I can't stop watching these hilariously bad AI-generated videos of celebrities like Will Smith and Scarlett Johansson
- The internet is freaking out over clips generated by AI text-to-video generator ModelScope.
- Twitter users are calling the clips — like one of Will Smith eating spaghetti — "cursed."
The internet is freaking out over AI-generated videos that are so bad you can't look away.
There's a nightmarish compilation of "Will Smith eating spaghetti" that has racked up 1.8 million views on Twitter and depicts a cartoon-like version of the "Men in Black" star shoving handfuls of spaghetti into his mouth. A similar video was posted of actress Scarlett Johansson devouring pasta.
Viewers have called the videos "cursed," "deeply unnerving," and "super creepy."
The videos were made using ModelScope Text to Video Synthesis, a free AI video generator that AI firm Hugging Face released to the public last week. Users can enter a prompt like "Beyonce walking 30 dogs" or "Spiderman surfing," and the bot spits out a clip in a matter of minutes.
The generator can also depict violence in graphic detail. One Reddit user generated a video of "Joe Rogan fighting a bear." The dramatic 17-second clip depicts the podcast host boxing a bear to its death, turning into one, and then eating its fleshy remains.
As one Reddit user puts it: "This is some fever dream shit."
Redditors also shared clips of Jordan Peterson drinking apple cider vinegar, Emmanuel Macron fighting an orangutan, and a Zombie apocalypse film.
Despite the AI's impressive capabilities, the videos are low quality and stamped with a Shutterstock watermark. That may be because the Modelscope tool is trained on images scraped from stock photo sites, according to Vice.
A Hugging Face spokesperson referred Insider to a blog post explaining the model behind the generator. The post says that the model was trained on public data sets like Webvid and that it "was not trained to realistically represent people or events."
The clips come as AI-image generators like DALL-E 2, Stable Diffusion, and Midjourney have taken the internet by the storm. While the images they produce can look eerily realistic, they aren't perfect. Tools like DALL-E, for example, were found to add extra fingers to hands, The New Yorker reported.
AI image generators have also been found to produce images that reflect racial and gender stereotypes, and can spread misinformation. Just this past week, a deepfake of the Pope wearing a luxurious puffer coat went viral, sparking conversations about the ethics of AI image generators.