Research
DirCast: a single streaming video-language model directs and commentates live Dota 2 esports
DirCast decides every second whether to cut between cameras and then comments on the chosen view, trained on 27 professional Dota 2 matches. It reportedly wins 29.2 percent of commentary comparisons against GPT-4o, versus 19.4 percent for the strongest of eleven streaming commentators.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
DirCast: Live Esports Directing and Commentary with a Single Streaming Video-Language Model