Generative AI art remains hugely controversial – as Transport for Ireland just discovered when it had to apologise for using AI-generated images for a Halloween ad campaign. Nevertheless, video is widely seen as the next frontier. Premiere Pro Generative Extend is now available in Beta and and an Adobe AI video model is expected to roll out soon.
While John Wick maker Lionsgate has brought in Runway to develop a bespoke AI video model for its own use and Blumhouse is working with Meta Movie Gen, others are having fun using AI to generate videos of characters from GTA San Andreas. Because why not? But while they look kind of realistic, it's got to the point where realism in AI video no longer impresses me.
We've come a long way from just two years ago when someone used nascent AI image generation to 'upgrade' 90s video game characters from Virtua Fighter. Now there's the possibility of making characters move. And the results are as realistic looking as some of the best deepfakes.
"So cool! Games will become more immersive as NPCs have realistic routines, personalities, and evolving relationships with the player," one person suggests in response to Angry Tom's post of AI GTA characters on X. Some people are pointing out that CJ isn't very convincing, but, in general, the experiment is another example of how fast AI video is advancing.
The user didn't use one of the best-known AI video generators but a browser-based app called Hailuo AI. I was unable to find any information on its website about what model it's using or how it was trained, which is one of the issues drawing a big question mark over whether AI video generation can be safe to use commercially.
But I feel that we're reaching a point with generative AI imagery where the tech itself has lost its surprise value, but it hasn't yet got to the point where it's really good for anything other than saying: 'wow, AI was able to generate 5 seconds of almost usable footage'.
