OpenAI’s new ChatGPT image generator makes faking photos easy

For nearly two centuries, convincing photo manipulation has demanded specialized skills or tools, whether it be darkroom techniques or adeptness with software like Photoshop. However, OpenAI has now streamlined this process to a simple sentence input with the launch of its latest tool. On Tuesday, OpenAI introduced GPT Image 1.5, an advanced AI image synthesis model that significantly outpaces its predecessor, generating images up to four times faster and at a reduced cost of approximately 20% via the API. This rollout is not the first of its kind; Google had previously launched its own image-editing model, Nano Banana, which garnered considerable excitement in the AI community, prompting OpenAI to accelerate its own development. What sets GPT Image 1.5 apart is its classification as a “native multimodal” image model, integrating image generation within the same neural network that interprets language prompts. Unlike earlier models such as DALL-E 3, which utilized a different image creation technique, GPT Image 1.5 treats images and text as comparable data chunks, referred to as “tokens.” This approach allows the model to seamlessly process both visual and textual information. For instance, if you provide a photo and request to modify it—like dressing someone in a tuxedo at a wedding—the model interprets the text and image simultaneously, producing altered visuals in the same manner it would generate text. This innovative technique enables GPT Image 1.5 to manipulate images more effectively than its predecessors, whether it involves changing poses, angles, removing objects, or altering styles—all while maintaining facial accuracy throughout edits. Users can interactively communicate with the AI about their images, refining the results much like one would edit a draft in ChatGPT, making the art of image manipulation accessible to everyone, regardless of their photographic expertise.