Enterprise-grade AI image generation in 2 seconds is here: Krea 2 Raw and Turbo available as open weights under custom license

· Source: VentureBeat · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, long

Summary

AI creative tools startup Krea has released Krea 2 Raw and Krea 2 Turbo, two new frontier AI image models, as open weights on Hugging Face under a custom license. Krea 2 Turbo achieves image generation in 2 seconds, making it one of the fastest available, while Krea 2 Raw serves as an undistilled base for custom training. The models are built on a 12-billion-parameter Diffusion Transformer architecture, designed for greater visual variety, prompt accuracy, and customization than typical generators. The custom license permits free commercial use for individuals and small businesses up to 50 seats, but requires larger enterprises to pay for usage and mandates all users implement technical safeguards against illegal content generation. Krea, founded in 2022 and having raised \$83 million, has evolved from an AI aggregator to a model provider, with over 30 million users across 191 countries as of June 2026.

Key takeaway

For AI Engineers or Creative Technologists seeking flexible, high-speed image generation, Krea 2 Raw and Turbo offer a compelling open-weight solution. You can train custom LoRAs on Krea 2 Raw for unique aesthetics, then port them to Krea 2 Turbo for near-instant 2-second outputs. Be aware that the custom license requires implementing content moderation safeguards, and enterprises over 50 seats need a paid commercial agreement. This approach provides artistic control while balancing commercial viability and safety.

Key insights

Krea 2 offers fast, customizable AI image generation via open-weight models with a hybrid licensing and mandatory safety framework.

Principles

Method

Krea 2 uses a 12B-parameter Diffusion Transformer. Raw is an undistilled base for LoRA training, while Turbo is a distilled variant for rapid 8-step inference, enabling 2-second 2k resolution image generation.

In practice

Topics

Best for: CTO, Machine Learning Engineer, Computer Vision Engineer, AI Engineer, Director of AI/ML, Creative Technologist

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by VentureBeat.