FLUX.1-dev is a powerful 12-billion parameter rectified flow transformer designed for generating high-quality images from text prompts. It delivers cutting-edge output quality, just slightly below the flagship FLUX.1 [pro] model, and matches or exceeds many closed-source competitors in prompt adherence. The model is trained using guidance distillation, making it more efficient and accessible for developers and artists alike. FLUX.1-dev is openly available with weights provided to support scientific research and innovative creative workflows under a non-commercial license. It integrates smoothly with popular tools like the Diffusers library and ComfyUI, offering flexible options for local and cloud-based inference. The model supports generating large, detailed images up to 1024x1024 resolution with customizable parameters such as guidance scale and inference steps.
Features
- 12 billion parameter rectified flow transformer architecture
- Near state-of-the-art image quality, second only to FLUX.1 [pro]
- Competitive and accurate prompt following comparable to closed-source models
- Efficient inference enabled by guidance distillation training
- Open weights available to foster scientific and creative innovation
- Compatible with Hugging Face Diffusers and ComfyUI for flexible usage
- Supports high-resolution image generation up to 1024x1024 pixels
- Customizable parameters including guidance scale, inference steps, and random seed