Grok Imagine 2.0 is xAI's new image model, built so editing is part of the model instead of a bolt-on. Point at a region and change only that. Composite from five references in one pass. Text-to-image generation is live on reAPI today.
required · be specific about layout and text
1–12 · billed per delivered image
7 ratios · square to full vertical
single mode · already the default
Try one of these prompts
Credit-based — 1 credit = $0.001 USD. Pay only for completed generations.
One sample at the cheapest tier (1 image).
Grok Imagine 2.0 is the image generation and editing model xAI shipped on 7 August 2026 as the new Quality Mode inside Grok. It was built for one thing: images you can use in real work. Grok Imagine 2.0 follows instructions down to the detail, plans typography and layout the way a designer would, and keeps small text sharp in dense, multi-part visuals. The real break from a plain text-to-image model is editing. A magic wand changes only the region you point at. Segmentation picks precise areas. Background removal exports any subject on transparency. And multi-reference editing takes up to five input images in a single generation. On the Arena boards cited in xAI's announcement, Grok Imagine 2.0 ranks second in the world in both text-to-image and image editing. Grok Imagine 2.0 runs today inside Grok's own apps, and text-to-image generation is now live on reAPI — the console on this page is connected, and one request returns up to 12 images. The editing capabilities described above are not exposed through the API yet.
Real-world workflows and production use cases you can build and ship with this model.
The first generation is rarely the final asset. Point at a label, a sky, a jacket — Grok Imagine 2.0 changes that region and leaves everything you already approved untouched. No more regenerating a whole frame to fix a corner of it.
Read the docs
Feed a character, a location and a prop into Grok Imagine 2.0 at once and get one composed frame back. Up to five inputs per generation, so the manual layer-stitching pass disappears from your workflow.
Posters, packaging and UI mockups fail on typography first. Grok Imagine 2.0 plans type and layout deliberately, so dense compositions hold together and small text stays sharp in Grok Imagine 2.0 output instead of melting into texture.
Magic wand, segmentation and background removal are part of Grok Imagine 2.0 rather than a post-process, so an edit preserves what you approved instead of regenerating it.
Grok Imagine 2.0 accepts five reference images per generation, replacing the manual compositing step that used to sit between a generation and a usable asset.
Small text stays legible and multi-part layouts hold in Grok Imagine 2.0 — the two things that usually send AI imagery back for a redo.
xAI's own announcement puts gpt-image-2 first on both Arena boards and Grok Imagine 2.0 second — and gpt-image-2 already ships on reAPI. Here is how they line up on the capabilities xAI has published.
Comparison reflects publicly documented behavior at the time of writing: xAI's Grok Imagine 2.0 announcement (7 August 2026) and OpenAI's published gpt-image-2 capabilities. Grok Imagine 2.0 is not yet callable on reAPI, so its side describes announced capabilities rather than measured behavior. Rankings are the Arena leaderboards cited by xAI as of 7 August 2026.
Sign up and generate a key in your reAPI console — the same key calls Grok Imagine 2.0. Starter credits are included so you can test first.
OpenPost a prompt, an image count and an aspect ratio to the shared images endpoint — or use the console on this page, which builds the same request body for you.
OpenEvery image model on reAPI speaks the same async request shape, so the integration you write today is the one that will call Grok Imagine 2.0.
Opencurl -X POST https://reapi.ai/api/v1/images/generations \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "grok-imagine-image-2-0",
"prompt": "An editorial product poster with crisp small type",
"n": 1,
"size": "3:2"
}'Rejected synchronously with the failing field named. Check enum values, ranges, and URL fields — nothing is charged.
401 means a missing or invalid Bearer key; 402 means the reserve exceeds your balance. Manage keys and credits in the console.
Reference material and outputs pass automated moderation. Rejected tasks fail with a clear error and a full refund.
A task that reaches a failed state is never charged — the reserve refunds automatically. Keep the task id and retry when ready.
Common questions about this model.
Explore more models in the same category.
SpaceXAI
From $0.005 per image
From $0.014 per image
OpenAI
From $0.023 per image
From $0.015 per image
Try it in the playground or grab an API key to integrate now.