Inspiration
I learned that streamers and live broadcasters can lose up to 90% of their audience during periods of inaction - whether that's breaks in a Twitch streamer's gameplay, ad breaks in between NBA quarters, or anything else that interrupts the action. This translates to actual, tangible financial losses.
What it does
So we built the first in a series of mini games that will engage, reward, and retain viewers during periods of downtime. These mini games can range from structured to unstructured, simple to complex, and are generated in real-time in response to the live video.
How we built it
We used openCV, a Python library, AWS Rekognition for image-labeling and text recog, etc. Used that to build a helper desktop app which interfaces with the Twitch streamer's broadcasting software (in this case, OBS) and runs the rekognition models and sends the generated game data into our Twitch extension. This Twitch extension loads directly into users' streams and then lets them play specifically during downtimes, when the streamer is between games.
Challenges we ran into
Models, especially, high-fidelity ones, take a long time to train. We had to balance efficacy of Rekognition model with limited time.
Accomplishments that we're proud of
We had to learn so many different new technologies to get this to work. We're particularly proud of getting the initial Rekognition code to work in under an hour.
What we learned
Desktop app development, working with AWS Rekognition, and Twitch extensions, all new stuff.
What's next for Playtest
Well, we're supposed to talk to the founder of Gen.G, one of the largest eSports orgs in the world which to this day is still the 2nd most successful LCK (Korean League of Legends regional league) team of all-time. Hoping that we can continue to find a foothold as we consolidate the value we're providing to live content creators.
Log in or sign up for Devpost to join the conversation.