New state-of-the-art image generation model. It is a natively multimodal language model that accepts both text and image inputs, and produces image outputs.