FLUX.3 Image Klein arrives in Diffusers

Black Forest Labs says Klein supports text-to-image generation, image editing and multi-reference generation using as many as 10 reference images.[2] Hugging Face describes an added layout-control feature: prompts can use bounding boxes to specify where image elements should appear or where content from a reference image should move.[1] Black Forest Labs also says Klein shares capabilities with the full FLUX 3 Image model, including native 4K generation, pixel-perfect editing and bounding-box controls.[2] These capabilities make the model relevant to teams evaluating both new-image creation and image-editing workflows.

The FLUX.3 Image Klein collection contains two open checkpoints.[1] Hugging Face says the distilled checkpoint is time-step distilled for fast inference, while the base checkpoint runs 50 steps with classifier-free guidance and is the version to fine-tune.[1] Black Forest Labs describes the release in similar terms: a base model for fine-tuning and a step-distilled model for low-latency generation.[2] Hugging Face also says DreamBooth LoRA training scripts for both tasks ship in Diffusers.[1] Teams should choose between the checkpoints based on whether their immediate priority is inference speed or fine-tuning.

Hugging Face reports that the distilled checkpoint generates a 1024-by-1024 image in about two seconds on an H100.[1] Black Forest Labs says the generator weights require 11.2 GB of memory in BF16 or 5.6 GB in FP8, and that its INT8 build runs on NVIDIA GeForce RTX GPUs with as little as 8 GB of VRAM through ComfyUI or Diffusers.[2] For the distilled INT8 model, the company reports generation times of about four seconds for a 1-megapixel image on an RTX 5090 and under nine seconds on an RTX 4070 through ComfyUI.[2] These figures can help teams scope hardware tests, but they describe specific configurations.

Black Forest Labs lists API prices of $0.008 per image at 768×768, $0.009 at 1K, $0.011 at 1.5K, $0.014 at 2K and $0.036 at 4K; it says text-to-image and editing cost the same at each resolution.[3] The weights use the FLUX Kommunity License, which the company says permits free commercial use for individuals and companies with annual revenue under $5 million; larger organizations can use commercial self-hosted licensing options or its API.[2] Day-one integrations include ComfyUI, Diffusers, TensorRT-LLM, SGLang Diffusion and vLLM-Omni.[2] Black Forest Labs says it used dataset filters and targeted fine-tuning to help suppress harmful content and behavior.[2]