Editing is a major part of the release. A magic wand tool lets users target a particular part of an image while leaving other areas unchanged, while segmentation provides more precise control over which portions are modified. Image 2.0 also includes background removal with transparent exports and can take as many as five reference images in one generation, allowing elements from multiple inputs to be incorporated without manually compositing them first.
A new Smart Resize feature addresses another common production task. Users can select a different aspect ratio and have the model extend and recompose the existing scene to fit the new dimensions. xAI showed the same source image adapted across formats ranging from tall 1:2 and 9:16 layouts to square, landscape and 2:1 banner compositions.
Alongside the model, xAI is introducing templates that package frequently used image workflows into predefined starting points. The initial selection covers tasks including photo editing, product color changes, e-commerce photography, professional headshots, background replacement, icons, merchandise, character sprites, game assets and social content. Users provide the necessary inputs while the template handles the underlying workflow configuration.
Consistency across separate generations is another area xAI is emphasizing. The company demonstrated characters, environments and objects created independently while retaining a common visual style, an approach intended to help users assemble multiple assets that belong to the same visual world.
xAI also published benchmark results for Image 2.0. Based on Arena Image Edit and Text-to-Image leaderboards as of August 7, 2026, the company says Grok Imagine Image 2 ranks second globally in both categories. In the figures provided by xAI, a lower-cost configuration of Grok Imagine Image 2 recorded an Elo score of 1,439 for image editing, behind OpenAI's GPT Image 2 at 1,463. In text-to-image generation, the same Grok model scored 1,320 compared with GPT Image 2 at 1,380.
Those results place xAI's latest model among a competitive group that also includes image systems from OpenAI, Meta, Microsoft, Google, ByteDance and Alibaba. The rankings vary depending on the model configuration and leaderboard, with xAI's Grok Imagine Image Quality appearing separately from the Image 2 low-cost configuration in the results it published.
The Image 2.0 release builds on xAI's broader expansion of Grok Imagine beyond still-image generation. The company made Grok Imagine Quality Mode available through its API on May 6, while its July 31 developer release added text-to-video, image-to-video and reference-to-video capabilities to grok-imagine-video-1.5. Native 1080p output is available for text-to-video and image-to-video generation.
xAI has also structured its API pricing around different levels of image quality. The company lists grok-imagine-image at $0.02 per 1K or 2K output image. Grok Imagine Image Quality costs $0.05 for a 1K output and $0.07 at 2K. Grok Imagine Video 1.5 is priced by duration and resolution, ranging from $0.08 per second at 480p to $0.25 per second at 1080p.
With Image 2.0, xAI is putting more emphasis on the steps that come after an initial prompt. Precise regional edits, multiple references, transparent backgrounds, resizing and reusable templates are designed around workflows where an image is repeatedly changed before it becomes a finished asset. The result is a version of Grok Imagine aimed less at one-off generations and more directly at recurring creative and production work.
This analysis is based on reporting from xAI.
Image courtesy of Grok.
This article was generated with AI assistance and reviewed for accuracy and quality.
