OpenAI's advanced image generator now available via API
What's the story
OpenAI has extended the availability of its improved image generation capabilities in ChatGPT, to developers via its API.
The feature, which was first made available to most ChatGPT users in late March, drew a lot of attention for generating realistic Ghibli-style photos.
Its launch led to a spike in new ChatGPT sign-ups and heavily stressed OpenAI's infrastructure.
User engagement
ChatGPT users create over 700 million images
In its first week of availability, the new image generator feature was used by over 130 million ChatGPT users who created over 700 million images between them.
The high level of user engagement highlights the demand and interest in advanced image generation technology.
The surge in usage also underscores the challenges faced by OpenAI in managing such a rapid increase in demand for its services.
AI model
gpt-image-1: The driving force behind image generation
The image generation capability in OpenAI's API is powered by an AI model called "gpt-image-1."
This multimodal model can create images in different styles, follow custom guidelines, use global knowledge, and render text.
Developers using gpt-image-1 can generate multiple images at once and control the quality of generation which directly affects speed.
Safety measures
gpt-image-1 adheres to OpenAI's safety protocols
OpenAI has promised gpt-image-1 adheres to the same safety protocols as ChatGPT's image generation.
These include safeguards that prevent the model from generating content that violates company policies.
Developers can tweak moderation sensitivity, opting for "auto" for standard filtering or "low" for less restrictive filtering.
The company also claims all images generated with gpt-image-1 are watermarked with C2PA metadata, making them identifiable as AI-generated by supported platforms and apps.
Cost
Pricing details and current usage
OpenAI's pricing for gpt-image-1 is set at $5/million input tokens for text, $10/million input tokens for images, and $40/million output tokens.
This roughly amounts to 2¢, 7¢, and 19¢ per generated image for low-, medium-, and high-quality square images, respectively.
Companies such as Adobe, Airtable, Wix, Instacart, GoDaddy, Canva, and Figma are already leveraging or testing gpt-image-1 in their platforms.