OpenAI’s New Image Generator Within Their GPT-4o Model is a Beast, Here’s a First Look

Two AI chat bots trying to have a conversation can be quite entertaining, while OpenAI’s new image generator within their GPT-4o model takes things to the next level. Unlike earlier models where image generation was handled by separate systems like DALL-E 3, this is built into GPT-4o itself.

It excels at rendering text within images, which was a weak spot for earlier models. This new image generator can also handle complex prompts with up to 10-20 objects in a single image, maintaining accuracy in how they relate to each other (something called “binding” that older models struggled with). You can also refine images through conversation. For example, you might start with a basic image, then ask it to tweak the style (say, turn it into an anime or Studio Ghibli look), add elements, or adjust details, all the while keeping the core consistent across edits.

The catch? It’s a bit slower than previous generators (taking about a minute per image due to the higher quality), and heavy demand has reportedly strained OpenAI’s GPUs, leading to temporary rate limits. This launch has sparked a lot of buzz and social media is already flooded with creative outputs, especially the Ghibli-style makeovers.

We trained our models on the joint distribution of online images and text, learning not just how images relate to language, but how they relate to each other. Combined with aggressive post-training, the resulting model has surprising visual fluency, capable of generating images that are useful, consistent, and context-aware,” said OpenAI.