Over the past few weeks, you may have noticed some impressive AI-created video memes. Harry potter Nightmarish footage of Will Smith eating spaghetti, reimagined as a Balenciaga commercial, recently went viral. , highlights how problematic some uses of technology can be.
Looking at these videos, last year programs like Craiyon (previously known as DALL-E Mini) produced crude and often unrealistic, but recognizable images such as baby surveillance footage. I remember the moment when AI image creation tools became popular. Gas station robbery, Darth Vader courtroom sketch, Elon Musk eating crayons.
Craiyon was an open-source knockoff of OpenAI (the company behind ChatGPT)’s DALL-E 2 image generator, which was then carefully restricted. This tool was the first to demonstrate the ability of AI to take text prompts and transform them into what look like real photographs or human-drawn illustrations. Since then, DALL-E has become available to everyone, and as programs such as Midjourney and Dream Studio have developed and improved similar tools, it has become relatively easy to create complex, realistic images with just a few taps of the keyboard. can now be created in
see more
These image creation tools are becoming a reality as engineers tweaked the knobs and levers of the algorithms behind these image generators, added more training data, and bought more GPU chips to run it all. has become very good at counterfeiting Check out Alex Jones at the Gay Pride Parade or The Box of Covenants at his sale at the Yard, to name a few examples from the subreddit dedicated to weird AI imagery.
Widespread access to this technology and its sophistication has forced us to rethink how we view online images. It was highlighted last month when an AI-generated image of him purportedly showing the arrest of Donald Trump went viral. Due to this incident, Midjourney has announced that it will no longer offer free trials of its services. While this fix may deter some Cheapskate malicious attackers, it leaves the broader issue untouched.
As WIRED’s Amanda Hoover wrote this week, algorithms still struggle to generate compelling videos from prompts. Creating many individual frames is computationally expensive. Also, as today’s jittering and sputtering videos demonstrate, it is difficult for algorithms to maintain sufficient consistency between them to produce meaningful videos.
But AI tools are much more adept at editing videos.Balenciaga meme and the version it references friend and breaking bad, was created by combining several different AI tools. I generated a static image first, then added a simple animation effect. But the end result is still impressive.
Runway ML, a startup developing AI tools for professional image and video creation and editing, announced this week a new, more efficient technique for applying style changes to videos. I used it to create a dreamlike footage of my cat Leona walking in a ‘cloudscape’ from an existing video in just a few minutes.
Various machine learning techniques open up new possibilities. For example, a company called Luma AI uses a technique called Neural Radiance Fields to turn 2D photos into detailed 3D scenes. Feed some snapshots into the company’s app and you can play with fully interactive 3D scenes.
These clips suggest that we are at an inflection point in AI video creation. Similar to AI image generation, following the proliferation of memes, the quality and controllability of AI video could greatly improve, embedding the technology everywhere. AI could be a muse for some writers. Runway’s tools were used by a visual effects artist working on the Oscar-winning piece. All at once, anywhereDirected by Darren Aronofsky whale, black swanand Pi again runway fan.
But to understand where AI video is headed, Midjourney and Dream Studio’s images are so sophisticated now, and how difficult it becomes to distinguish between real and fake clips. It is enough to see Of course, people can already manipulate video with existing technology, but it’s still relatively expensive and difficult to do.
Rapid advances in generative AI could prove dangerous in an age when social media is weaponized and deepfakes are propaganda toys. As Jason Parham wrote in his WIRED this week, we need to seriously consider how generative AI can bring back and reuse ugly stereotypes.
For now, our instinct to trust video clips is mostly trustworthy, but it may not take long for the footage we see to be less credible and authentic than it used to be.
