Loading model details…
A production-focused image generation and editing model with strong instruction following, typography, local edits, multi-image composition, and visual consistency.
Grok
| Item | Specification |
|---|---|
| Input | Text, image |
| Output | Image |
| Modes | Text-to-image, single-image editing, multi-image editing |
| Resolution | 1K, 2K |
| Quality Levels | low, medium, auto |
| Model Code | grok-imagine-image-2.0 |
Grok Imagine Image 2.0 is a generally available image generation and editing model released by SpaceXAI on August 7, 2026 under the API code grok-imagine-image-2.0. It supports text-to-image, natural-language editing, and multi-image composition with controls for aspect ratio, 1K or 2K resolution, quality, and the number of outputs.
The model targets visual assets suitable for real projects, improving complex instruction following, typography, small-text clarity, composition, and content preservation during edits. One edit can combine up to five source images across photography, design, and illustration. Selection and segmentation workflows can change a specified area while preserving the rest of the image.
| Feature | Benefit |
|---|---|
| 1K and 2K Output | Supports fast drafts and higher-detail production assets. |
| Up to Five Source Images | Enables complex composition, multiple subjects, prop references, and style alignment. |
| Typography and Small Text | Serves posters, social assets, infographics, and text-heavy designs. |
| Precise Regional Editing | Changes selected elements while reducing unintended edits elsewhere. |
| Broad Aspect-Ratio Control | Covers square, landscape, portrait, photography, mobile, and ultrawide formats. |
| Batch and Deferred Generation | Requests multiple variations and supports asynchronous handling for longer jobs. |
It supports text-to-image, single-image editing, and multi-image editing with up to five sources. It fits photography, illustration, graphic design, product scenes, character assets, and text-bearing visuals.
The official generation guide lists 1K and 2K resolutions with low, medium, or auto quality. Because auto lets the service select the active level, specify one explicitly when repeatable output characteristics matter.
Provide the exact copy, hierarchy, type style, position, and alignment, and avoid too many competing text regions. Review spelling and small text after the first result, then use local editing for corrections.
Name the target object clearly and use selection or segmentation tools where the product interface exposes them. Also state which content must remain unchanged to reduce drift outside the target area.
The current official capability accepts up to five sources. Assign a clear role to each image—identity, clothing, prop, environment, or style—to reduce ambiguity.
Lock the aspect ratio, quality level, character description, palette, camera language, and style terms, then continue from approved images through editing or references. A shared prompt template is more reliable than redescribing every asset from scratch.
Grok Imagine Image 2.0 is available on LinkModel. Use the public model ID and the generation or editing endpoint, aspect ratio, resolution, quality, and image-input fields shown on LinkModel's live page.
Test text generation, single-image editing, multi-image composition, typography, multiple aspect ratios, and 1K or 2K output, and inspect result formats and errors. Do not assume native batch size, five-image input, quality levels, or deferred fields pass through unchanged.
| Resolution | Generation quality | Input image | Output image |
|---|---|---|---|
| 1K | low | $0.01 / image | $0.04 / image |
| 1K | medium | $0.01 / image | $0.06 / image |
| 2K | low | $0.01 / image | $0.06 / image |
| 2K | medium | $0.01 / image | $0.08 / image |
Estimate = requests × (reference images × input unit price + output images × output unit price). Image counts apply to each request. Published component rates are used; account discounts and actual usage may change the final charge.