OpenAI on Wednesday expanded access to its latest image generation technology, making the model known as “gpt-image-1” available through its application programming interface (API).
This move allows developers to embed the GPT-4o-based image creation and editing tools, previously rolled out within ChatGPT in late March, into their own applications and services. The API furnishes capabilities for producing photorealistic visuals, leveraging world knowledge, following custom guidelines, controlling styles, rendering text within images, and performing interactive adjustments.
According to OpenAI’s initial announcement about the underlying technology, “4o image generation is a new, significantly more capable image generation approach than our earlier DALL·E 3 series of models. It can create photorealistic output. It can take images as inputs and transform them.”
Developers utilizing the API can generate images at various resolutions, including 1024×1024, 1024×1536 (portrait), and 1535×1024 (landscape). Output options include JPEG or WEBP formats, along with support for transparency.
The model operates multimodally, processing text and images provided in sequence, a capability OpenAI describes on its blog by stating “gpt-image-1 is natively multimodal… It understands text and images in arbitrary sequence.”
This enables complex editing tasks, such as modifying existing images based on both the original picture and a new text prompt, or using ‘masking’ techniques to alter specific regions while leaving others untouched – features enhancing creative control.
OpenAI released gpt-image-1, a multimodal image generation model, now available via the Images API (with support in the Responses API coming soon), allowing developers and businesses to easily integrate high-quality image creation into apps and tools, with usage priced per token… pic.twitter.com/hdr1MYrLr2
— Tibor Blaho (@btibor91) April 23, 2025
Content Provenance and Safety Protocols
Addressing concerns about the origin and potential misuse of AI-generated visuals, OpenAI is embedding C2PA (Coalition for Content Provenance and Authenticity) metadata into every image created via the gpt-image-1 API.
C2PA is an open standard that allows creators to attach tamper-evident information about the content’s origin and editing history. This digital watermarking standard aims to identify content as AI-generated on platforms that support it. OpenAI’s use of C2PA began in early 2024 with DALL·E 3, and the company reinforced its commitment by joining the C2PA Steering Committee in May 2024.
However, the effectiveness of metadata relies on platform support and can be compromised by simple manipulations like cropping or screenshots, a limitation acknowledged within the industry and highlighted by research into watermarking vulnerabilities.
Beyond watermarking, the API includes content moderation filters designed to block requests that breach OpenAI’s policies, with developers able to choose between standard (“auto”) or less restrictive (“low”) sensitivity settings.
OpenAI also confirmed that customer prompts and images submitted through the API are not used to train its models, addressing a key data privacy concern for developers. The company maintains its policy against directly imitating the styles of living artists.
This approach aligns with broader industry efforts, including support from OpenAI, Microsoft, and Adobe for legislation like California’s AB 3211 bill proposing mandatory AI content labeling.
Developer Access, Performance, and Adoption
Accessing gpt-image-1 via the API involves costs based on token usage: $5 per million for input text tokens, $10 per million for input image tokens, and $40 per million for output image tokens. This translates to roughly 2 to 19 cents per generated image, depending on the chosen quality level. Some organizations might need to complete an identification verification process with OpenAI before gaining API access.
While powerful, the model isn’t instantaneous; complex prompts might take up to two minutes to process. Furthermore, while text rendering is improved over previous DALL-E versions, OpenAI documentation notes potential inconsistencies in precise text placement and maintaining visual coherence for elements like characters or logos across multiple generations.
The gpt-image-1 model is accessible through both OpenAI’s direct API and Microsoft’s Azure OpenAI Service, specifically via the Azure AI Foundry Image Playground. The Azure platform adds its own specific safety layers like content safety checks and abuse monitoring on top of OpenAI’s baseline measures. Developers testing the technology should note that using OpenAI’s own web-based image playground still incurs API usage costs.
Early Use Cases and Context
The API release follows substantial user engagement when the tools were first integrated into ChatGPT. OpenAI reported that over 130 million ChatGPT users generated more than 700 million images within the first week of the feature’s availability, initially gaining attention for producing Ghibli-style photos and AI action figures.
By opening up API access, OpenAI enables developers to build applications that directly compete with or supplement existing AI image tools from companies like Midjourney, Adobe (Firefly), and Stability AI.
Several companies, including Adobe, Airtable, Canva, Figma, GoDaddy, Instacart, and Wix, were named by OpenAI as already experimenting with or integrating the gpt-image-1 API. Examples cited include Figma embedding the tools into its design platform and Instacart testing image generation for visual aids in recipes and shopping lists.
