Grok Imagine Image 2.0 · xAI · Ranked #2 on Arena
Grok Imagine Image 2.0 — Try AI Image Generation Free Online
xAI shipped Imagine Image 2.0 on 7 August 2026: typography planned before render, edits that change only the area you point at, up to five reference images, and generative resizing across nine aspect ratios. Here is everything it does — plus a free generator you can use right now, no waitlist and no app install.
Capabilities
What Grok Imagine Image 2.0 actually shipped
xAI released Imagine Image 2.0 on 7 August 2026 as the new Quality Mode inside Grok. Six things it does that the previous Imagine image model did not.
Typography planned, not patched
Image 2.0 works out layout and type the way a designer would, before it renders a single pixel. Dense posters and infographics come back with the small captions still legible instead of dissolving into pseudo-letterforms.
Magic wand and segmentation
Point at any area of the image and change only that area. Segmentation picks out precise shapes, so recolouring a jacket or rewriting a sign leaves the rest of the frame exactly where it was.
Background removal built in
Export a subject on a transparent background straight out of the model. No separate cutout pass and no manual masking before the asset drops into a layout.
Up to five reference images
Feed as many as five inputs into a single generation to hold a character, a product, or a brand palette steady across a whole set, instead of re-rolling until things happen to match.
Smart resize across nine ratios
Recompose one image from a 1:2 banner all the way to 16:9 widescreen. It generates the parts of the frame that were never there rather than cropping your subject out of shot.
Quality Mode, 1K or 2K
Image 2.0 is the Quality Mode on grok.com/imagine and in the Grok iOS and Android apps, with 1K and 2K output and batch generation for sweeping through variants.
Comparison
Grok Imagine Image 2.0 vs GPT-Image-2
These two sit second and first on the Arena leaderboards. Here is where they actually differ — and which one is running when you generate on this page.
| Capability | Grok Imagine Image 2.0 | GPT-Image-2 |
|---|---|---|
| Built by | xAI | OpenAI |
| Arena text-to-image | Ranked #2 | Ranked #1 |
| Arena image editing | Ranked #2 | Ranked #1 |
| Max resolution | 2K | 4K |
| Editing approach | Magic-wand area selection, segmentation, one-step background removal | Prompt-driven image-to-image editing |
| Consistency across a set | Up to five reference images in one generation | Up to 16 reference images per edit |
| Aspect ratios | Nine, from 1:2 to 16:9, with generative resize between them | 1:1, 4:3, 3:4, 16:9, 9:16, auto |
| Where you can use it | grok.com/imagine Quality Mode, the Grok iOS and Android apps, or the Vercel AI Gateway at $0.05 per image | In the generator on this page — pick it in the model selector |
Arena rankings as reported on 7 August 2026. Grok and Grok Imagine are products of xAI; GPT-Image-2 is a product of OpenAI. Images 2.0 is an independent service and is not affiliated with either company.
Use Cases
Where a layout-aware image model earns its keep
A model that plans its own typography matters most wherever text and image have to survive together. Six jobs Imagine Image 2.0 is built for.
Posters & ad creative
Headline, subhead, and the small legal line all rendered in one pass, at the sizes they will actually be read at. Fewer round trips into a design tool just to fix the type.
Product & listing images
Background removal to a transparent export means a product shot drops straight onto a marketplace template. Multi-reference input keeps the same item recognisable across a whole listing.
Social posts & app screenshots
One composition, then smart resize into every ratio a platform asks for — 9:16 story, 1:1 feed, 16:9 header — with the frame regenerated instead of the subject cropped away.
Thumbnails & channel art
Short, punchy overlay text that lands sharp at thumbnail scale, and a reference image to keep a channel's visual system consistent from episode to episode.
Illustration & concept art
Editing one area at a time means an iteration is a local change, not a re-roll. Fix the hand, change the light, swap the background, and keep the composition you already liked.
Infographics & diagrams
The case the model was tuned for: many labels, several columns, and type at a range of sizes that all has to stay readable in the same frame.
Grok Imagine Image 2.0 FAQ
The questions people actually ask about xAI's image model