Welcome FLUX.3 Image Klein in Diffusers
huggingface ·
Hugging Face has integrated Black Forest Labs' FLUX.3 Image Klein model into the Diffusers library, offering a versatile solution for text-to-image and multi-reference image-to-image generation. The model introduces a novel layout control feature that allows users to define specific regions using bounding boxes within prompts for precise compositional guidance. Two primary checkpoints are available: a distilled version optimized for fast inference with fewer steps, and a base model designed for fine-tuning via LoRA. Performance benchmarks indicate that the distilled checkpoint can generate high-resolution images in approximately two seconds on an H100 GPU, while quantized variants support deployment on smaller hardware. The underlying architecture utilizes a two-stage transformer process that initially processes text, image, and reference tokens separately before merging them for joint attention. Additionally, the integration supports modular pipelines, enabling developers to customize workflows by inserting or removing processing blocks. This release provides comprehensive tools for both immediate usage and advanced customization through fine-tuning scripts.
huggingface