Grok Imagine Image 2.0: Spot Editing, Sharp Text & 5 Inputs
The magic wand edits only what you point at, backgrounds export transparent, and series hold one style. Ready-to-use prompts and four rules that save any generation.
👁 117 reads
Grok Imagine Image 2.0 is the first generator that can plan typography and layout like a designer, and small text comes out sharp. Add spot editing, background removal with transparency, and up to five input images at once. We’ve collected concrete prompts you can paste and test in five minutes.
Verdict in One Paragraph
This is the first image generator built for work, not for “wow” effect. You can see the difference in three things: it holds small text on a poster, it edits exactly the fragment you point at, and it maintains the same style through a series of images. The maker claims second place worldwide in text-to-image and editing according to the Arena ranking, as of August 7, 2026. For whom: posters, product photos, icons, game assets. Not for whom: if you’re looking for one pretty picture - anything will do for that.
Four Editing Tools That Change How You Work
| Tool | What It Does | When to Use It |
|---|---|---|
| Magic Wand | Changes the indicated area, rest stays untouched | When 90% of the image is good but one element isn’t |
| Segmentation | Marks precise areas for change | Changing the color of one object in a photo |
| Background Removal | Export with transparency | Product shot for a store, cutout for compositing |
| Multiple References | Up to 5 input images in one generation | Instead of manual stitching in graphics software |
The most important of these four is the magic wand. The previous way of working with generators looked like this: the image is 90% good, so you generate from scratch and lose that 90%. Spot editing eliminates this loop.
Ready-to-Use Prompts to Test
1. Poster with Small Text - The Test That Exposes Generators
Small text has always been a weakness of generators. This prompt checks whether claims have substance:
Synthwave band concert poster. Large retro typography at top, below a list of seven dates and cities in small print, label logo at bottom. Sharp small text, dark background, neon magenta and cyan. Vertical 2:3 format.
What to look for: whether seven lines of small print are legible, whether letters don’t fall apart, and whether the layout holds hierarchy. This is a test where most generators produce gibberish pretending to be text.
2. Product Photo with Transparent Background
Product photo: matte black cosmetic bottle on pure white background, soft studio light from left side, delicate shadow underneath, focus on label. No props, no hands, no text.
Then use background removal and export with transparency. You have a ready product shot for a store without a photographer.
3. Series in One Style - The Most Underrated Feature
The maker calls this world-building: a character, their locations, and objects generated separately but holding one style. It looks like this:
Character: woman in raincoat, 1970s, grainy analog photography, muted palette, cold light.
Location: empty bus station at night, same style: grainy analog photography, muted palette, cold light.
Object: leather suitcase on bench, same style: grainy analog photography, muted palette, cold light.
The secret is repeating the style description word for word in each prompt. Not “same style as before” - literally the same phrase. This works in all generators, not just this one.
4. App Icons
Set of line icons: envelope, calendar, clock, magnifying glass. Uniform line thickness 2 px, rounded ends, square grid, no fills, single color. On transparent background.
Nine Formats from One Image
Intelligent resizing extends the frame instead of cropping it. Available ratios: 1:2, 9:16, 2:3, 3:4, 1:1, 4:3, 3:2, 16:9, 2:1.
This is an opinion: this is a feature that genuinely saves time when publishing. One image for a horizontal post cover, vertical reel, and square for social media, without three separate generations and without cropping people’s heads when framing.
Fifteen Ready Templates
Templates pack typical tasks into a ready starting point - you supply input, you get a finished result:
- Photos and photography: photo editing, photo transformation, collage, background removal and change, profile photo
- Product and marketing: product color change, product poster, e-commerce photos, user-style photos, merchandise
- Design: mascot, icon generator
- Games and streaming: character sprite, prop and UI set, emoji generator
Via API
The model is called grok-imagine-image-2.0. The call is short:
response = client.image.sample(prompt="...", model="grok-imagine-image-2.0")
It returns an image URL. That’s enough to plug generation into an existing script.
Four Rules That Save Any Prompt
These are our rules from working with generators, not from the maker’s documentation:
- One object, lots of empty background. A composition with three things at once almost always comes out worse than with one.
- Describe light, not mood. “Soft light from the left, one long shadow” works. “Atmospheric” means nothing.
- List what you don’t want. “No text, no people, no logo” removes most typical mishaps.
- Don’t write prompts in languages with diacritics if the image should contain text. Special characters in generated typography fall apart in almost every model.
Most Common Questions
Where Can I Use This?
In quality mode at grok.com/imagine and in iOS and Android apps. Via API as grok-imagine-image-2.0.
Is This Better Than Other Generators?
The maker claims second place worldwide in both Arena rankings, as of August 7, 2026. This is their data, which we haven’t verified - but the advantage in small text and spot editing is verifiable in five minutes with your own prompt.
How Many Input Images Can I Provide?
Up to five in one generation. This replaces manual stitching in graphics software.
Can You Get a Transparent Background?
Yes, background removal exports the object with transparency - ready to insert into other work.
How Do You Maintain One Style in a Series?
Repeat the style description literally in each prompt, word for word. References like “as before” don’t work.
Sources and Verification Date
Features, template list, ratios, API model name, and ranking position read September 21, 2026 from the official announcement of August 7, 2026: x.ai/news/grok-imagine-image-2. The Arena ranking position is the maker’s claim, not our measurement. Example prompts and the four rules at the end are ours - they come from working with image generators, not from the maker’s materials.
More About Grok 4.7
- Guide: full benchmark and pricing table ·
- Grok 4.7 vs Fable 5.1 vs GPT-5.6 Sol: we calculated how much a point of advantage costs ·
- How much does Grok 4.7 cost? Full pricing: text, voice, image and video ·
- 500,000 tokens of context in Grok 4.7: what really fits in there
More on this from QUANTSLANT
guides·Grok 4.7 in Cursor: How to Enable and When to Switch ModelsA day of coding for $3.20 instead of $20. Decision table, three test tasks, and four tips that improve quality regardless of which model you choose.