# GPT-image-2

OpenAI's GPT-image-2: strong at following complex instructions, text and designed layouts.

Last updated: 2026-10-05

**Key facts**

- GPT-image-2 is OpenAI's image model from the generation before GPT Image 2.5, strong at following long, structured instructions and rendering text in designed layouts.
- GPT-image-2 offers resolution 1K, 2K or 4K and quality low, medium or high in Sloosh; the GPT Image 2.5 models add xhigh and max.
- GPT-image-2 takes up to 16 reference images in Sloosh Studio and prompts up to 32,000 characters.
- GPT-image-2 suits designed pieces with several blocks of text, such as recipe cards, menus and posters.

Web page: https://sloosh.ai/docs/gpt-image-2

GPT-image-2 is the generation before GPT Image 2.5. It follows long, structured instructions well and renders text cleanly, which suits designed pieces such as cards, menus and posters.

## When to use it

- Designed layouts with several pieces of text: recipe cards, menus, posters.
- Iterating on one image over several edits.
- For new work, compare it with [GPT Image 2.5 Flare](https://sloosh.ai/docs/gpt-image-2-5-flare.md), which is faster.

## How to prompt it

- Lay the prompt out the way the image is laid out: top to bottom, with each block's text in quotes.
- Numbered steps and short lines render more reliably than long paragraphs of copy.
- Name the fonts by look ("bold rounded", "simple sans-serif") and give the colours.
- Quality goes up to high. Use high for small text.

[Prompting GPT Image](https://sloosh.ai/docs/prompting-gpt-image.md) has the full guide.

## Example

A recipe card with a title and three numbered steps, all from the prompt.

![A recipe card for an iced matcha latte: a photo of the drink, the title ICED MATCHA LATTE and three numbered steps](https://sloosh.ai/docs/m-gpt-image-2.webp)

_2:3, quality high, Enhance off._

**Prompt**

```text
A clean, modern recipe card for an iced matcha latte, for a cafe's Instagram. Top: a photo of the drink in a tall glass with ice and a green-and-white swirl. Below it, the title "ICED MATCHA LATTE" in a bold rounded font. Then three numbered steps in a simple sans-serif:
1. Whisk 2 tsp matcha with 60 ml hot water.
2. Fill a glass with ice and 200 ml oat milk.
3. Pour the matcha over the top.
Soft cream background, matcha-green accents, plenty of white space.
```

## Settings in Sloosh

- Aspect ratio: Auto, 1:1, 3:2, 2:3, 16:9, 9:16, 4:5 or 5:4.
- Resolution: 1K, 2K or 4K.
- Quality: low, medium or high.
- At 2K or 4K, an Auto aspect ratio renders square.
- References: up to 16 images.
- Prompt: up to 32,000 characters.

Open it in [Image](https://sloosh.ai/studio/image) and pick GPT-image-2 in the model chip. The Generate button shows the cost first.

## Frequently asked questions

### What is GPT-image-2 best for?

GPT-image-2 is best for designed layouts with several pieces of text, such as recipe cards, menus and posters, and for refining one image over several edits. It follows long, structured prompts well. For new everyday work, compare it with [GPT Image 2.5 Flare](https://sloosh.ai/docs/gpt-image-2-5-flare.md).

### GPT-image-2 vs GPT Image 2.5 Flare: which should I use?

Try GPT Image 2.5 Flare first for new work: OpenAI says it is up to 50% faster than GPT Image 2 at comparable quality, and it adds xhigh and max quality. Keep GPT-image-2 when a layout prompt already works well on it. Both take the same prompts and 16 references in Sloosh.

### How do I make a menu or recipe card with GPT-image-2?

Lay the prompt out the way the image is laid out, top to bottom, with each block's text in quotes. Use numbered steps and short lines rather than long paragraphs, name fonts by look such as "bold rounded", give the colours, and set GPT-image-2's quality to high for small text.

### What quality setting should I use for small text with GPT-image-2?

Use high. GPT-image-2 offers Auto, low, medium and high in Sloosh, and high is the setting for small or dense text. Higher quality is slower, so draft on Auto or medium and switch GPT-image-2 to high for the final.

## Sources

- [OpenAI cookbook: image model prompting](https://developers.openai.com/cookbook/examples/multimodal/image-gen-models-prompting-guide)