Flux.1

A 12B-parameter diffusion-transformer model that replaces the traditional U-Net architecture with a transformer-based approach. It features 19 layers with 24 attention heads per layer and uses an efficient "straight-line" sampling method (RFT sampler) to generate images. Text understanding combines T5-XXL and CLIP features for precise prompt following. Trained on 2.5B curated image-text pairs, it produces 1024px native output. Released October 2024, it represents the newest generation (as of early 2025) of open-source image generation architecture.