FLUX 3 Image Klein: 5.6B Open Weights Model for ComfyUI

FLUX 3 Image Klein

Text-to-ImageImage EditingMulti-ReferenceOpen Weights

Black Forest Labs' first open weights model in the FLUX 3 family: a 5.6B parameter image generator for text-to-image, image editing and multi-reference generation with up to 10 reference images. Ships as a base checkpoint for fine-tuning and a step-distilled checkpoint for low-latency generation, in BF16, FP8, INT8 and NVFP4 builds, with core ComfyUI support.

Blog Model Page HF (distilled)HF (base)HF (ComfyUI)

Developer	Black Forest Labs
Release Date	2026-10-15
Architecture	5.6B flow matching transformer (per-mode early blocks + 28 joint blocks), Qwen3-VL-4B text encoder, FLUX.2 VAE
License	FLUX Kommunity License
Capabilities	Native 4K text-to-image, pixel-perfect editing, bounding-box control, up to 10 reference images, optional prompt upsampler
Variants	Base for fine-tuning · step-distilled for speed · BF16 / FP8 / INT8 / NVFP4

FLUX 3 Image Klein shares its core capabilities with the full FLUX 3 Image model: native 4K output, editing that leaves the rest of the frame untouched, and bounding boxes that control where elements go. The distilled checkpoint is the practical entry point, since it is the same size as the base model but runs in a fraction of the steps.

A snow leopard on a rocky ledge, generated by FLUX 3 Image Klein. Sample by Black Forest Labs.

ComfyUI support merged in PR #17064 on release day, together with four official templates (text to image and image edit, for both the base and the distilled checkpoint). Weights go to models/diffusion_models and models/text_encoders, and the VAE is the existing FLUX.2 VAE. See the news article for the file list, the benchmarks and more samples.