OpenAI launches ChatGPT image-generator

OpenAI has launched 4o image generation within its flagship ChatGPT platform. Use cases highlighted by Open AI for 4o include creating 'workhorse' imagery including logos, diagrams and illustrations, which point to several applications across the marketing communications industry. Here are some of its key features:

More accurately rendered images: AI-generated imagery has a history of being more visually intriguing than precise. With 4o, OpenAI has been guided by the principle that image generation should not just be beautiful but useful. Creating consistent logos should be easier with 4o image generation which renders text on images far more accurately. This is a particularly significant development at a time when distorted or incorrect text is often a telltale sign of AI-generated imagery.

Better binding and adherence to prompts: In the context of image generation, 'binding' refers to the precision with which elements within a prompt are incorporated into a final output. Better 'binding' capabilities mean that 4o image generation is likely to deliver images based on text inputs with a better adherence to prompts than earlier versions.

Being able to iterate on images: Uploaded images can be used as either an inspiration or a basis for image generation. With 4o image generation incorporated within GPT-4, users now have the option to refine an image through a natural conversation to ensure consistency. For instance, a character’s appearance could remain consistent even as various other elements in an image undergo changes: vital for creating multiple iterations of a single hero visual in an ad campaign.

The ability to render more complex visuals: The number of objects that can be in a visual also increases with 4o image generation. A statement from Open AI says, “While other systems struggle with ~5-8 objects, GPT‑4o can handle up to 10-20 different objects. The tighter binding of objects to their traits and relations allows for better control.”

A wider variety of styles: The models have been trained on a vast array of images including photorealism, allowing for more convincing looking outputs.

Like all AI-powered models, Open AI acknowledges that 4o image generation is not perfect. The company has also put in place limitations to the model to prevent it from generating illegal imagery, and to include metadata which will help identity the provenance of an image as being AI-generated.

4o image generation is available across large parts of the ChatGPT suite of products, with it likely to be launched soon for the Enterprise and Edu tiers. Other companies that have highlighted the logo-creation abilities of their image generation software include Google with Imagen, a part of its Gemini suite of AI tools.