While many fear a future in which AI-powered media becomes unique from traditional media, destroying society and/or civilization in the process, we are not quite there yet. Exhibit A is a surreal AI-generated beer ad that went viral over the weekend.
entitled “Synthetic buzzer,” The 30-second video first appeared on instagram about a week ago, created by Helen Power and Chris Boyle of London manufacturing company called Privateisland.tv. The couple weren’t available for comment before this story was posted, but judging by the look of the video, it looks like they probably created it using a new version of Runway. generation-2 An AI model that can create short video clips based on written cues, similar to how Stable Diffusion can create still images.
Set to a raucous crowd backing track and “All Star” Smash Mouth, the video features simulacra of people having fun at a stereotypical American backyard barbecue, sometimes physically blending in with impressionistic beer vessels. The women laugh with their jaws open. Beer glasses turn into beer cans. The blazing grills take on the status of a columnar fire tornado and arc across the yard. It is a vision of a surreal hell that is both familiar and impressively alien.
Why is this so weird? Currently, AI video generators are still primitive. As their creators train the models, they work with a much smaller set of inputs than AI still image synthesis models, and these models require significantly more computational resources to run. The impressionistic take on beer ads likely comes from imbibing the essence of real beer ads in the Gen-2 dataset. Runway has not disclosed the dataset used for Gen-2 training, but in paper for Gen-1 (older model) it specified “an internal dataset of 240 million images and a user dataset of 6.4 million video clips”.
We’ve been experimenting with Gen-2 (which is currently in closed testing), and creating even weird alien clips like these still requires human perseverance, running and discarding many generations to even get a satisfying result. Even so, the resulting clip is only a few seconds long. In the case of Synthetic Summer, Privateisland.tv created the clips, selected the best ones, and combined the segments into a sequence, adding music and sound effects.
But wait, beer isn’t the only product fictionalized by AI for memetic purposes. On April 24, someone called “Pizza Later”. tweeted a video for a fictional restaurant called “Pepperoni Hug Spot”, heavily AI-generated, which includes distorted video clips of people eating pizza created by Gen-2 Runway. In addition, its creator reportedly generated script with GPT-4, used Midjourney for still images and voice-over Eleven Laboratories. They put it all together with the help of Adobe. after effects.
Definitely wasted 3 hours of my life doing this today… Everything from voiceovers to videos and images is created by artificial intelligence. Compiled in After Effects. More details below. pic.twitter.com/CXv6gWM8gj
— Pizza Later (@Pizza_Later) April 24, 2023
Both these human-initiated and human-collected works show that generative AI still has a long way to go before it independently dazzles the masses with society-changing memes. Humans are still driving these alien jobs, and we could potentially get some semblance of comfort from that. May be
However, no video can match the purity and grandeur of AI-created Will Smith eating spaghetti, which will forever remain in our hearts as our first AI-created nightmare video meme.