OpenAI has moved its upgraded image generation technology beyond ChatGPT and into the API, giving developers a way to integrate the same broader capability into their own apps and services.
The API version is powered by gpt-image-1, a natively multimodal model that can generate images in different styles, follow custom guidelines, use world knowledge, and render text. For developers, the launch turns a consumer-facing feature into an application-building tool.
What OpenAI is making available
The new API capability is based on the image generation feature that launched for most ChatGPT users in late March. That ChatGPT feature quickly drew attention for producing realistic Ghibli-style photos and AI action figures, becoming one of the company’s most visible recent product moments.
The response was large enough to create both growth and operational pressure for OpenAI. According to the company, over 130 million ChatGPT users created more than 700 million images in just the first week of the tool’s availability.
Now, the underlying technology is available to developers through gpt-image-1. Instead of using image generation only inside ChatGPT, companies can connect it to their own workflows, interfaces, and products through OpenAI’s API.
What gpt-image-1 can do
OpenAI describes gpt-image-1 as a natively multimodal model. In practical terms, that means the model is designed to work across more than one type of input or output, rather than treating image generation as a narrow add-on.
The model can create images across different styles, follow custom guidelines, leverage world knowledge, and render text. Those capabilities matter because app developers often need more than a single prompt-to-image flow. They may need outputs that match product rules, brand requirements, or a specific user task.
Developers can also generate multiple images at a time. That supports use cases where a user may need options, variations, drafts, or several related visuals from the same interaction.
OpenAI is also giving developers control over generation quality. That quality setting affects speed, so developers can make tradeoffs depending on the experience they are building. A fast interactive feature may need different settings than a workflow where higher-quality output is more important.
Safety, moderation, and image identification
OpenAI says gpt-image-1 uses the same safety guardrails as image generation in ChatGPT. Those safeguards are intended to restrict the model from creating content that violates the company’s policies.
Developers can adjust moderation sensitivity. The setting can be set to auto for standard filtering or low for less restrictive filtering. According to OpenAI documentation provided to TechCrunch, low filtering limits fewer categories of potentially age-inappropriate content.
That gives developers some control over how strict filtering should be in their own products, while still keeping the model within OpenAI’s policy framework. The moderation setting is therefore part of the product design surface, not just a background technical option.
OpenAI also says all images created with gpt-image-1 are watermarked with C2PA metadata. The purpose is to let supported platforms and apps identify those images as AI-generated. That metadata is an important part of how generated media can be labeled or recognized after it leaves the original app.
How pricing works
OpenAI has published pricing for gpt-image-1 in token terms. Tokens are the raw bits of data the model processes, and the price depends on whether the input is text or image data, and whether the output is an image.
The listed API pricing is:
- $5 per million input tokens for text
- $10 per million input tokens for images
- $40 per million output tokens for images
OpenAI says that translates to around 2 cents, 7 cents, and 19 cents per generated image for low-, medium-, and high-quality square images, respectively.
For developers, that makes image quality a product and cost decision at the same time. Higher-quality image generation costs more per image, while lower-quality generation is cheaper and faster.
Who is already using it
OpenAI says companies including Adobe, Airtable, Wix, Instacart, GoDaddy, Canva, and Figma are already using or experimenting with gpt-image-1.
Two examples show the range of possible use cases. Figma Design now lets users generate and edit images through gpt-image-1. Instacart is testing the model for images for recipes and shopping lists.
Those examples point to a broader shift in how image generation may appear inside software. Rather than sitting in a standalone creative tool, it can become part of design, commerce, productivity, and planning workflows where users already work.
The API release also changes who can build with OpenAI image generation. ChatGPT users can create images inside OpenAI’s own product, while developers can now embed gpt-image-1 into services that have their own users, interfaces, constraints, and business models.
The central development is simple: OpenAI’s upgraded image generator is no longer limited to ChatGPT. With gpt-image-1, developers now have access to image creation, editing-oriented workflows, quality controls, moderation settings, C2PA metadata, and published pricing through the API.