--- title: "Welcome FLUX.3 Image Klein in Diffusers" description: "The Black Forest Labs FLUX.3 Image Klein model is now available in Diffusers, combining text-to-image, multi-reference image-to-image, and bounding-box layout control into one pipeline. Two open checkpoints are offered: a distilled version using 16 inference steps without classifier-free guidance for fast generation, and a base version requiring 50 steps with guidance. The distilled checkpoint produces a 1024x1024 image in about two seconds on an H100, and quantized NVFP4 and FP8 weights can be loaded via from_single_file to reduce memory. The model is built as a Modular Diffusers pipeline, enabling users to remove, replace, or insert custom components, such as a block that captures intermediate clean-image previews. LoRA fine-tuning scripts for both text-to-image and image editing are included, along with a prompting guide for optimal scene descriptions." --- # Welcome FLUX.3 Image Klein in Diffusers [Hugging Face](https://yomu.fyi/company/hugging-face) · Aritra Roy Gosthipaty, Sayak Paul, Linoy Tsaban · Oct 15, 2026 **Type:** Announcement ## Summary The Black Forest Labs FLUX.3 Image Klein model is now available in Diffusers, combining text-to-image, multi-reference image-to-image, and bounding-box layout control into one pipeline. Two open checkpoints are offered: a distilled version using 16 inference steps without classifier-free guidance for fast generation, and a base version requiring 50 steps with guidance. The distilled checkpoint produces a 1024x1024 image in about two seconds on an H100, and quantized NVFP4 and FP8 weights can be loaded via from\_single\_file to reduce memory. The model is built as a Modular Diffusers pipeline, enabling users to remove, replace, or insert custom components, such as a block that captures intermediate clean-image previews. LoRA fine-tuning scripts for both text-to-image and image editing are included, along with a prompting guide for optimal scene descriptions. ## Approach / What changed Black Forest Labs released two open checkpoints of FLUX.3 Image Klein: a distilled version (16 steps, no CFG) and a base version (50 steps, CFG). The model supports text-to-image, multi-reference image-to-image via tokens, and layout control through JSON bounding-box regions in the prompt. Quantized NVFP4 and FP8 single-file weights are available for reduced memory usage. The pipeline uses the Modular Diffusers system, allowing users to remove or insert components. LoRA fine-tuning scripts for text-to-image and image editing are included. ## Takeaways - FLUX.3 Image Klein combines text-to-image, multi-reference image-to-image (using tokens), and layout control (via bounding boxes) in a single model. - Two checkpoints are offered: a distilled checkpoint (16 steps, no CFG) for fast inference and a base checkpoint (50 steps, CFG) for fine-tuning. Quantized NVFP4 and FP8 formats are available for smaller GPUs. - The pipeline is built with Modular Diffusers, enabling users to pop out the text encoder to encode once and generate multiple images, or insert custom blocks like Flux3PreviewStep to capture intermediate outputs. ## FAQ ### How does layout control work in FLUX.3? Layout control is done by appending a JSON list of regions to the prompt. Each region has an id, a bounding box (on a 0–1000 scale), and a description. The caption references each region by its id. For edits, regions can include a src\_bbox and tgt\_bbox to move content from a reference image. ### What are the differences between the distilled and base checkpoints? The distilled checkpoint uses 16 inference steps with no classifier-free guidance, generating a 1024x1024 image in about 2 seconds on H100. The base checkpoint requires 50 steps and supports CFG, making it suitable for fine-tuning. ### How can I fine-tune FLUX.3? LoRA fine-tuning scripts for both text-to-image and image editing are included in the release. The base checkpoint is recommended for fine-tuning. **Tags:** [Generative AI](https://yomu.fyi/topic/generative-ai), [Hugging Face](https://yomu.fyi/topic/hugging-face), [Inference](https://yomu.fyi/topic/inference), [LoRA](https://yomu.fyi/topic/lora) - Source: [Hugging Face](https://huggingface.co/blog/flux-3) - Source URL: https://huggingface.co/blog/flux-3 - Ingested by Yomu: 2026-10-15T18:00:15.003Z [Read original post](https://huggingface.co/blog/flux-3)