GTA research - the reference corpus for a GTA Vidrush

17 Sep 2026, 12:55 · videos a script-to-footage tool could actually rebuild
1,019usable references
160channels
1.3Bviews behind them
5script formats
1,379rejected as un-rebuildable
The one test every video had to pass. If we hand its script to the tool, can the tool rebuild it out of footage that already exists? Vidrush narrates a script over sourced clips. It films nothing, it has no face, and it cannot reproduce a performance. So a video whose value IS the creator - his own speedrun, his own mod build, his own reaction, his friends' banter - is useless to us however many views it has. A video that explains, counts, tells or argues, with interchangeable GTA footage underneath, is exactly what we need. And no mods, at all. GTA 6 ships on PS5 where modding is not possible, so a video whose premise needs a modified game teaches us nothing we could ever reproduce - not a graphics mod, not a ghost, not a character who is not in the real game, not a weapon you cannot actually pick up. Those are cut on sight even when the views are huge.

The bar: 8 minutes or longer, 300,000 views or more, English, faceless channel, real GTA 5. Where it came from: 13,529 videos, 1,538 channels read one by one, then 306 survivors read again video by video. The face check is finished. Every channel in this corpus had nine real frames pulled from its own videos and looked at. It overturned nine of my own calls - six channels keep a webcam in the corner or a host in frame (Hazardous, Sernando, The Leaderboard, Joe Bart Games, Fenton, Ball Creative) and three were the wrong language behind English titles (Awesome Genome in Hindi, VonOrdona in Tagalog, GameMagz in Indonesian). All nine are out.

The corpus

TitleChannel ViewsSubs DateMinFormatBuildFace