RunwayML is an AI-powered video editor. Its Gen 1 model is now available in an iPhone app.
The iPhone app offers features from RunwayML’s Gen-1 AI model. The New York-based startup unveiled the model last February: It can give videos new styles, similar to the filter technology popularized by Instagram, but with much stronger visual intervention and greater flexibility.
For example, a realistically photographed person can be transformed into a claymation character or a vivid drawing. The city background becomes an apocalyptic scene from a Roland Emmerich movie, and the lapdog becomes a tiger (see cover image). You can choose from a variety of styles, enter a text command, or specify a reference image to find your preferred style.
RunwayML is now available for free on the App Store for newer iPhones. Those who already have a Runway account for the web version can use it to exchange assets between smartphone and PC.
Gen-2 can already generate short videos
RunwayML is already a step ahead in terms of AI video: the newer Gen 2 model can generate short videos from text input or from a combination of image and text. Gen 2 is currently available in a beta version on the web. The first users are sharing some examples on Twitter.
My very first text to video with #Gen2
text_prompt: a woman wearing a white dress meditating, standing on water, peaceful day, the most beautiful scene
— Kris Kashtanova (@icreatelife) April 20, 2023
So practical text-to-video generation just happened.
Here’s “futuristic car speeding” using Runway v2. It is a short clip, for now… pic.twitter.com/cGV4XRy4My
— Ethan Mollick (@emollick) April 20, 2023
It takes several minutes to process a few seconds of video, however, and the videos still contain glitches and are somewhat jerky. In addition to Runway, Google is working on Phenaki, a text-to-video AI system, and Meta is working on Make-a-Video. Meta has another model in the works, Dreamix, specifically for text-to-video editing.
The question will be whether the rapid development of image AI systems will also apply to video AI systems. If so, at least video databases for stock footage or short social media clips will soon face serious AI competition.