# Grok Image: how to create and edit pictures with Grok Imagine — guide and prompts

> Grok Image is image generation in xAI's Grok. The current model, Imagine Image 2.0, launched on August 7, 2026 and debuted at #2 on the Arena leaderboard: it renders sharp text, edits a selected area, removes backgrounds and builds one frame from up to 5 photos. In AI2Face Grok Image is available without a SuperGrok subscription.

Published 2026-10-13 · Updated 2026-10-28
URL: https://ai2face.com/blog/grok-image

In two years Grok Image went from a meme generator inside X to a model xAI pitches as a work tool: posters with small text, product shots, in-place photo edits. We made a poster, a portrait, a product photo, a before/after edit and a two-photo composite in Grok Image 2.0, and went through the xAI docs on what the model can do, how to write prompts and where the limits are. For an overview of both xAI models, for images and for video, see [Grok Imagine](/models/grok-imagine); the image model itself is covered in detail on the [Grok Imagine Image 2.0](/models/grok-imagine-image) page.

## What is Grok Image

Grok Image is image generation and editing in xAI's Grok chatbot. In the interface and the API it's part of Grok Imagine, the visual side of Grok, which also covers [Grok video](/blog/grok-video).

| Model | When | What matters |
|---|---|---|
| Aurora | December 9, 2024 | xAI's first in-house image model, an autoregressive mixture-of-experts; launched on X |
| Grok Imagine Image | before Image 2.0 | Fast base model, $0.02 per image in the API |
| Imagine Image 2.0 | August 7, 2026 | Quality mode in Grok: typography, in-place editing, up to 5 references; $0.04 per image in the API |

On launch day **Imagine Image 2.0** ranked #2 in the world on the Arena leaderboards for both image generation and image editing. As of September 24, 2026 it was #6 out of 80 models.

## How to generate a picture in Grok

1. **Open Grok Imagine** — grok.com/imagine or the Imagine section of the Grok app for iOS and Android. You can also ask for a picture in regular chat, but Imagine gives you aspect ratios and editing tools.
2. **Turn on Quality mode** — that's Image 2.0. The fast mode is fine for drafts.
3. **Pick an aspect ratio.** Image 2.0 supports 16 of them: from 1:1 to tall 9:16, 9:19.5 and 9:20 and panoramic 21:9 and 5:2. The API defaults to `auto`, which picks a ratio from the prompt.
4. **Write the prompt** using the formula below.
5. **Generate several variants.** Through the API you can get up to 10 images per request.
6. **Refine with edits, not new generations.** It's better to fix a good frame in place than to rewrite the prompt from scratch.

## What Grok Image 2.0 can do

**Text and layout.** xAI says Image 2.0 plans typography and layout like a designer: dense, multi-part visuals hold together and small text stays sharp. In our test, a poster with a title, a subtitle, a dot separator and a date range came out without a single mistake on the first try.

![Retro poster SAILS & SALT: white sailboats on a turquoise sea, a yellow sun, red-roofed houses on cliffs, with the line SUMMER FESTIVAL · JULY 12–14 under the title](https://ai2face.com/_astro/poster.ZYt5AmjY_f9Bn0.jpg "Title and a small date line with a separator — no mistakes on the first try. Grok Image 2.0, 2K, 3:4")

The poster prompt puts the exact text right in the prompt and rules out any other text:

```
Minimalist retro travel poster for a fictional seaside town, screen-print style:
white sailboats on a turquoise sea under a pale yellow sun, bold title SAILS & SALT
at the top in large condensed letters, and a smaller line SUMMER FESTIVAL · JULY 12–14
below it, limited palette of navy, turquoise, cream and coral, no other text
```

**Photorealism and product shots.** Skin, fabric and stone texture are strengths. Image 2.0 takes lens (85mm, 50mm) and lighting cues (golden side light, hard morning sunlight) literally: depth of field and shadows come out as prompted.

![Left: an elderly fisherman with a grey beard mending a net on a wooden pier at dawn; right: a matte sage-green cup on travertine with palm-leaf shadows](https://ai2face.com/_astro/photoreal-product.CSEEsYhQ_2dYakm.jpg "A portrait with real skin texture and a product shot with hard shadows. Grok Image 2.0, 2K")

**Templates.** Alongside Image 2.0, xAI added 15 templates: photo editing, product shots, headshots, icons, game assets and emoji.

## Grok image edit: how to edit photos in Grok

Editing is what sets Image 2.0 apart from earlier versions. The model preserves everything you didn't ask to change, across generations and edits.

- **Text edits.** Upload a photo and describe the change: “make the coat red, make it a sunny day, keep the face, pose and street unchanged”.
- **Magic wand.** Point at an area and the model changes only that.
- **Segmentation** — precise selection of areas to edit.
- **Background removal** — exports the subject on a transparent background, ready for a product listing or collage.
- **Smart resize** — pick a new aspect ratio and the model fills in the rest: a vertical photo becomes a horizontal banner.
- **Multiple sources.** Up to 5 photos per generation: combine subjects, transfer a style, compose a scene.

![Left: a woman in a beige trench coat on a wet cobblestone street on an overcast day; right: the same woman in the same pose in a red trench coat under a blue sky](https://ai2face.com/_astro/edit-before-after.Bw9A-iBn_ZH93BY.jpg "One-prompt edit: the coat and weather changed, while face, hair, pose, sneakers and buildings stayed the same. Grok Image 2.0")

Combining several photos works the same way: refer to the sources in order — “the woman from the first image”, “the cup from the second”.

![Left: two source images, a woman in a beige trench coat and a green cup; right: the result, the same woman at a marble table by a cafe window holding that cup](https://ai2face.com/_astro/multi-reference.cKdvX7e-_Z17ijPm.jpg "Two sources, one frame: the woman, her trench coat and the cup are all recognizable. Grok Image 2.0")

In the API, edits are billed on both ends: you pay for the input image and the result. Other ways to change a photo with AI — in our guide on how to [edit photos with AI](/blog/izmenit-foto-nejroset).

## Grok image prompts

The formula: **subject and action → place and background → light → style or camera → constraints**. English works best for fine details and precise typography. Write any text that should appear in the image right in the prompt, in the language you want, and add “no other text”.

| Task | What to include |
|---|---|
| Photorealism | lens (35mm, 85mm), light source, “realistic skin texture” |
| Product shot | material, surface, type of light and shadow, “no logos, no text” |
| Poster, cover | exact text, its position and size, palette, “no other text” |
| Art, illustration | technique (watercolor, screen-print, 3D render), palette |
| Edit | what changes and what stays: “keep the face and pose unchanged” |

Six prompts for typical jobs:

**Profile headshot:**

```
Professional headshot of a man in his 30s in a navy sweater, soft window light from the left,
neutral light grey background, 85mm lens, shallow depth of field, realistic skin texture, no text
```

**Product photo for a listing:**

```
Product photo of white wireless earbuds in an open charging case on a smooth concrete surface,
soft diffused light, subtle reflection, minimal composition, no logos, no text
```

**Article cover or thumbnail:**

```
Wide editorial illustration of a person working on a laptop at a desk surrounded by floating
icons of charts and chat bubbles, flat vector style, teal and orange palette, no text
```

**Food:**

```
Overhead photo of a rustic breakfast table: shakshuka in a cast iron pan, fresh bread, herbs,
morning sunlight with hard shadows, linen napkin, realistic, no text
```

**Interior:**

```
Scandinavian living room with a light oak floor, a beige linen sofa, a large window with
sheer curtains, indoor plants, soft afternoon light, wide angle, photorealistic, no text
```

**Fantasy art:**

```
A lighthouse on a cliff made of giant seashells, glowing windows at dusk, stormy sea below,
painterly digital art, rich colors, cinematic composition, no text
```

More ready-made templates — in our collection of [image prompts](/blog/promty-dlya-kartinok).

## Why a picture doesn't come out right

- **Background text is gibberish.** Image 2.0 renders the main headline accurately, but signs and lettering in the background that the prompt doesn't mention sometimes fill up with nonsense characters. Fix it with “no readable text on signs” or a targeted edit.
- **Too many details.** Ten objects with attributes get mixed up. Put the main thing first and keep the rest short.
- **Clashing styles.** “Photorealistic” and “watercolor” in one prompt give a muddy result. Pick one style.
- **No light or background.** Without them the model uses neutral studio light, and the image looks flat.
- **A series looks inconsistent.** To keep one character or style across several pictures, edit with the source as a reference rather than regenerating from the same prompt.
- **Moderation refusal.** The model won't fulfil prompts that break xAI's rules.

## Grok image limit and pricing

**In the Grok apps.** There's no fixed number of images. Since June 2026 paid SuperGrok plans share one weekly pool across all products: chat, Imagine, voice. Images, and video even more so, use it faster than chat; usage by product is shown in Settings on the Usage tab. When the pool runs out you can buy Extra Usage Credits — only on the web, from $5 — or upgrade. xAI doesn't publish free-tier limits for Imagine.

**Watermark.** Images from the Grok apps carry a watermark that can't be turned off.

**In the xAI API:**

| Model | Price per image | Notes |
|---|---|---|
| grok-imagine-image-2.0 | $0.04 | 1K and 2K, low or medium quality, editing, up to 5 sources |
| grok-imagine-image | $0.02 | previous-generation fast model |

Edits are billed for both the input image and the result. Quality defaults to `auto`: low for generation, medium for editing.

## Grok Image vs other AI image generators

| | Imagine Image 2.0 | [GPT Image 2.5](/models/gpt-image-2-5) | [Nano Banana 2](/models/nano-banana-2) | [Nano Banana Pro](/models/nano-banana-pro) |
|---|---|---|---|---|
| Arena rank (Sep 24, 2026) | 6 | 1 | 9 | 14 |
| In-place editing | magic wand, segmentation | by mask | by prompt | by prompt |
| Background removal | yes | transparent background | — | — |
| References | up to 5 | several | up to 14 | up to 14 |
| API price | $0.04 | per token | from $0.045 | from $0.134 |

Grok Image 2.0 ranks above Nano Banana 2 and Pro and is handier for targeted edits and cutouts. For the most accurate text, GPT Image 2.5 leads; for complex scenes with a dozen references, use [Nano Banana](/models/nano-banana). For a detailed comparison of generators, see our overview of [AI image generators](/blog/nejroseti-dlya-izobrazhenij).

## Grok Image in AI2Face

SuperGrok is a separate subscription with a weekly cap. In AI2Face, Grok Image is available right in the generator:

- **no SuperGrok subscription** — sign in with email or Google;
- **pay by card, crypto or another convenient method** — the cost of each generation is shown before you run it;
- **credits don't expire** — besides a subscription you can buy a credit pack, and the balance carries over to the next month;
- **no AI tags in the file** — AI2Face strips the AI-generation metadata from the finished image and writes regular camera EXIF instead, so social networks won't see an “AI-generated” label in the file;
- **every model in one subscription** — Grok Image, [GPT Image](/models/gpt-image), Nano Banana and video models: better value than separate subscriptions, and easy to compare pictures on the same prompt;
- **welcome credits after sign-up** — enough for your first pictures.

## Bottom line

Use Grok Image 2.0 for posters and covers with text, product photos and, above all, editing: swap a detail, remove the background, change the aspect ratio, build a scene from several photos — without losing the face or composition. It's in the top ten and ranks above Nano Banana. If you need a paragraph of small text with zero mistakes, compare with GPT Image 2.5; if you need a dozen references in one scene, try Nano Banana Pro. A finished Grok picture can be animated right away — see our [Grok video](/blog/grok-video) guide.

## FAQ

### Can Grok generate images?

Yes. Grok Imagine handles images in Grok: today that's the Imagine Image 2.0 model, the Quality mode on grok.com/imagine and in the apps. You can generate both in regular Grok chat and in the Imagine section.

### How do I edit a photo in Grok?

Upload the photo and describe the change in text; for a targeted edit, select the area with the magic wand. Image 2.0 keeps everything you didn't ask to change, can remove the background and can combine up to 5 source photos into one frame.

### What is the Grok image limit?

xAI doesn't state a fixed number of images. Since June 2026, paid plans share one weekly pool across all Grok products, and images and video use it faster than chat. Your current usage is shown in Settings, on the Usage tab.

### What resolution does Grok Image support?

1K or 2K. There are 16 aspect ratios, from square to panoramic 21:9 and 5:2. Through the API you can get up to 10 images per request.

### Can I use Grok Image for free?

You can try Grok for free, but xAI doesn't publish free-tier limits for Imagine. AI2Face gives welcome credits after sign-up, enough for your first Grok Image pictures.

### Do Grok images have a watermark?

In the Grok apps, yes: generated images and videos carry a Grok watermark, and there's no setting to turn it off.

## Sources

- https://x.ai/news/grok-imagine-image-2
- https://x.ai/news/grok-image-generation-release
- https://docs.x.ai/developers/model-capabilities/images/generation
- https://docs.x.ai/developers/model-capabilities/imagine
- https://docs.x.ai/developers/models
- https://docs.x.ai/grok/faq
- https://arena.ai/leaderboard/text-to-image