How Generated Worlds builds a short drama platform: the models we run, the measurements behind each release, and what every one of them still does badly. Written by the people who ship it.
updated September 15, 2026|9 posts, written when something ships|rss
Where people are watching
plotting…
AI models
Which models we run, where they sit in the pipeline, and the ones we measured and switched off.
Our episode analysis asks a language model which line of dialogue a clip should end on, and looks the time up itself. Asking the model for the timestamp produced numbers that fit nothing.
Transcription runs on faster-whisper rather than a hosted API, with voice-activity detection on and the model kept in memory between episodes. What that choice fixes, and when it is the wrong one.
Following the subject with a face detector holds the subject slightly better and looks worse. On a catalogue with burnt-in captions it also destroys the captions.
One control on the player does two things: a tap sends a free heart to the creator, a hold sends a token from the viewer's wallet. Both reach the studio that made the episode.
Viewer Journey shows a short drama creator how far viewers get through their series, which episodes hold them and what they did next — with a worked example of how to read the charts.
Full screen on a computer now puts the vertical picture on the left and the episode list on the right, instead of a phone-shaped video floating in black.
Viewers can now pay for the next episode, or the rest of a series, without creating an account first. A password comes afterwards, by email, if they want one.