Image GenerationActive
openai / gpt-image-2
gpt-image-2
Generate or edit images with GPT Image 2 from OpenAI. It delivers strong instruction following, sharp text rendering, and flexible sizing up to 4K.
Text to ImageImage to ImageImage Editing
Model ID
gpt-image-2
Provider
openai
Runs
123296
Updated
1778677509
wiro playground—openai/gpt-image-2
Updated 1778677509
## Overview
GPT Image 2 is OpenAI’s image generation and image editing model. It turns a written description into 1 or more finished images. You can also upload reference images and ask for targeted edits, with optional masking. It’s useful when you need readable on-image text, accurate layouts, and high-fidelity edits.
## What you can build
- Product photos with accurate labels, packaging text, and consistent lighting
- Marketing creatives with dense typography like posters, flyers, and social ads
- UI mockups and app screens with small text and structured layouts
- Image editing tools for object removal, background changes, and retouching
- Style transfer and composition from multiple reference images
- Localized image variants by replacing text while keeping the design
## Inputs
- A required written description of the image you want to generate, or the edit you want applied. Wiro accepts up to 32,000 characters.
- Optional reference images to guide the result or to edit directly. Wiro supports up to 16 images per request.
- Optional mask image for edit jobs. Use it to limit where changes can happen. The mask should match the first input image dimensions.
- Output size selection. You can let the model pick automatically, or choose a fixed aspect ratio such as square, landscape, or portrait.
- A quality level selection that trades off draft output versus final output.
- Background handling. You can keep it automatic, or force an opaque background.
- Output file format choice: PNG, JPEG, or WebP.
- Optional compression control for JPEG or WebP outputs, on a 0 to 100 scale.
- How many images to return in one request. Wiro supports 1 to 10 images.
- Moderation strictness selection to control how aggressively content gets filtered.
## Outputs
- One or more generated images in the format you selected (PNG, JPEG, or WebP).
- For each image, you get a downloadable file plus basic metadata such as width, height, and MIME type.
## Recommended settings
- Drafting and fast iteration: use low quality and a standard square or auto size.
- Final assets with readable text: use high quality and a fixed size that matches your layout needs.
- Edits that must preserve details: keep the output size on auto when editing, unless you need a resize.
- JPEG or WebP for smaller files: set a compression value below 100 if you can accept artifacts.
## Limitations
- Transparent backgrounds aren’t supported for GPT Image 2. Use an opaque background.
- Very large outputs are supported up to 4K, but sizes above typical 2K outputs are considered experimental.
- Masks guide edits, but the edited region may not follow the mask shape perfectly.
- Character and brand consistency can still drift across separate generations.
- Precise layout placement can fail in grid-heavy or pixel-perfect compositions.
- Low-quality reference images, heavy JPEG artifacts, screenshots with noise, or inconsistent lighting can cause unwanted changes during edits.
## Safety & compliance
- OpenAI filters prompts and generated images under its image safety policies.
- Use the moderation control to choose standard filtering or less restrictive filtering.
- Don’t use this model to generate disallowed content, including harmful or exploitative imagery.
- Only upload images you have rights to use, especially for people, logos, and branded designs.
Example prompts
Great starting points for gpt-image-2.
A hyper-detailed, high-resolution photographic edit based on image. The clear rocks glass, the metallic spoon, and the rich, dark mahogany wooden table with its complex grain and precise geometric shafts of warm sunlight must all remain in their exact positions and forms. Inside the glass, the simple iced coffee is replaced by a perfectly layered Café Bombón with complex, high-clarity details. A thick, smooth layer of sweetened condensed milk forms the base, topped by a precise layer of dark, rich espresso, and finally an elegant cap of airy, micro-foamed milk. The simple ice cubes are replaced by large, geometrically precise, slow-melting artisan ice spheres showing internal crystal structures and fractures, all glistening with heavy, detailed condensation on the glass. The metallic spoon is highly polished to a mirror-like finish, reflecting the layers of the drink and the wood. The dust motes floating in the sunbeams are rendered with extreme fidelity as delicate, golden, complex particles. The entire scene is polished to perfection, with the contrast between foam, liquid, glass, and wood grain maximized. The lighting is intensified, making the glass and its contents glow from within, all while maintaining the exact existing shadow patterns.Image Generation
A detailed, realistic high-resolution photograph of the building facade from image, viewed from the same low-angle perspective. The terracotta-ochre color of the building is preserved, but aged, showing minor imperfections, weathered plaster, and patches of peeling paint. The original grid of windows and ornate wrought iron balconies are retained in their positions. However, multiple clotheslines with actual laundry (towels, linen shirts, socks) are strung between the balconies on all levels, adding life. The potted plants on the middle balcony are more diverse (geraniums, lavender, jasmine). An older woman with a patterned headscarf is partially visible, tending to plants on one of the larger balconies on the second level from the bottom. The ground level, slightly more visible, now shows a parked vintage bicycle, a small cafe chair, and some scattered cafe menus on a worn wooden table. The lighting is shifted from harsh daylight to a warm, soft, late afternoon golden hour, with long, diffused shadows across the facade. The sky is a gradient of deep blue to warm orange. The overall atmosphere is bustling, authentic, and lived-in. All textures (iron, wood, stucco) are highly detailed.Image Generation
A detailed, realistic photograph of the telecommunications tower from image, viewed from the same perspective. The clear, gradient sky is completely replaced by a dense, pervasive layer of cool, blue-grey morning fog that shrouds the entire scene. The detailed lattice structure of the tower is still visible, but softened and partially obscured, silhouetted against a ghost-like, heavily diffused disc of the rising sun. Moisture glistens on the metal poles of the street lights and on the leaves of the palm trees and other foliage at the base. The street lights are on, casting soft, diffused pools of light. The large foreground street light is prominent but soft in the fog. The air is thick, and a few detailed, moist-looking cobwebs are visible on the lower lattice. The overall feeling is vast, cool, and still. The texture of damp metal and wet foliage is paramount.Image Generation
A hyper-detailed photograph capturing a perfect infinity loop of ancient, moss-covered library bookshelves. A spiral staircase made of dark, polished mahogany descends within the loop, leading toward a solitary vintage globe at the absolute geometric center. A single window, situated outside the entire structure on the far-left wall, casts a defined, warm light only onto the globe, while the bookshelves remain in cool, diffused shadow.Image Generation
An oil painting in the style of Salvador Dalí depicting \"The Bureaucracy of Time.\" The scene is a vast, open desert where the sand is entirely composed of miniature, ancient wristwatches. Large, melted clocks are not dripping, but are instead solidifying into a massive, architectural filing cabinet. A man made entirely of flowing water stands before the cabinet, hopelessly trying to organize a stack of papers that are rapidly disintegrating into desert sand.Image Generation
API quick start
Run gpt-image-2 with a single API call.
POST https://api.wiro.ai/v1/Run/openai/gpt-image-2
{
"prompt": "A hyper-detailed, high-resolution photogr…",
"resolution": "1k",
"ratio": "1:1",
"quality": "low"
}