OpenAI Beefs Up ChatGPT's Image Generation Model

OpenAI launched a caller representation procreation AI exemplary connected Tuesday, dubbed ChatGPT Images 2.0. This exemplary tin make much than 1 representation from a azygous prompt, similar an full survey booklet, arsenic good arsenic output text, including successful non-English languages, similar Chinese and Hindi. This merchandise is disposable globally for ChatGPT and Codex users, with a much almighty mentation disposable for paying subscribers.

When immoderate large AI institution releases a caller representation model, it tin revive involvement and boost usage, particularly if societal media users follow a meme-able trend, transforming images of themselves. Last year, Google's motorboat of the Nano Banana exemplary was a large infinitesimal for the company, particularly erstwhile users started posting hyperrealistic figurines of themselves online. Earlier this year, ChatGPT Images made waves connected societal media arsenic users shared AI-generated caricatures.

What’s Different?

Since the caller exemplary tin pat into ChatGPT’s “reasoning” capabilities, Images 2.0 tin hunt the net for caller accusation and make much than 1 representation astatine a time. In essence, the bot tin usage further steps to output much thorough generations from a azygous prompt. Images 2.0 besides has a much caller cognition cutoff date: December 2025.

This besides means that outputs from the caller exemplary are much granular. For example, I generated an infographic with San Francisco’s upwind forecast for the adjacent day, arsenic good arsenic activities worthy doing. The representation ChatGPT generated included close upwind details for the rainy day, on with accurate-looking drawings of the Ferry Building, Castro Theater, Painted Ladies houses, and Transamerica Pyramid.

Additionally, Images 2.0 is much customizable for users who privation unsocial facet ratios for representation outputs. The caller exemplary tin make images, ranging from 3:1 wide to 1:3 tall, and users tin set the image’s size arsenic portion of their punctual to the AI tool.

First Impressions

After a fewer hours of generating images with the caller model, I was mostly impressed with the substance rendering capabilities, successful English astatine least. Not that agelong ago, representation outputs featuring text, from immoderate of the large models, often included galore malformed characters oregon words with errant other letters. ChatGPT struggled to label images accurately 2 years prior, truthful the cleaner, much analyzable outputs from Images 2.0 are a motion of continued improvement. Google has besides focused connected improving representation outputs featuring substance successful its recent iterations of Nano Banana.