Grok Imagine 2.0 Is Live: What Image 2.0 Actually Changes

xAI shipped Grok Imagine 2.0 today, August 7, 2026. Imagine Image 2.0 is generally available right now as the new Quality Mode on grok.com/imagine and in the Grok iOS and Android apps. It is the first Grok image release aimed at production work rather than at feeds: xAI says it plans typography and layout "the way a designer would", holds dense multi-part visuals together, keeps small text sharp, and preserves what you feed it across generations and edits. The practical headline for creators is the editing suite that came with it, including region-level edits and up to five reference images in a single generation. The catch is the last line of the announcement: "API access is coming soon." Nothing outside xAI's own apps can reach this model yet.

Grok Imagine Quality Mode is now Image 2.0

There is no new toggle to find. xAI's wording is that Image 2.0 "is now generally available as the new Quality Mode", so the tier you were already paying attention to is the door. If you opened Quality Mode yesterday you were routed to the older grok-imagine-image-quality model. Open it today and you get 2.0.

That matters for anyone who tested Grok on text a few weeks ago and wrote it off. Quality Mode was already the better tier for typography, and our own Grok Imagine prompt guide told people to use it whenever words were part of the brief and to proofread every character anyway. Image 2.0 raises that ceiling rather than replacing the advice. Proofread the output; just expect to find fewer problems.

The editing suite is the actual news

Text-to-image quality moves in small increments now. Editing control does not, and this release is mostly an editing release. Four tools shipped with it:

• Magic wand. Point at a region and it edits that region, leaving the rest untouched. This is the difference between fixing one thing and regenerating an image you already approved.
• Segmentation. Select precise areas of the image to change, rather than describing them in prose and hoping the model picks the right object.
• Background removal. Export any subject on a transparent background, ready to drop into other work.
• Multi-ref editing. Up to five input images in a single generation, which xAI frames as removing the need for manual compositing.

The five-reference number is the one to note. xAI's current image API documents that "multi-image editing supports up to 3 source images in a single request", so 2.0 is a real jump in how much you can hand the model at once: a product, a model, a background, a logo, and a style plate, composed in one pass. Whether five references actually beat three in practice depends on giving each one a distinct job, which is the same discipline that governs the older model. We cover that in the Grok image editing guide.

Smart resize and 15 templates

Smart resize is the quieter feature and probably the one you will use most. Pick a ratio and the model fills in the frame rather than cropping the composition you liked. xAI shows nine ratios: 1:2, 9:16, 2:3, 3:4, 1:1, 4:3, 3:2, 16:9, and 2:1. For anyone shipping the same creative to a story, a feed post, and a display banner, that is the tedious half of the job.

The release also adds 15 templates that package a workflow into a starting point: Photo Edit, Product Color Change, Editorial Product Poster, Reimagine, Photo Collage, Mascot Maker, BG Removal & Change, E-Commerce Photos, UGC Photos, Professional Headshot, Icon Maker, Character Sprite, Props & UI Kit, Emoji Creator, and Merch Maker. Read that list as a statement of intent. Nine of the fifteen are commerce, marketing, or product work. xAI is aiming Imagine at the people who currently pay for design hours, not at the people making wallpapers.

One more capability is aimed at video teams: xAI describes generating a character, her locations, and the props she carries separately while holding one style from image to image, then building a world for video out of them. That is asset-set consistency, and it is the workflow behind every serialized short-form channel.

Where Grok Imagine 2 ranks against GPT Image 2

xAI claims Image 2.0 "ranks second in the world in both text-to-image generation and image editing", citing the Arena leaderboards as of August 7, 2026. The published order:

Two details are worth reading carefully before you treat this as settled. The entry ranked second is labelled grok-imagine-image-2 (low), meaning the low-compute setting, and xAI models appear on Arena under the name SpaceXAI. And the previous Quality Mode model, grok-imagine-image-quality, still sits mid-table in both boards, which is a fair measure of how far this jumped in one release.

The nuance that survives all of this: Arena is aggregate human preference across a broad prompt mix, and it does not tell you which model wins your brief. GPT Image 2 still leads both boards. Nano Banana Pro remains the model we reach for on multi-label infographics and print-bound layouts. Second place across two general leaderboards makes Grok a default candidate for far more jobs than it was yesterday. It does not make it the answer to every job.

What this changes for your workflow this week

If you produce commercial creative, the useful move today is a bake-off rather than a migration. Take three assets you already shipped, a product poster with real copy on it, a lifestyle composite built from separate references, and a headshot, and rerun them in Quality Mode. You will learn more from three of your own briefs than from any leaderboard, and the region-level editing tools mean a near-miss is now a fix rather than a reroll.

What you should not do is rebuild a pipeline around 2.0. It lives inside grok.com and the Grok mobile apps, which means manual work, manual export, and no way to run a set of thirty variants against a brief. That is a consumer surface, and it stays one until the API lands.

The API is not here yet, and that gates everything

xAI ends the announcement with "API access is coming soon" and gives no date. We checked xAI's models and pricing page today: it lists grok-imagine-image at $0.02 per image, grok-imagine-image-quality at $0.05 per image, grok-imagine-video at $0.050 per second, and grok-imagine-video-1.5 at $0.080 per second. There is no grok-imagine-image-2 entry, no price, and no rate limit published for it.

Until that changes, every studio, agency tool, and automation that routes to Grok is running the older models, including this one. Anyone telling you otherwise today is describing Quality Mode from before this morning. We are tracking the rollout in the Grok Imagine API post, which is where the model id and pricing will land when they are published.

What Dream Pixel Forge runs today, and what changes when the API drops

Being precise about this matters more than being fast about it. Dream Pixel Forge already runs Grok Imagine as one adapter behind its provider router: xai/grok-imagine-image, at 2 credits per image, with editing from up to three input images and seven aspect ratios. It is the automatic flexible baseline, the cheap lane you iterate in before spending premium credits on a keeper. Grok video runs alongside it at 24 credits for a 6-second 480p clip. None of that is Image 2.0. It is the standard model, and it is what the router picks today.

Image 2.0 is on the list for the day xAI ships the API, on the same footing as every other model in the registry: a model entry, a credit price set from real per-image cost, and a place in the fallback chain. When it lands, the parts of this release that are hardest to reproduce elsewhere, the five-reference compositing and region-level editing, become things your agent or the freeform generator can call directly instead of things you do by hand in a chat window. Until then, the honest recommendation is to test 2.0 in Grok, keep production on models you can actually script, and use background removal and the editing ops you already have.

The short version

Grok Imagine 2.0 is live today as Quality Mode, it ranks second on both Arena boards behind GPT Image 2, and the real upgrade is editing control: point-and-edit regions, segmentation, transparent-background export, five references in one generation, and smart resize across nine ratios. It is a consumer release for now. Test it on your own briefs this week, do not rebuild anything around it, and watch for the API.