Krea 2 Turbo

Krea 2 Turbo is a text-to-image diffusion model that generates images from natural-language descriptions, optimized for creative, commercial, and developer workflows. It is a 12-billion-parameter Diffusion Transformer, released as an open-weight checkpoint by Krea.ai, Inc. on June 22, 2026. The model family includes two checkpoints: Krea 2 Raw (the base release) and Krea 2 Turbo, which benefits from additional fine-tuning and distillation for improved performance. Krea 2 Turbo is designed to support image generation, concepting, design exploration, and integration into applications and creative tools. Training data combined publicly available sources, licensed data, and proprietary synthetic data, with filtering to remove harmful content and improve quality. Safety measures include targeted fine-tuning to reduce susceptibility to harmful outputs, and multiple rounds of internal and external safety evaluations covering sexually explicit content, non-consensual imagery, child-safety risks, and other high-risk categories. Risks include unpredictable outputs, potential inaccuracies, and variable prompt following depending on phrasing. The model is not intended for factual information or fully automated decision-making affecting legal rights. Deployers must implement content filtering as required by the Krea 2 Community License. Users are solely responsible for their generated outputs and any third-party rights implications.