For my sins, I do occasionally scroll through TikTok. In between the usual suspects of ear worm music loops and memetic dances I generally lack the co-ordination to recreate but not the determination, there is a spattering of AI-generated content. On my 'For You' page, this usually takes the form of image filter videos that twist users' selfies into vaguely resembling a frame from a favourite anime. Well, ChatGPT's recent image generation update can now spit out pictures that look like it's traced Studio Ghibli's homework.
The update specifically brings image generation to OpenAI's GPT-4o model, refining a number of things AI image generators have historically struggled with, such as photorealism and rendering legible text, or even a full glass of wine, according to PCWorld. The update also allows users to refine image results through the chat interface, and the AI is apparently now better able to generate consistent variations on a theme. For example, one of OpenAI's demo videos shows GPT-4o generate a penguin mage in various styles, including a low-poly look, a reflective metallic get-up, and looking like a wargaming miniature.
