After Stable Diffusion announced its AI video generator last week, users began animating memes that were once still photos.
In a possible sign of things to come and like something out of a Harry Potter movie, classic memes like “distracted boyfriend” and “disaster girl” are now moving and coming to life.
Unsettling pic.twitter.com/WkxqzPde9H
— Know Your Meme (@knowyourmeme) November 27, 2023
Disasterpic.twitter.com/DwdSiQBjQ8
— Know Your Meme (@knowyourmeme) November 27, 2023
People have been using Stable Video Diffusion which can generate up to 25 frames of video from a still image (depending on the model) to breathe fresh life into the classic memes and it has sparked a reaction across the internet.
The first instance of meme animation appears to have come from @fofr on X (the platform formerly known as Twitter) who experiments with generative AI, especially video.
Stable Video Diffusion (SVD) is currently only for research purposes and those wanting to use the video diffusion model have to contact Stability AI to request waitlist access. But it’s not just SVD that people have been using to animate memes, Runway ML can complete similar tasks.
first world problems pic.twitter.com/rovwTCjI7c
— Know Your Meme (@knowyourmeme) November 27, 2023
Another one pic.twitter.com/NCrX2m4lo2
— Know Your Meme (@knowyourmeme) November 27, 2023
Although the videos are intriguing, not everyone is thrilled they are here — reflecting the unease people have with generative AI technology.
“I can only imagine some AI tool is paying you to post all this crap cuz it sucks,” writes one disgruntled X user beneath a Know Your Meme thread of the animated gifs.
Others pointed out how terrible the people in the images look after being given AI treatment, something that AI video sorely struggles with.
At the moment, Stable Diffusion’s video model is limited: the video it can generate is only short (about four seconds) and the company admits that it cannot yet achieve “perfect photorealism.”
But just like we witnessed with AI image generators, it’s likely that AI video models will vastly and rapidly improve in the coming years — or even months.