OpenAI Beefs Up ChatGPT’s Image Generation Model
OpenAI launched a new picture technology AI mannequin on Tuesday, dubbed ChatGPT Images 2.0. This mannequin can generate a couple of picture from a single immediate, like a complete examine booklet, in addition to output textual content, together with in non-English languages like Chinese and Hindi. This launch is accessible globally for ChatGPT and Codex customers, with a extra highly effective model obtainable for paying subscribers.
When any main AI firm releases a brand new picture mannequin, it could possibly revive curiosity and improve utilization, particularly if social media customers undertake a meme-able development, reworking photographs of themselves. Last yr, Google’s launch of the Nano Banana mannequin was a significant second for the corporate, particularly when customers began posting hyperrealistic figurines of themselves on-line. Earlier this yr, ChatGPT Images made waves on social media as customers shared AI-generated caricatures.
What’s Different?
Since the brand new mannequin can faucet into ChatGPT’s “reasoning” capabilities, Images 2.0 can search the web for current data and generate a couple of picture at a time. In essence, the bot can use extra steps to output extra thorough generations from a single immediate. Images 2.0 additionally has a more moderen data cutoff date: December 2025.
This additionally implies that outputs from the brand new mannequin are extra granular. For instance, I generated an infographic with San Francisco’s climate forecast for the following day, in addition to actions value doing. The picture ChatGPT generated included correct climate particulars for the wet day, together with accurate-looking drawings of the Ferry Building, Castro Theater, Painted Ladies homes, and Transamerica Pyramid.
Additionally, Images 2.0 is extra customizable for customers who need distinctive side ratios for picture outputs. The new mannequin can generate photographs starting from 3:1 broad to 1:3 tall, and customers can regulate the picture’s measurement as a part of their immediate to the AI software.
First Impressions
After just a few hours of producing photographs with the brand new mannequin, I used to be usually impressed with the textual content rendering capabilities, in English not less than. Not that way back, picture outputs that includes textual content, from any of the most important fashions, usually included quite a few malformed characters or phrases with errant further letters. ChatGPT struggled to label photographs precisely two years prior, so the cleaner, extra advanced outputs from Images 2.0 are an indication of continued enchancment. Google has additionally centered on bettering picture outputs that includes textual content in its recent iterations of Nano Banana.
