OpenAI upgrades ChatGPT's image generation, Sam Altman says some stuff may 'offend people'

OpenAI on Tuesday rolled out an upgraded image-generation feature in ChatGPT, enabling users to generate images with greater precision, including the ability to render text accurately and follow complex prompts.

Unlike previous iterations, the tool can maintain consistency across multiple image generations, making it particularly useful for applications such as game design and storytelling.

OpenAI CEO Sam Altman announced the development on X, calling it an “incredible technology” and a step forward in creative freedom.

"We think people will love it, and we are excited to see the resulting creativity,” he said, adding that some stuff may “offend people”.

“What we'd like to aim for is that the tool doesn't create offensive stuff unless you want it to, in which case within reason it does.”

In a blog post, OpenAI highlighted the model’s ability to analyse and incorporate user-uploaded images into new creations.

The model can handle complex scenes with 10-20 objects, surpassing the 5-8 object limit of rival systems, and it integrates user-uploaded images to inspire or refine outputs.

Trained on a vast dataset of online images and text, GPT-4o offers "surprising visual fluency," capable of maintaining consistency across iterative designs, like a video game character, through natural conversation.

The model also supports customisation, allowing users to specify details such as hex-code colours, aspect ratios, or transparent backgrounds. However, its detailed rendering means images may take up to a minute to generate.

The company also outlined its approach to safety, saying all AI-generated images would be embedded with metadata, which will identify an image as coming from GPT‑4o.

“ We’ve also built an internal search tool that uses technical attributes of generations to help verify if content came from our model,” it said.

Safeguards against misuse

OpenAI revealed it has put safeguards in place to prevent the creation of harmful or misleading content, including restrictions on deepfake imagery and graphic violence.

“When images of real people are in context, we have heightened restrictions regarding what kind of imagery can be created, with particularly robust safeguards around nudity and graphic violence,” it said.

The rollout has started for Plus, Pro, Team, and Free-tier ChatGPT users, with Enterprise and educational access expected soon. Developers will also gain access to the API in the coming weeks, expanding the tool’s reach across different applications.