AI image models: xAI Grok Imagine

aitamer.news is produced with Grok Bot, an xAI product. xAI has no say over this review.

AI image models, episode 3. Previously: FLUX and Nano Banana vs GPT Image. Next: Midjourney vs Ideogram.

This series explains one or two AI image models per post, using the same fourteen questions every time. Facts come from the vendor’s own pages unless marked as a user report or our opinion. Prices were checked on 29 October 2026, and we generated no images for this episode.

Grok Imagine is the image and video generator from xAI, whose developer docs and terms now carry the SpaceXSI name. We covered the video side in AI movie models: xAI Grok Imagine. The current image model is Grok Imagine Image 2.0 (grok-imagine-image-2.0), released on 7 August 2026. In the Grok apps it is the “Quality” mode. Two older API models matter for this review: Grok Imagine Image 1.0 (grok-imagine-image), still on sale and cheaper, and the earlier Quality model (grok-imagine-image-quality), which xAI retires in the API on 2 November 2026.

What it does differently. xAI built 2.0 for layout and text. Its launch post says the model “plans typography and layout the way a designer would, so dense, multi-part visuals hold together and small text comes out sharp”, and the app adds templates for headshots, product shots, icons, game assets, mascots and emoji. It is also inside the Grok and X apps, so many people reach it through a subscription they already have.

If you use the API, act before Sunday. xAI’s migration guide says that from 2 November, requests to grok-imagine-image-quality (and the older grok-imagine-image-pro alias) are served by 2.0 at quality: low. xAI says 2.0 low costs $0.01 less per image at every resolution. 1.0 is not affected.

Grok Imagine Image 2.0 against the checklist

1. Max resolution. Three sizes: 1k (the default), 1.5k and 2k (xAI API reference and pricing). xAI does not publish exact pixel sizes. 2K is the ceiling, below the 4K of Nano Banana 2.1 and the 3840 x 2160 of GPT Image 2.5.

2. Aspect ratios. Fifteen fixed ratios plus auto: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 2:1, 1:2, 19.5:9, 9:19.5, 20:9, 9:20, 21:9 and 5:2. The phone-shaped 19.5:9 and 20:9 are unusual. When you edit, the output follows the first input image’s ratio unless you set one (xAI docs). In the apps, “smart resize” reframes an image to ratios from 1:2 to 2:1.

3. Reference types. Through the API, an edit takes source images as a URL, base64 or an uploaded file ID, and the prompt refers to them as <IMAGE_0>, <IMAGE_1> and so on. That covers style transfer, combining subjects and multi-turn editing, where each result becomes the next input. The apps add a “magic wand” to edit one region, segmentation, and background removal with a transparent background (xAI launch post). xAI’s API docs show no mask parameter, and the OpenAI SDK’s images.edit() call is not supported, so you send JSON.

4. References at once. Up to 5 input images per edit, raised from 3 in August (xAI release notes). That is well below Nano Banana 2.1 (14) and GPT Image 2.5 (16). One request can return 1 to 10 images.

5. Price (checked 29 October 2026). API prices per output image for 2.0:

Resolution	low	medium
1K	$0.04	$0.06
1.5K	$0.05	$0.07
2K	$0.06	$0.08

Each input image on an edit adds $0.01. quality: auto is the default, which xAI says means low for new images and medium for edits, and you are billed for the quality actually served. For comparison, 1.0 costs $0.02 per image at 1K or 2K, and the retiring Quality model cost $0.05 to $0.07. The Batch API works with 2.0. In the apps, the free tier includes “limited” image generation, and paid plans (Plus $8, Premium $30, Super $100, Ultra $200 a month) share one weekly usage pool across chat, Imagine and voice. API use is billed separately.

6. Speed and leaderboard rank. xAI publishes no generation-time figure. The API allows 6 requests a second and offers deferred generation for slow jobs. Ranks, read on 29 October:

• Artificial Analysis, text to image: #6 with 1166 Elo, tied with MAI-Image-2.6 and behind GPT Image 2.5 Sunburst (1222), Flare (1207), GPT Image 2 (1186) and Nano Banana 2.1 (1176). Artificial Analysis prices it at $60 per 1,000 images, using medium at 1K.
• Artificial Analysis, image editing: #17 with 1117. The leaders are Sunburst (1199), MAI-Image-2.7-Preview (1193) and Flare (1182).
• Arena, text to image: #6 with 1317, marked preliminary (board dated 26 October).
• Arena, image edit: #5 with 1439, also preliminary (board dated 23 October).

So the two boards agree on creation but disagree on editing. xAI’s launch post said 2.0 “ranks second in the world” on Arena for creation and editing. That was true on 7 August, and newer models have passed it since. In September, Arena said Microsoft’s MAI-Image-2.6 “outperforms Grok Imagine Image 2.0 while also costing less”.

7. Character consistency. xAI says 2.0 “preserves what you put in across generations and edits”, and the app’s “Build a world for video” tool makes a character, her locations and props “generated separately, holding one style”. We found no independent test of character drift. xAI’s own prompting guide is candid about one trap: an edit that pastes in a full new prompt “restages the scene: the subject drifts, a second copy of a prop appears”. Keep edit prompts short and specific.

8. Style and scene consistency. The templates and the world builder are the tools for keeping a look across a set. The prompting guide also says to lock aspect_ratio across a series and to use medium at 2k when fine detail matters. Artificial Analysis’s #17 editing rank suggests edits change more than you asked for more often than the leaders do. Arena’s #5 points the other way, so test this on your own images.

9. Compared with the previous version. Clearly better on both boards. On Artificial Analysis, 1.0 ranks #34 for creation (1022) and #26 for editing (1073), and the Quality model ranks #20 and #19. When 2.0 arrived on Artificial Analysis in September it debuted at #4, “the highest outside OpenAI”, after a 14-place climb. Against 1.0 you pay two to four times as much per image. Against the Quality model, 2.0 low is cheaper.

10. Text rendering. This is the model’s selling point. xAI promises sharp small text and planned layouts, and Artificial Analysis said in September that 2.0’s strongest categories were knowledge, text rendering and reasoning. Use it for posters, menus, infographics and product labels.

11. What users think (user reports). Our X pass found mostly promotion from resellers and side-by-side posts with no verdict. One reseller lists it at $0.06 an image. We found no detailed independent hands-on review to quote, so treat the board results above as the best outside signal for now.

12. General quality. We did not generate images for this episode. On the boards, 2.0 is a top-six model for creation, a step behind OpenAI’s GPT Image 2.5 pair and Nano Banana 2.1, and level with MAI-Image-2.6. Where it stands for editing depends on the board. Examples are in xAI’s launch post.

13. Licence, commercial use, open weights. Closed model, no weights. Under xAI’s consumer terms (last updated 27 October 2026), the heading reads “You Own Your User Content”, covering input and output. The same terms also say that “When using Output or SpaceXSI’s name, logos, trademarks, or other brand elements, you are required to obtain our permission and attribute your generation of the Output to the Service”. Read literally, that asks for permission as well as attribution, so check xAI’s Brand Guidelines before commercial use. Another clause says you may be required to add an AI disclosure to output, or SpaceXSI may add one. Images from the Grok apps carry a Grok watermark. xAI’s FAQ says there is no setting to remove it, and removing it breaks the Acceptable Use Policy. API use falls under xAI’s enterprise terms.

There is also a safety record to weigh. In January, the UK regulator Ofcom opened an investigation into X over sexualised images made with Grok. Its 14 October update says xAI has since blocked images of real people in revealing clothing such as bikinis, and blocked UK users from pornographic content in the standalone Grok image tool. Ofcom says the Online Safety Act does not let it investigate standalone Grok image creation. Its investigation into X continues.

14. Our verdict (opinion). Grok Imagine Image 2.0 is a strong, mid-priced all-rounder. Choose it for text-heavy designs, or if you already pay for Grok and want images in the same place. For 4K, more than five references or the best blind-vote scores, go to Nano Banana 2.1 or GPT Image 2.5. For bulk drafts, 1.0 at $0.02 is still worth a look. If your code calls grok-imagine-image-quality, switch the model name this week and check the new outputs before 2 November does it for you.

Summary

#	Checklist	Grok Imagine Image 2.0
1	Max resolution	2K (1k, 1.5k, 2k)
2	Aspect ratios	15 fixed plus auto, 1:2 to 21:9
3	Reference types	edit, style transfer, multi-turn; app: region edit, background removal
4	References at once	5 (up to 10 outputs per call)
5	Price (29 Oct)	$0.04 (1K low) to $0.08 (2K medium), +$0.01 per input image
6	Speed, rank	no speed figure; AA #6 create, #17 edit; Arena #6 and #5
7	Character consistency	claimed by xAI, untested; full-prompt edits drift
8	Style/scene consistency	templates, world builder; boards disagree on edits
9	vs previous	far above 1.0 and Quality; Quality retires 2 Nov
10	Text rendering	the main strength
11	Users	little independent review yet
12	Quality	top six for creation
13	Licence	you own output; attribution and permission clause; watermark
14	Verdict (opinion)	the text and layout pick at a mid price