The introduction of “Images in ChatGPT” by OpenAI constitutes a revolutionary advancement that brings image generation capabilities directly to the ChatGPT platform. The new GPT-4o model powers this development, which empowers users to generate images directly during their chats, representing a breakthrough in AI-generated content creation.

“Images in ChatGPT” offers advanced image generation capabilities across ChatGPT’s free and subscription-based tiers, like Plus, Pro, and Team, to make this technology accessible to more users. OpenAI spokesperson Taya Christianson explained that free tier users will have usage limits similar to DALL-E 3, allowing for about three images per day, but mentioned that these limits might change depending on demand. OpenAI ensures that DALL-E enthusiasts will maintain access through a specialized GPT model.

OpenAI’s research lead Gabriel Goh described GPT-4o as a transformative “omnimodal” foundation that can manage multiple data types like text, images, audio, and video. The model shows significant progress through its improved ability to manage image “binding,” which resolves typical issues faced in AI image creation. GPT-4o proves to be reliable when handling 15 to 20 objects because it avoids mixing their colors or shapes, unlike previous models that often failed to maintain object-attribute relationships.

The model’s enhanced text rendering stands out as a significant advancement. AI-generated images have typically displayed text that appears scrambled or meaningless. Goh explained that the development required continuous iteration, which stretched on for many months until they achieved the correct outcome. The team reached a point where text rendering in digital images is consistently usable, despite acknowledging that achieving perfect text rendering for smaller text elements remains out of reach.

The system utilizes an autoregressive architecture, which represents a distinct departure from the diffusion models used by most image-generating systems. The sequential image generation from left to right and top to bottom mimics text generation processes and may enhance both text rendering and binding features.

OpenAI demonstrated during their presentation that their system can create scientific diagrams with exact labels like Newton’s prism experiment and produce comics with multiple panels displaying consistent characters and dialogue, as well as design informational posters containing precise text. The system demonstrated practical applications by producing transparent background images for both stickers and restaurant menus, as well as logos.

As multimodal product lead at ChatGPT, Jackie Shannon discussed how the system makes use of extensive world knowledge. She explained that when she creates an image, she faces her personal skill constraints but utilizes all of her accumulated world knowledge. The model incorporates world knowledge, which means users can request an image of Newton’s prism experiment without needing to describe the experiment to receive the correct image.

OpenAI claims that the upgraded quality and advanced features make the slight delay in image generation worthwhile. Shannon acknowledged existing latency issues but emphasized that the superior image quality and world knowledge features make up for additional waiting time.

OpenAI responded to potential misuse worries by implementing strong protective measures. The system protects users by preventing watermark removal and blocks creation of sexual deepfakes while refusing requests for CSAM content. The standard C2PA metadata, which identifies images as OpenAI creations, will be embedded into all generated images even though there are no visual watermarks present. Internal tools for image verification remain operational within the company.

While no system achieves perfection in this area, Shannon stated that they are constantly developing better safeguards and see their current efforts as initial measures. Users who generate images with ChatGPT maintain ownership rights and can freely utilize these images according to OpenAI’s usage policies.

By integrating “Images in ChatGPT,” OpenAI strengthens its main product while advancing AI-driven creative potential through a new visual expression tool in its conversational interface.