xAI's New Image Model Lands Second in the Arena Rankings, and Brings Photoshop-Style Editing to Grok
Imagine Image 2.0 scores just behind OpenAI's GPT-Image-2 on both public leaderboards. The more interesting part is the editing tools, which point at where image AI is actually heading.
xAI has released Imagine Image 2.0, a new image generator available as a “Quality Mode” on grok.com/imagine and in the Grok apps for iOS and Android. API access is coming, the company says, without a date. On the public Arena leaderboards as of August 7, the faster “low” variant takes second place worldwide in both categories.
A word on what those rankings mean, because they get quoted a lot. The Arena leaderboards work by showing people two images from two anonymous models and asking which is better. Thousands of those votes produce an Elo rating, the same system used to rank chess players. Imagine 2.0 scores 1,439 in the Image Edit Arena against GPT-Image-2’s 1,463, and 1,320 in Text-to-Image against 1,380. Reve 2.1, Meta’s Muse-Image, Alibaba’s Qwen-Image-3.0-Pro, Google’s Gemini and ByteDance’s SeedDream all rank below. Two things follow: second place here is genuinely close to first, and the whole ranking measures which picture people prefer at a glance, not accuracy, not usefulness, and not whether the model gave you what you asked for.
The editing tools are the part worth paying attention to. Magic Wand changes only a selected area of an image rather than regenerating the whole thing. A segmentation feature lets you pick out precise regions, and background removal exports a subject on transparency. Multi-Ref Editing combines up to five input images into one generation, and Smart Resize converts an existing image to a different aspect ratio by having the model fill in the new space. xAI is also shipping templates, preconfigured starting points for common jobs like product photography, marketing material, game assets and streaming emojis, plus a feature that generates characters, locations and props separately while holding the visual style steady across all of them. The company frames that last one as groundwork for video production.
That shift is the real story across the whole field. The first few years of image AI were about the quality of a single generation from a single prompt, which is a demo. Actual work is iterative: you get something close, then change one thing, then change another, and keep the parts that worked. Tools for selecting, masking, and reusing references are what turn a novelty into something you can finish a job with, and they matter more than a 24 point Elo gap.
What this means for you: if you already make images with AI and keep hitting the wall where a small fix means regenerating everything and losing what you liked, these are the features that help, and the consistency tools are the most useful of the set for anything with a repeated character or product. If you are new to this, do not pick a tool by leaderboard position. The rankings between the top few models are close enough that workflow, price and whether it is available to you should decide it. A fair caveat: these are xAI’s own announcement and third-party leaderboard scores, not independent testing, and API pricing is not public yet.
Sources
Five Big Tech Companies Agree on One Format for AI Agent Add-Ons. Anthropic Is Not Among Them
Amazon, Cursor, Microsoft, OpenAI and Vercel launched Agent Plugins, a shared package format for AI agent extensions. It bundles Agent Skills and MCP servers into one folder.
