OpenAI has announced the launch of image generation capabilities based on the “gpt-image-1” model via API, allowing developers to integrate the feature into their own apps and services. This same model is already available in ChatGPT and has gained popularity for creating Studio Ghibli-style images and various “AI action figures.” According to the company, over one hundred thirty million ChatGPT users created more than seven hundred million images in just the first week of this feature’s launch.
The “gpt-image-1” model enables the creation of images in different styles, following given instructions, using world knowledge, and accurately reproducing text. Developers can generate multiple images at once, choose the quality, and thus the speed of results. The model uses the same safety measures as ChatGPT, including protection against generating unwanted content. Moderation sensitivity can be adjusted — a standard filter or a less strict mode for a limited range of content categories.
All images created with this model contain watermarks in the form of C2PA metadata, allowing platforms and apps to identify that the image was generated by AI. The cost of using the model is five dollars per million input tokens for text, ten dollars for images, and forty dollars per million output tokens for images. According to OpenAI’s calculations, this is approximately two, seven, and nineteen cents per image of low, medium, and high quality, respectively.
Already, companies like Adobe, Figma, Canva, Wix, Instacart, and GoDaddy are using or testing the integration of “gpt-image-1” in their products. For example, in Figma Design, users can generate and edit images using simple prompts, change styles, add or remove objects, expand backgrounds, and more. Adobe provides access to this feature in its Firefly and Express apps, allowing users to experiment with different styles to create creative ideas.
