https://github.com/krea-ai/krea-2 https://huggingface.co/krea/Krea-2-Raw Krea 2

Two models are provided. Krea 2 RAW is the base model—a pretrained checkpoint without any distillation, offering high flexibility and adaptability, making it ideal for fine-tuning, post-training, and LoRA training. Krea 2 Turbo is an 8-step distilled checkpoint designed specifically for fast, high-quality text-to-image generation.

These two models are designed to work together. You can train a LoRA model on RAW and then apply it to Turbo; LoRA models trained on RAW run seamlessly on Turbo as well. We strongly recommend training LoRA models on RAW and applying them to Turbo for inference.