Accessible AI
RunwayML is a program that allows anyone to access machine learning models and techniques without any explicit coding knowledge. Though I have some reservations about any system that works like a magic black box, RunwayML uses its simplicity not as a way to dilute what it can do, but rather to empower others to bring complicated AI techniques into other fields. The program targets itself towards “creatives” with a lot of featured models for visual work, like automatically coloring old photos or removing backgrounds. You can train models or just use some right out of the gate.
Instead of doing anything creative or useful, I decided to get comfortable using the program with the Selfie to Anime model. Running this model was incredibly simple, especially compared to the work that I had put into some previous experiments. After a few minutes of setting up the program, it works seamlessly, allowing me to upload pictures from my computer and have them nearly instantly come out processed by the program. The results were… exciting?
One note about RunwayML for folks who intend to try it out. Right now it is in beta and is what I would call pseudo-free. Technically they charge to use their remote GPUs to do all the processing, but you get $10.00 of starting credit and it only costs $0.05 a minute. My anime experiment cost me less than $0.50 to run and I could have cut that way down with better planning on my part.
I am super excited to see where this program goes, especially with the promise of local processing via Docker and even a browser-based version! They also have an educator hub and I am looking forward to coming up with ways to use this in my classes.