One question underneath all of these pages: which GTA videos could a script-to-footage tool actually rebuild? Not the ones with the most views - the ones whose value is the script and the footage, not the person who made them. Everything here was read by hand: 1,538 channels one at a time, then the survivors again video by video, then a second harvest of 1,778 more that the original collection never held. Nothing was decided by a keyword rule.
Every kept video also carries a Build badge saying how hard it would be to make. Easy is an argument or a news piece, where the script leads and any relevant footage carries it. Medium has to keep the thing being explained on screen while it is explained. Hard is a list: twenty items, twenty specific shots.
The videos a GTA Vidrush should learn from, sorted into the five script formats the tool can produce. Thirteen years of footage means the tool has everything it needs to illustrate almost any script.
The short list, not the corpus. Every video here was read as a sheet of sixteen frames across its runtime and kept only if our tool could rebuild it from footage it can download. GTA 5, Online and 6 only, because that is what the library holds. His own five gradings sit at the top.
The same test applied to the fortnight after the reveal. Half of it was people filming their own faces. What worked was counting the details in footage everyone already had - and there is only half an hour of that footage in the world.
The whole 13,529-video database with the junk moved out of the way - kids channels, big personalities, mods, roleplay servers, silent compilations, other languages, brand uploads. Nothing deleted: every removal carries its reason and can be undone in one filter.
Every video that survived, GTA 5 and GTA 6 together, with a tick box that remembers what you have already watched. This is the one to open when you actually want to sit and go through them.
Counted on both datasets, with the sampling bias spelled out. In the only fair comparison we have, long form wins on median, mean and total views.
Built from the GTA research archive. Every page carries a Face column: green means that channel was verified on nine real frames pulled from its own videos, grey means not yet. The check has already overturned nine of my own calls, so treat grey as unknown rather than as clean.