What is the story about?
For most of the past two years, you could spot an AI-generated image if you looked closely enough. Something was slightly off about the hands, text that
made no sense, lighting that felt artificial and backgrounds that dissolved into blur. The tells were always there. OpenAI is now working to eliminate every single one of them.
What is GPT image 2?
The company is developing gpt-image-2, a new image generation model for ChatGPT that is specifically designed to produce visuals so sharp, detailed, and naturally composed that telling them apart from a real photograph becomes genuinely difficult. The upgrade moves beyond the dreamy, animated styles of earlier models and focuses on sharper details and natural-looking results with better text placement and layouts that actually make sense.
This is a meaningful step beyond where ChatGPT's image capabilities currently stand. The existing GPT-4o image generation, which OpenAI launched earlier this year, already represented a major improvement over the DALL-E 3 era. GPT-4o brought natively multimodal, precise, and photorealistic outputs into ChatGPT for the first time, and the response from users was immediate. The viral Studio Ghibli-style images that flooded social media in March were generated using this model, and they gave millions of people their first real taste of how far AI image quality had come.
But gpt-image-2 appears to push that further. Where GPT-4o improved the overall quality and accuracy of outputs, the new model is specifically targeting photorealism as its primary goal the ability to generate images that look like they were taken with a real camera in a real environment, not created by a machine interpreting a text prompt.
Upcoming dedicated image feature in ChatGPT
OpenAI has also launched a new dedicated Images feature within ChatGPT, designed to make image generation faster and more intuitive with preset filters, prompts updated regularly to reflect emerging trends, and a model that makes precise edits while keeping details intact, generating images up to four times faster than before.
For everyday users this opens up genuinely useful possibilities more believable product mockups, more convincing travel visuals, more practical photo editing without needing professional software. A marketing team that once spent days on a campaign shoot could potentially generate the same visual assets in minutes.
But the implications don't stop at convenience. Images that are functionally indistinguishable from photographs raise serious questions about misinformation, identity, and trust. Images generated by ChatGPT now carry C2PA metadata, a provenance standard that embeds information about how an image was created, but metadata can be stripped, and most people viewing an image online never check it.
OpenAI hopes this leap toward realism will help ChatGPT stand out against rivals like Google and Anthropic, both of whom are building their own image generation capabilities at pace. The competition is real and moving fast.
What's coming is not just a better image tool. It's a fundamental shift in what counts as visual evidence and the world isn't quite ready for that conversation yet.
