Back to Model
F

FLUX 3 Image Klein

Model Oct 15, 2026

Black Forest Labs' downloadable image-generation and editing model, with a 5.6B image transformer, multiple references and prompt-based layout boxes.

Visit primary link 4 sources attached
Hugging Face's generated knitted knight and dragon scene, with red boxes marking the knight's source and destination positions.
Hugging Face's published FLUX 3 layout-editing example. This is an author-generated sample with annotated boxes, not an AITrending reproduction or a guarantee of unchanged pixels. Source: Hugging Face Diffusers team
multimodalself-hosted

What happened lately

FLUX 3 Image Klein timeline

Black Forest Labs released FLUX 3 Image Klein on October 15, 2026. The image transformer has 5.6B parameters; its Qwen3-VL-4B-Instruct text encoder is additional. The model accepts a text prompt and up to ten reference images. A base checkpoint supports fine-tuning, while the distilled checkpoint targets generation and editing with a default 16-step path.

Creators can describe positions using normalized boxes and move elements between source and target boxes. Those boxes guide local generation rather than enforcing a hard pixel mask. The hosted API’s pixel-restoration behavior is a separate feature. Diffusers provides modular inference and LoRA training workflows. The weights are subject to the FLUX Kommunity License, including its revenue-based commercial-output exception; they are not an unrestricted permissive-license release.