- OpenAI has added its image generation model (gpt-image-1) to its API
- The API allows businesses to integrate ChatGPT's image generation into their own tools
- Pricing is based on tokens: $5 per million text tokens, $10 per million image input tokens, $40 per million image output tokens
- Major companies like Canva, GoDaddy, and Airtable are already testing the integration
- Images include C2PA metadata to identify AI-generated content
- OpenAI promises not to use customer API data to train its models
In a big move that will change how businesses use AI pictures, OpenAI just made ChatGPT's popular image maker available through their API. This means companies can now put those cool AI pictures right into their own tools without sending users somewhere else. The announcement came on April 23, 2025, and it's already got lots of tech folks excited about what they can build.
What Is the New Image Generation API?
The new API gives access to "gpt-image-1," which is the model that powers ChatGPT's image generation. It lets developers make high-quality, professional images directly in their apps or websites. You might've seen those Studio Ghibli-style pictures floating around social media - that's this same technology!
OpenAI says the model is super flexible. It can make pictures in different styles, follow specific instructions, use general knowledge, and even write text correctly in images. This makes it useful for all kinds of businesses and projects.
The image generator has been a huge hit since OpenAI added it to ChatGPT. More than 130 million people have used it, making 700 million pictures in just the first week! That popularity even had OpenAI's CEO Sam Altman joking that their "GPUs are melting" from all the use.
How Does the Pricing Work?
The pricing for the API works differently than the regular ChatGPT service. OpenAI splits the cost into three parts:
- Text input tokens (your prompts): $5 per million tokens
- Image input tokens: $10 per million tokens
- Image output tokens (the pictures): $40 per million tokens
This is pretty expensive compared to some competitors. Stability AI uses credits that cost about $0.01 each, with their best model costing eight credits per image (about $0.08). Google's Imagen model charges paying users $0.03 per image through the Gemini API.
The higher price might be because OpenAI thinks their model makes better images or because they need to cover the huge computing costs - those "melting GPUs" Altman mentioned don't come cheap!
How Are Businesses Using It?
Some big companies are already testing the new API:
- Canva is looking at ways to add it to their Canva AI and Magic Studio Tools
- GoDaddy is testing it to help customers create logos
- Airtable is using it to help marketing teams manage visual content
These early adopters show how the API can fit into different workflows. For Canva, it's about enhancing their design tools. For GoDaddy, it helps small businesses create professional branding. And for Airtable, it streamlines how teams work with images.
The main advantage is keeping everything in one place. Before this, a business might have needed separate tools for text and images. Now they can generate both with the same API calls, making their apps simpler and more powerful.
Safety Features and Image Metadata
OpenAI isn't just throwing this tool out there without thinking about the problems it might cause. They've added the same safety rules that ChatGPT has to stop people from making harmful or inappropriate images.
One cool thing they've done is add special metadata to every image. This metadata follows the C2PA standard (that's the Coalition for Content Provenance and Authenticity). This invisible marker shows that AI made the image and tracks who owns it. This helps stop people from passing off AI art as human-made or using it in misleading ways.
OpenAI sits on the steering committee for C2PA, working with other tech companies to set standards for tracking digital content. With more and more fake images showing up online, these kinds of markers are getting really important.
Businesses can also adjust the content filtering to match their brand's needs. This means a children's app could set stricter rules than a creative design tool for adults.
Data Privacy Promises
In a move that should make privacy-minded users happy, OpenAI has promised not to use any customer data from the API to train future models. This includes both the prompts people send and the images that get made.
This is different from how some other AI services work. Many companies collect user data to improve their models, which makes some people and businesses nervous about their ideas getting copied or leaked.
The promise means businesses can use the API without worrying that their unique prompts or brand assets will end up training OpenAI's next model. This could be especially important for companies working on unique visual styles or creating images of products that haven't been released yet.
How This Compares to Previous OpenAI Image Tools
Before this API launch, OpenAI had already been working on AI image generation for a while. They started with DALL-E, then improved it with DALL-E 2 and DALL-E 3. The previous models were diffusion-based, which means they gradually refined random noise into an image.
The new gpt-image-1 model works differently. It's built into the GPT-4o model, which means it has a deep understanding of both text and images. This "multimodal" approach helps it create images that better match what people are asking for.
OpenAI added DALL-E 3 to ChatGPT last year, but now they've moved to this newer technology. The switch seems to have paid off with more people using the image generation feature than ever before.
Some users have noticed that the new model is especially good at certain styles, like the Studio Ghibli-inspired images that took over social media. It's also better at including text in images, which was a common problem with earlier AI image generators.
What This Means for the Future of AI Images
The API launch shows that OpenAI sees image generation as a key part of their business. By making it available to developers, they're helping spread AI-generated images across more apps and services.
This could change how we think about images online. In the past, every picture was either taken with a camera or drawn by hand. Now, more and more images will be created by AI based on text descriptions. This might be good for businesses that need lots of custom images, but it also raises questions about art, authenticity, and what we can trust online.
The C2PA metadata is one way OpenAI is trying to address these concerns, but as the technology spreads, we'll need to think more about how to tell AI images from human-made ones.
For developers, the API opens up new possibilities. They can now create apps that generate custom illustrations, design mockups, product visualizations, and more - all without needing human artists. This could make certain kinds of visual content much cheaper and faster to produce.
What's Next for OpenAI and Image Generation?
OpenAI hasn't said what's next for their image generation tools, but based on their pattern of rapid improvements, we can expect the technology to keep getting better. Future versions might create more realistic images, understand more complex prompts, or generate images in specific styles with greater accuracy.
One area that might see improvement is animation. Right now, the API creates still images, but the next step could be generating short animations or even video clips. Other companies are already working on AI video generation, so OpenAI might be looking at that too.
Another possibility is more customization options for businesses. The current API lets companies generate images based on text prompts, but future versions might let them fine-tune the model on their own brand assets or style guides. This would help companies create images that match their exact visual identity.
Whatever comes next, the launch of this API marks a significant step in making AI image generation a standard tool for businesses rather than just a cool demo. As more companies integrate it into their products, we'll see how people use it and what new issues might arise.
Frequently Asked Questions
How much does OpenAI's image generation API cost?
The API uses token-based pricing: $5 per million text input tokens, $10 per million image input tokens, and $40 per million image output tokens.
Can I use the API to generate images in any style?
Yes, the model can create images in various styles and follow specific guidelines for artistic direction.
Do images created with the API include metadata?
Yes, all images include C2PA metadata that identifies them as AI-generated and tracks ownership.
Will OpenAI use my prompts or generated images to train future models?
No, OpenAI has promised not to use customer API data, including prompts and generated images, to train its models.
Which companies are already using the image generation API?
Canva, GoDaddy, and Airtable are among the early adopters testing the API for different business purposes.
How does this compare to other image generation APIs?
It's more expensive than competitors like Stability AI ($0.08 per image) and Google's Imagen ($0.03 per image), but may offer different quality or features.
Can I control what kinds of images the API generates?
Yes, companies can adjust content moderation settings to align with their brand needs and prevent inappropriate content.
How many people have used OpenAI's image generation features?
Over 130 million users have used the image generation feature in ChatGPT, creating 700 million images in the first week alone.