This year has seen some rapid advances in AI image generation models, with Google's Nano Banana Pro going viral last month. But OpenAI is hoping to have the last word: On Tuesday, the company announced the newest version of ChatGPT Images, an AI image model with improvements that should help it follow instructions and create more detailed images.
You can take advantage of the new model beginning today. It's available to all ChatGPT users globally and available in the API as GPT-Image-1.5. You can use the model in a new dedicated space in ChatGPT for images in the sidebar, with preset styles and trending designs, built specifically for image creation -- a major upgrade from simply talking with ChatGPT to generate images.
"Many people's first experience with ChatGPT involves turning a text prompt into a picture. It's a magical way to see what this technology can do, but the chat interface wasn't originally designed for this," Fidji Simo, CEO of applications at OpenAI, wrote in a blog post. "Creating and editing images is a different kind of task and deserves a space built for visuals."
(Disclosure: Ziff Davis, CNET's parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
OpenAI says that the new model is four times faster than the previous one and better able to follow instructions and handle precise editing. You'll be able to use those editing tools to add in specific objects, change styles of objects, try on specific pieces of clothing and remove elements.
This image, generated with GPT-Image-1.5, shows improvements in text generation in images. OpenAI
OpenAI says the new 1.5 model is also better at generating legible text in AI images. This is a notoriously tricky task for generative media models, with even the first generation of OpenAI's model falling flat on similar promises. Test images made with 1.5 seem to highlight improvements, but we'll have to wait to see how our hands-on testing goes.
This infographic made with GPT-Image-1.5 has decently legible text. OpenAI
Don't miss any of our unbiased tech content and lab-based reviews. Add CNET as a preferred Google source on Chrome.
All of those updates would be massive upgrades compared to the original model. When I reviewed the first model, it was noticeably slower than competitors and lacked essential editing tools to refine your images. The improvements in the 1.5 model strive to fix those annoyances.
... continue reading