Imagine Image 2.0 is xAI's image generation and editing model, live as the Quality Mode on grok.com/imagine and in the Grok apps for iOS and Android. At launch it held second place worldwide on both Arena leaderboards, behind OpenAI's GPT Image 2. You point at a region, describe the change, and everything else in the frame stays untouched. Free generation inside Grok is now heavily capped, so a paid plan is the realistic path for regular work.
The magic wand in Imagine Image 2.0 changes only the region you point at and leaves the rest of the canvas alone. Image segmentation picks out precise areas to modify, and background removal exports any subject with clean transparency, ready to drop into other software.
Here is what that looks like in practice: you generate a poster, one letter comes out wrong (the misspelled headline, a timeless classic), so you circle the word, type the fix, and the background does not shift by a pixel. Multi-reference editing goes further, blending up to five source images in a single generation.
| Tool | What it does | What it replaces |
|---|---|---|
| Magic wand | Changes only the pointed region | Regenerating the whole image |
| Segmentation | Selects a precise area to edit | Rough manual cut-outs |
| Background removal | Exports the subject on transparency | A separate editing app |
| Smart resize | Recomposes across nine aspect ratios | Crops that chop off the subject |
| Multi-reference | Blends up to five source images | Compositing by hand |
At launch, Imagine Image 2.0 sat second worldwide on both Arena leaderboards, scoring 1,439 Elo in image editing against 1,463 for GPT Image 2. One detail raised eyebrows: the model is listed there under the SpaceXAI name, a consequence of xAI moving under the SpaceX umbrella. Same team, new banner.
An early independent head-to-head published after launch put it ahead on photorealism and character consistency, and behind GPT Image 2 on multilingual text and style transfer. Ready-made templates of this kind exist across other AI image editing tools, but keeping a character identical from one generation to the next is still uncommon.
Imagine Image 2.0 runs from grok.com/imagine or the Grok app once you pick Quality Mode. Opening an account costs nothing, but generation without a subscription is tightly capped, and Quality Mode frequently redirects to a paid plan.
That leaves the developer side of the wallet question. xAI says API access is on the way, with no date or official rates yet. Meanwhile, third-party platforms such as fal bill the Quality tier per image, around $0.05 at 1K and $0.07 at 2K. Those figures shift quickly, so check xAI's official pages before spending anything.
Partly. Anyone can open Quality Mode in Grok on the web, iOS and Android, but generation without a subscription is heavily restricted and often ends at an upgrade prompt. A paid Grok plan lifts those caps, while the API has not opened yet.
GPT Image 2 still tops both Arena leaderboards, with Imagine Image 2.0 right behind in generation and editing. An independent comparison run after launch found xAI's model ahead on photorealism and character consistency, with GPT Image 2 winning on multilingual text and style transfer. The gap between the two is small.
Not officially yet. xAI lists API access as coming soon, without a date or published pricing. A preview version already circulates on third-party gateways such as Vercel AI Gateway and fal, billed per generated image. For production work, waiting for the official release and its final terms is the safer bet.
No, this model handles still image generation and editing only. Video runs through Grok Imagine's separate video models, which carry their own version numbers. The confusion comes from naming: people searching for Grok Imagine 2 often mean the video side, while Imagine Image 2.0 is xAI's image model.
Verdict: A botched letter on a poster no longer means starting over: designers on deadline, online sellers and marketing teams will claw back hours of retouching, and developers can keep an eye on the API opening to wire it into their own products.
