidia fine-tuning recipes.