It draws like it knows stuff
Matt V
March 27, 2025
OPENAI
OpenAI has come out with a new feature called “Images in ChatGPT”, significantly upgrading image generation inside ChatGPT and leveraging GPT-4o.
It’s available across all plans - even the free tier, though limits still apply.
The tool comes with a few upgrades.
One is improved “binding,” which means it’s much better at keeping object shapes and colours accurate, even when you ask to generate busy scenes with 15 to 20 items.
Older models often got this wrong, mixing up when multiple elements were involved.
Text on images has also seen a big improvement.
Instead of generating the whole image at once (like DALL-E), this model draws it bit by bit, left to right, top to bottom, similar to how we write.
That small shift makes a big difference when it comes to spelling and layout.
Images do take a bit longer to generate now, but OpenAI thinks the boost in quality is worth the wait.
In early demos, it produced posters, comics, scientific diagrams and even restaurant stickers, all with accurate text and design, without needing much explanation from the user.
Three things to know:
Image generation is now upgraded for all users, including the free tier.
GPT-4o handles more complex prompts with clearer text and better object accuracy.
All images include invisible metadata to mark them as AI-generated.
Watermark drama? Sorted with metadata
On the safety front, OpenAI says it has measures in place to stop harmful or explicit content, prevent watermark removal, and tag each image with hidden metadata to show it’s AI-generated.
There’s no visible watermark, but OpenAI can trace images internally if needed.
And yes, anything you create is yours to use, as long as it follows their content guidelines.
AI: “Give me 5 seconds and a prompt.”
Me: gives chaotic prompt
AI: “Here’s your scientific diagram.”