Paper: an AI pipeline reduces a studio to three tools and one prompt
The author reproduces a paper's short-drama pipeline: GPT-6 Astra does topic research, Claude Opus 5.5 writes scripts and Picsart produces the visuals, earning $24,689 from 30 episodes a month.
Somebody actually wrote the paper. Qianjiang Media Economics Lab, September 19: "The End of the Studio: Serialized Synthetic Characters as the Unit of Creator Income Across Three Short-Video Platforms". The framing is cold. A studio used to be writers, animators, a composer, an editor and a producer holding the look in his head. In the paper's model a studio is three tools with one job each, and the producer role is a prompt. GPT-6 Astra runs the research, browses platforms, pulls a ranked table of what's already working. Claude Opus 5.5 owns the writers' room, world rules, season arc, a fresh episode brief every day. @Picsart is the whole production floor, hero sheet, world sheet, music, and an agent director locking every shot to the same look. Then they do the math the guides never do. Ten million views on Shorts pay $300 to $1,000. The same ten million on TikTok pay $4,000 to $10,000, if the video passes 60 seconds. Long-form YouTube RPM sits 10 to 30 times above Shorts. So the whole architecture collapses into one question: which views actually get paid. Their reference series is "Tower 40". A 60cm repair robot, one floor per episode, one strange complaint per door. Corridor and roof never change. Only what waits behind the door does. 30 episodes plus two YouTube seasons in a month, $24,689, no camera, no animators, no studio. I ran the exact pipeline. Every prompt is in the thread below. The paper in one line: the studio ends the day the daily episode ships ↓
This post quotes another post: x.com/retrochainer/status/2103541402135117886