You may have noticed some great AI-generated videos over the past few weeks. Harry Potter Reimagined as a Balenciaga ad Terrifying shots Will Smith eats spaghetti The two have recently gone viral. They point to how quickly the ability of AI to create videos is advancing, as well as how difficult some uses of the technology are.
These videos remind me of when AI imaging tools exploded last year, when programs like Craiyon (formerly DALL-E Mini) allowed anyone to conjure up recognizable, albeit primitive and often surreal like Surveillance footage of kids robbing a gas station, Darth Vadar courtroom drawings, and Elon Musk eating crayons.
Craiyon was an open source emulated version of it which was then carefully restricted DALL-E 2 Image generator from OpenAI, the company behind ChatGPT. The tool was the first to demonstrate the ability of artificial intelligence to direct text and transform it into what looks like real images and human-drawn illustrations. Since then, DALL-E has been open to everyone, and programs like Medvoyage and Dream Studio have developed and perfected similar tools, making it relatively easy to create complex, photorealistic images with just a few clicks of the keyboard.
Show more
As engineers tweak the algorithmic buttons and levers behind these image generators, add more training data, and drive more GPU chips to run it all, these image creation tools have gotten incredibly good at simulating reality. For some examples of a subreddit dedicated to weird AI images, check out Alex Jones at the gay pride parade or L’Arche d’Alliance at a garage sale.
Widespread access to this technology and its sophistication is forcing us to rethink the way we view images online, as was highlighted after AI-generated images purporting to show the arrest of Donald Trump became viral last month. The incident led to Midjourney announcing that it would no longer be offering a free trial of its service – a solution that may deter some cheap bad actors but leaves the wider problem untouched.
As WIRED’s Amanda Hoover wrote this week, algorithms still struggle to create a compelling video from a prompt. Generating many individual frames is computationally expensive, and as today’s choppy, jerky video shows, it’s difficult for the algorithms to maintain enough consistency between them to produce video that makes sense.
However, AI tools are becoming more adept at editing videos. Mimi Balenciaga, as well as reference versions friends And very bad, combining a few different AI tools, first to create static images and then adding simple animation effects. But the end result is always impressive.
Runway ML, a startup that develops artificial intelligence tools for professional photo and video creation and editing, this week launched its new, more efficient technology for applying stylistic edits to videos. I used it to create these dreamlike snapshots of my cat, Leona, walking through “cloudcape” from an existing video in just a few minutes.
Various machine learning techniques open up new possibilities. company called Loma IA, for example, uses a technique known as neural radiation fields to transform 2D images into detailed 3D scenes. Insert a few screenshots into the company app and you’ll get a fully interactive 3D Scene file to play with.
These clips indicate that we are at an inflection point for the AI ​​video industry. As with AI image creation, the rise of memes could be followed by dramatic improvements in the quality and controllability of AI videos that bring the technology to all sorts of places. Artificial intelligence can become a source of inspiration for some authors. Tools were included Used by visual effects artists Working on an Oscar Everything is everywhere at once. Darren Aronofsky, Director WhaleAnd black SwanAnd Goodbye it is also track fan.
But you only need to look at how advanced Midjourney and Dream Studio’s advanced footage is now to get an idea of ​​where AI video is heading – and how hard it is to tell the real from the fake. clips. Of course, people can already process video with today’s technology, but it’s still relatively expensive and difficult to achieve.
Rapid advances in generative AI can be dangerous in a time when social media has been weaponized and deepfakes have become a propaganda game. As Jason Parham wrote for WIRED this week, we also need to think hard about how generative AI can restore and reuse ugly stereotypes.
For now, we can rely on the instinct for confidence in the videos, but it may not be long before the images we see become less solid and honest than before.
