Black Forest Labs releases FLUX 3 Image Klein with layout control
Black Forest Labs published FLUX 3 Image Klein on 15 October 2026: 5.6B parameters, editing with up to 10 references, box-based layout control, an 8 GB INT8 build and a revenue-capped license.
Black Forest Labs released FLUX 3 Image Klein on 15 October 2026. It is a 5.6-billion-parameter diffusion model. That count covers the image transformer only, because the Qwen3-VL-4B-Instruct text encoder is separate.
What is new
- One model, several jobs. It handles text-to-image, instruction-based editing and composition from up to 10 reference images.
- Layout control. You place bounding boxes on a 0 to 1000 grid in the prompt, then move, recolor, replace or remove the elements inside them.
- Output. It renders text inside images and produces up to 4K.
- Two open checkpoints. The distilled Klein runs in 16 steps without classifier-free guidance and is the one to self-host. The undistilled Klein Base takes 50 steps and is meant for fine-tuning, LoRA training and research.
- Prompt upsampler. The model was trained on long, detailed captions, so a short prompt sent as typed gives generic results. Black Forest Labs provides an upsampler, a 9B vision-language model fine-tuned from Qwen3.5-9B under Apache 2.0, and the reference code, Diffusers and the ComfyUI workflows enable it by default.
Memory and speed
The BF16 generator weights are about 11 GB and the FP8 ones about 5.6 GB, and an INT8 build runs on NVIDIA GPUs with as little as 8 GB of VRAM through ComfyUI or Diffusers. NVFP4 builds need a Blackwell GPU. The pipeline also needs the separate Qwen3-VL-4B text encoder, about 9 GB in BF16. Hugging Face’s Diffusers team measured a peak of 20.7 GiB, about 22 GB, with BF16 weights and no CPU offloading. Black Forest Labs says a 1-megapixel image takes about 4 seconds on an RTX 5090 and under 9 seconds on an RTX 4070. Day-one support covers ComfyUI, Diffusers, SGLang Diffusion and vLLM-Omni. Running image and video models locally explains how to set one up.
In its own internal evaluations Black Forest Labs ranks the model first among open image models for text-to-image, multi-reference generation and editing. Those results are the developer’s and have not been independently confirmed.
License
The weights use the FLUX Kommunity License: free for non-commercial use, and free for commercial use only if you and your affiliates have annual revenue under US$5 million. Larger organizations need a separate commercial license or the paid API. Commercial users must also filter or review outputs and add AI disclosures where the law requires them. The download is gated on Hugging Face: you accept the license and are approved automatically. See how it compares with Qwen-Image-2.1, which is research-only.
Sources
- Black Forest Labs: FLUX 3 Image Klein, open weights, 5.6B parameters (15 Oct 2026) bfl.ai
- FLUX 3 Image Klein model card (Hugging Face) huggingface.co
- Reference inference code (GitHub) github.com
- BFL release notes docs.bfl.ml
- Hugging Face blog: FLUX.3 Image Klein in Diffusers (memory measurements) huggingface.co
Models in this story
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| FLUX 3 Image Klein Black Forest Labs | 5.6B | – | FLUX Kommunity (Conditional) | 8 GB Laptop or phone |