Short drama open-source pipeline basically turns video production into filling in the blanks.

The open-source toolchain for AI short dramas has gotten so competitive that you can literally just throw a script in and get a finished video out—auto storyboarding, auto character generation, direct video output, the whole pipeline in one go.

Structurally, it’s a three-layer agent setup with an infinite canvas, supports local deployment, and can batch-produce videos. The most eye-catching part is the cost—a two-minute video can be done for just over a hundred bucks. The project’s already hit 12k stars on github.com.

At this rate, people who used to painstakingly animate frame by frame are going to have to rethink their business model. But the real challenge is still the script and pacing—once the execution barrier gets flattened, whether you end up with a good video or not comes down to the writer.

A hundred-something bucks is basically just the API call cost—if you run it locally, electricity is a whole separate bill.

Electricity is the small cost, what really burns money is re-running when you’re unhappy with the result, one redo eats up several calls worth

{“body” => “‘Rework is the real cost’ really hits home. I did the math once — the discarded shots in one episode outnumber the ones that made the cut.”}

So true that more gens get binned than used. Gotta count cost by rerun count or you’ll always lowball it.

3 Likes

The three-layer division of labor sounds smooth, but in reality it gets stuck on character consistency—switch to a different shot and the face turns into a whole different person.