Understand Grok Imagine Image 2.0 before you build on it.
Grok Imagine Image 2.0 is SpaceXAI's image generation and editing model, published on 7 August 2026 and shipped as the new Quality Mode inside Grok Imagine on the web and in the iOS and Android apps. The launch post states the goal plainly — make images you can use in real work — and sells the model on three behaviours: following instructions down to the detail, planning typography and layout the way a designer would, and preserving what you put in across generations and edits.
Every figure on this page comes from SpaceXAI's own launch post for Imagine Image 2.0, its API documentation for the grok-imagine-image-2.0 model, its Imagine pricing table and its developer FAQ. The vendor's product pages are branded SpaceXAI; the same site's developer documentation still uses xAI in places. SpaceXAI publishes no standalone licence or output-ownership document for this model, and this page does not fill that gap.
What the named templates produce
The launch post introduces fifteen named templates — packaged starting points for photo editing, product shots, headshots, icons and game assets — and shows a cover frame for fourteen of them. Six of those covers are reproduced here. SpaceXAI publishes the prompt and the settings behind none of them, so the captions below say what each frame shows and which template and group it belongs to, and stop there.
Example images belong to SpaceXAI and are reproduced with credit: Grok Imagine Image 2.0 launch post
How SpaceXAI describes Grok Imagine Image 2.0
The launch post opens by naming the audience rather than the technology: the model is built to produce images you can use in real work. Everything the post claims follows from that — tight instruction following, in-frame layout, sharp small text, and inputs that survive a chain of edits. Precision, not aesthetics, is the axis the vendor competes on.
Built around real work, in the vendor's words
SpaceXAI states the goal as making images you can use in real work, and says 2.0 follows instructions closely down to the details.
Typography and layout planned the way a designer would
The post claims dense, multi-part visuals hold together and that small text comes out sharp — a layout capability rather than a font-rendering one.
Editing as a first-class capability
Grok Imagine Image 2.0 was trained for fidelity across photography, design and illustration, and the post treats editing as one of the model's primary jobs rather than a bolt-on.
Nothing lost between generations
SpaceXAI's claim is that the model preserves what you put in across generations and edits, which is what makes an iterative workflow practical.
Benchmarked, on the vendor's own terms
The launch post cites the Arena Image Edit and Text-to-Image leaderboards as of 7 August 2026 and states that 2.0 ranks second in the world in both. No other evaluation is offered.
What the launch post and the API docs document
This section carries only the behaviours SpaceXAI names — in the launch post for what the model does inside Grok Imagine, and in the API documentation for the parameters a developer can set.
The magic wand
Edits the region you point at and leaves the rest of the image untouched, which the post presents as the answer to the first generation rarely being the final asset.
Segmentation
Selects precise areas of the image to change, given as a separate tool from the wand and used the same way — name the region, then name the change.
Background removal
Exports any subject with a transparent background, ready to drop into other work. This is the one editing tool the post describes by its output file rather than by its on-screen control.
Multi-reference editing
Accepts up to five input images in a single generation, which SpaceXAI frames as removing the need for manual compositing. The API documentation confirms five and notes that the output ratio follows the first input unless you override it.
Smart resize
One image, any size: pick a ratio and the model fills in the frame. The app offers nine ratios; the API accepts fifteen, including the 21:9 and 5:2 widescreen options that arrived with this model.
One world, built image by image
A character, her locations and the props she carries, generated separately and holding one look across all of them — the post's example of consistency applied across a set rather than within one frame.
Documented specifications
Every row below is stated in SpaceXAI's launch post, its API documentation, its pricing table or its developer FAQ. Rows marked as unpublished are gaps the vendor has not filled, not values this page has estimated.
- Developer
- SpaceXAI
- Model
- Grok Imagine Image 2.0
- Launch date
- 7 August 2026
- Consumer surface
- Quality Mode in Grok Imagine on the web, iOS and Android
- API model id
- grok-imagine-image-2.0
- Reference images per edit
- Up to 5
- Image resolution (API)
- 1k by default; 2k available
- Quality parameter
- low, medium or auto; auto serves low for generation and medium for editing
- Smart resize ratios (in app)
- 1:2, 9:16, 2:3, 3:4, 1:1, 4:3, 3:2, 16:9, 2:1
- Aspect ratios (API)
- Those nine plus 19.5:9, 9:19.5, 20:9, 9:20, 21:9, 5:2 and auto
- API price
- $0.04 per image, flat regardless of prompt length; edits are billed for input and output
- Templates
- 15 named, grouped into Photo Tools, Product, Marketing, Design Tools, Game Assets and Streaming
- Watermark
- A Grok mark is applied to output; the vendor states there is no setting to remove it
- Slug retirement
- grok-imagine-image-quality retires on 2 November 2026 and is served by 2.0 at low quality
- Licence
- Not published
- Output ownership
- Not published
Documented access channels
SpaceXAI names two surfaces for this model — its own consumer product and its API — and publishes the per-image rate for the second. What it does not publish is which model each in-app template uses, or any regional restriction on either channel.
Grok Imagine on the web
The consumer surface, where Image 2.0 is the option labelled Quality Mode. The launch post says images land here first.
OpenGrok on iOS and Android
The same model shipped to the mobile apps in the same release, according to the launch post. SpaceXAI does not publish a separate model list per platform.
OpenThe Imagine API
Reached as the model id grok-imagine-image-2.0, with the parameter list, ratio table and quality modes documented per endpoint.
OpenImagine pricing
The vendor's own rate table, which lists this model at $0.04 per image and also carries the per-second video rates for the sibling model.
OpenGrok Imagine Image 2.0 FAQ
Can I use Grok Imagine Image 2.0 output commercially?
SpaceXAI does not publish an output-ownership or licence statement for this model on its launch post, its API documentation or its pricing page, so this page cannot answer that. What governs your use is the terms attached to whichever channel you reach the model through, together with the vendor's Acceptable Use Policy.
Is the Grok watermark removable?
No. The vendor's own developer FAQ states that generated images and videos carry a Grok watermark to indicate the content was created with AI, that there is no setting to remove it, that labelling AI-generated content is legally required in some jurisdictions, and that removing or obscuring the watermark is prohibited under the Acceptable Use Policy.
What does the API cost?
$0.04 per image for grok-imagine-image-2.0, charged per generated image and flat regardless of prompt length. An image edit is billed for both the input image and the output image. The vendor states that 2.0 at low quality is $0.01 per image cheaper than the model it replaces, at every resolution.
What happens to grok-imagine-image-quality?
It retires on 2 November 2026, with a 60-day notice period that began on 2 September 2026. After that date requests to the retired slug are served by grok-imagine-image-2.0 with quality set to low. The slug continues to resolve, so nothing breaks, but you are billed at the 2.0 low rate rather than choosing a quality yourself.
Which model do the in-app templates use?
Not published. The templates ship inside Grok Imagine, which serves Image 2.0 as its Quality Mode, but SpaceXAI does not state which model or settings any individual template uses. Treat the covers as output from the product, not as per-template specifications.
Is this xAI's model or SpaceXAI's?
Both names appear on the vendor's own pages. The site and the model pages are branded SpaceXAI, the footer reads © 2026 SpaceXAI LLC, and the launch post's own benchmark note says its models are listed on Arena under SpaceXAI. The same site's news item dated 2 February 2026 is headed xAI joins SpaceX and records SpaceX announcing it had acquired xAI, and the developer documentation still uses xAI in places. This page treats SpaceXAI as the publisher because that is the name the product pages currently carry.