The technology will be initially available on the X platform in select countries, with a full rollout planned within a week.
The Aurora model — an advanced mixture-of-experts network — was trained on billions of internet examples to generate photorealistic images with precise text instruction following. xAI said in a press release that the system demonstrates particular strength in rendering detailed visual representations of real-world entities, including text, logos, and human portraits.
Key features of the new image generation system include high-fidelity image creation across multiple domains, ability to render precise visual details, native support for multimodal input, and the capability to take inspiration from or edit user-provided images (as illustrated by the Ghibli-style image of xAI CEO Elon Musk above).
xAI emphasized the model’s training approach, which involved processing interleaved text and image data to develop a comprehensive understanding of visual representation. The technology aims to distinguish itself from existing image generation models by offering more nuanced and accurate rendering.
Image editing functionality will be released to X platform users in the near future, providing additional creative control for users.