OpenAI has introduced a new feature, integrating image generation capabilities into ChatGPT, effective March 25, 2025. You can find this feature under the subhead Images in ChatGPT. ChatGPT is allowing its users to utilize GPT-4o for directly generating images in ChatGPT. 
 

This news is a blessing for those who need this service a lot, as it solely focuses only on image creation. You can use it on ChatGPT Plus, Team, Free subscriptions tiers, Pro and more. Though the free tier limit is not static nor limited to DALL-E, representative Taya Christianson narrated. However, they are still figuring out the fixed number of images that could be shared. It will be changeable with time for sure as per them. 
 

Going ahead with the published ChatGPT frequently asked questions, a user can generate a maximum of three images a day with DALL-E 3. Christianson has claimed that fans could still use a custom GPT. 
 

Gabriel Goh, the researcher head of ChatGPT, claimed it to be an improvisation from the earlier models. Here, he claimed that the GPT-4O unimodal is generating any asked data, whether text, audio, video, or image. He has further claimed it comes with binding, where Open AI image generations maintain a balance between orders and objects. Maximum image models lag in delivering the expected images, especially the right colors and shapes, especially when you give several orders at a time. But with this new version, things are different, irrespective of the number of commands you give. It delivers exactly what you want with the uttermost accuracy and reliability. 

Users are likely to enjoy an improvised version when it comes to text rendering, and hence, it becomes easy to develop coherent text without any further typos in the image. For all other tools, you would have noticed texts get grabled simply. Goh also highlighted that commands with typo errors will deliver OpenAI image creation with errors. He said the present work is the hard work of many months. The work has yet to reach perfection, but they have created an image quality acceptable to the masses. 

The system seems to employ an auto aggressive approach where they are delivering images sequentially clockwise and top to bottom, exactly the way texts are written. They are not employing the diffusion model technique, which most image-generating apps commonly use. This model helps in Open AI image creation at once. Goh further claimed that images created through ChatGPT are superior due to their test binding and delivering capacity. 
 

The ChatGPT team displayed several examples before the launch, including the image of Newton's prism experiment with all major components, comics with specific situations and characters displaying dialogues in bubbles and unique posters with specific text. Overall, it is covering all kinds of images. In the months of hard work, they have focused on developing unique background images for menus, logos, stickers, etc. 
 

Jackie Shannon, the multimodel product head of ChatGPT, announced that this new feature does not demand many explanations for the images that already have been equated earlier, like Newton's prism experiment. 
 

The new system here is going to be faster and more efficient, and it will take less time to generate images. Shannon added that they will never close the space for improvisation in their techniques as, with time, the quality demanded will be changed, and so will the deliverables. 
 

Discussing the safeguard measures used here, after several unacceptable image generations, including Kamala Harris with a gun or political figures in compromised positions on different platforms, the OpenAI team came forward and claimed that the system is following robust safety measures to ensure there is no misuse. Sexual deepfakes, watermark removal and CSAM generation requests are not allowed through this tool. 
 

Further, OpenAI's new image feature doesn't come with any indicator that the image generated is through AI. Still, if we listen to Shannon, it has to follow standard c2PA metadata to get approval for it to be created through OpenAI. The company is also going to follow certain internal tooling systems to monitor the images on the tool. 
 

Perfection can never hit to anything, but as per Shanon, they are working hard to deliver a safe digital environment to users. This is just a beginning, he further added.