HiDream-O1-Image: An open 8B model that generates images directly in pixel space

HiDream released an MIT-licensed 8B model that operates directly on pixels without a VAE or separate text encoder. It supports text-to-image generation, editing, subject personalization, and storyboarding at up to 2,048 × 2,048. It scored 0.90 on GenEval, 89.83 on DPG-Bench, and 10.37 on HPSv3, and placed eighth in the Artificial Analysis Text-to-Image Arena.

Visit original source