Elon Musk’s SpaceXAI today (August 8) published a blog post announcing the launch of the Imagine Image 2.0 model. It is now available as a new Quality Mode on the Grok.com web version and the iOS and Android apps.
According to the official website, the Imagine Image 2.0 model follows user instructions more closely and can handle detailed requirements. For complex visual content, it plans text placement and layout structures to keep multi-part scenes consistent while improving the clarity of small text. The model can also preserve the subject and settings provided by users across multiple rounds of generation and editing. Related demonstration images are shown below:
Imagine Image 2.0 significantly enhances image-editing capabilities. SpaceXAI said that in real-world workflows, multiple rounds of revisions are typically needed, and the first generated image is rarely the final product. The model therefore includes several tools for making localized modifications and reusing assets:
The Magic Wand tool edits only the area indicated by the user while leaving other areas unchanged;
The segmentation function selects a precise area in the image for targeted modifications;
The background removal function exports the subject with a transparent background, making it easy to place in other works;
Multi-reference image editing supports entering up to 5 images in a single generation, reducing the need for manual compositing.
Smart resizing: after the user selects the target aspect ratio, the model completes the image frame so that the same image can meet different size requirements.
In terms of performance, Image 2.0 was trained for photography, design, and illustration scenarios. According to data from the Arena Text-to-Image and Image Edit rankings cited by xAI, as of August 7, 2026, Image 2.0 ranked No. 2 globally in both text-to-image generation and image editing, behind only OpenAI’s GPT Image 2 model.
