ai-short-drama

What it is in one line

A system that compiles short-drama tropes — "fake weakling who is secretly strong, rebirth, system, hidden boss, comeback" — into a story engine with arcs, character consistency, per-episode payoffs and binge hooks: an open-source Skill for Codex and Claude Code that owns the story and serialization state instead of stuffing an entire series into one generation prompt.

Who built it

Hao0321, an independent developer, open source on GitHub. The repo's content and docs suggest the author practices this exact "AI toolchain for short-drama production" workflow and distilled the lessons into a reusable Skill.

Read: the author is very likely a short-drama practitioner or heavy experimenter — the granularity (eight-dimension scoring, per-shot briefs, state deltas) is not something an outsider could invent. The honesty shows in the automation-boundaries section: login, payment, and compliance uncertainty must pause for a human, which means the author knows what AI must not touch.

What it actually does

• Scout → researches current tropes, platforms and cases
• Greenlight → an eight-dimension score picks concepts worth testing
• Bible → characters, world rules, ability costs, reveals, and the villain ladder
• Season / Episode → plans micro-arcs, writing episodes with a dominant turn, payoff, cliffhanger and state delta
• Studio → routes between serialized episodes, a complete micro-drama, a single-shot hit, grid moments, and continuous long shots
• Model-aware production → records model duration / reference quota, locks recurring-character three-views, global style, and per-shot timelines
• Produce → outputs a schema-validated production pack, fixed entity IDs, per-shot generation briefs and an edit handoff
• Audit → separately diagnoses concept, episode, season, production and packaging problems

Automation boundaries: the core Skill can independently finish concept, script, production pack and validation; media prompts and generation suggestions require ai-media-generator; assembling a final MP4 requires a separate edit executor; and login, extra payment, model-capability mismatch, compliance uncertainty and official public release must stop for confirmation.

What it deliberately does not do: include the author's private series bibles, character assets, or unpublished story settings; model capabilities are not hardcoded as permanent facts and must be re-verified before production.

What old behavior it replaces

Making AI short dramas used to mean two paths, both with fatal flaws:

One prompt for the whole series — hand "write a 10-episode short drama" to a model; the opening works, then characters drift and the world logic collapses, because a single prompt's context window cannot hold the state of a serial. This is the root of "one good episode is easy; ten that hold together is the hard part."

Human-managed serialization — character sheets, ability costs, villain ladder, per-episode hooks, all tracked in documents and spreadsheets, re-aligned by hand for every episode. It runs on human discipline — a production-manager grind that breaks the moment the person changes.

ai-short-drama replaces both by moving serialization state — character consistency, style locks, per-shot timelines, inter-episode hooks — out of a single prompt and out of human memory, into checkable, resumable state files on disk. It attacks the real bottleneck of industrializing short drama: continuity.

Business model

None. MIT open source, free, no paid tier, no services. (The repo cites aizhuiguang.tech's public "generalizable product mechanisms" as the reference for the Studio layer, noting it is an independent reimplementation that copies neither copy nor code, with an evidence registry.)

Read: this is the classic "trade methodology for attention" play. Short drama is one of the hottest content tracks of 2026, and releasing the workflow as a reusable Skill is a bet that whoever defines the short-drama production pipeline owns the category's ecosystem slot. Monetization would be later — paid templates, custom production, or taking orders.

Hard numbers

• 48 stars / 9 forks (this batch's observation)
• MIT, Python scripts (drama_lint.py / studio_lint.py) plus schema validation and unit tests
• Install is a clone into ~/.codex/skills/ or ~/.claude/skills/
• Users and actual episodes produced: not disclosed
• Status: watching; the repo is being updated continuously

Four-way read

The call

It bet on the right bottleneck — continuity — and did the engineering properly. Generating a single episode stopped being the problem long ago; keeping ten episodes consistent is. Turning characters, style, timelines and hooks into checkable, resumable state files is a production-grade idea, not a demo-grade one.

Honesty is its most valuable asset. Writing out that login, payment, compliance uncertainty and public release must pause for a human — and that model capabilities must not be hardcoded as facts — is rare in AI-generated-content tooling and shows the author knows which steps AI must not touch.

But its reach is limited. Forty-eight stars says it is still a niche methodology repo, and it depends on coding agents like Codex and Claude Code as the execution environment — which filters out most short-drama practitioners, who are not programmers. The positioning is "people who use agents make short dramas," not "people who make short dramas use agents."

What to watch next

① Whether stars clear 200 in three months — word-of-mouth velocity for a methodology repo ② Whether any actually produced short drama publicly credits this workflow — stronger evidence than stars ③ Whether commercialization appears (paid templates, production services, a content account matrix)

What you can take from it

Product logic: when building AI content-production tools, decide first which steps must stop for a human to confirm, then talk about automation. Writing the automation boundary into the docs is not conservatism; it is professionalism, and it decides whether the tool is trusted in real production.

Verdict

Unproven. Right instinct, serious engineering, honest boundaries — but slow diffusion, a narrow audience, and no commercialization yet. It represents one sample of the "open-source AI content-production methodology" route. The continuity insight is worth remembering; its merits wait for real works and real revenue to verify.