DirCast directs and commentates live Dota 2 from raw player cameras
Original title: DirCast: Live Esports Directing and Commentary with a Single Streaming Video-Language Model
Trained on 27 professional matches, its commentary wins 29.2% of Proact-VL comparisons against GPT-4o.
DirCast is a single streaming video-language model that chooses camera cuts and commentates using only past frames from every player's raw camera. Each second it scans all cameras at low resolution, applies a hold-or-switch director, then reads the chosen view at high resolution before speaking. Context keeps what aired and what was said while unused views expire. Direction beats rule-based and zero-shot directors, including GPT-5.
- The dataset contains 27 professional Dota 2 matches with one camera per hero.
- Commentary wins 29.2% of comparisons against GPT-4o under Proact-VL.
- The strongest of eleven streaming commentators watching the human broadcast wins 19.4%.
- Expired unused views bound context cost for a stream of any length.
From the original article
1 min read at huggingface.co
A live esports broadcast needs a director who chooses which camera goes on air and a caster who talks over that choice, every second and without knowing what comes next. Streaming video language models have learned to commentate, but only over a broadcast that a human director has already produced.
By Zixian Gao, Lixin Xiu, Penghao Luo · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.