Computer Vision News - September 2022

6 AI Application “ Not only do we have great synthesis, but we also have good control because a text input controls our video engine, ” he reveals. “ We learn a mapping from the text to the respective video output and spend most of our time making it look good, controllable, and easy. It can’t take too long. People want quick feedback. ” Synthesia is around 120 people and raised $50 million in Series B funding at the end of last year. It has just started a research arm and is engaged in the academic community. Currently, it is working on new developments based on the latest research. It training, how-to videos, and product marketing . According to Matthias, the results are virtually indistinguishable from reality. “ We use complex technology, the latest stateof theart incomputervision, toachieve the realism we’ve got, ” he explains. “ Our customers aren’t very tolerant to mistakes, and with faces, if you make even a small mistake, people see it immediately . That sets the bar high, so it has to look good. Technically speaking, I think that’s our greatest challenge. ” Full details of the technology under the hood are a trade secret, but Matthias says Synthesia uses generative neural networks and keeps a close eye on the latest innovations coming out of computer vision conferences.