explaingit

jnmetacode/ai-shortfilm-prompts

17Audience · writerComplexity · 2/5ActiveSetup · easy

TLDR

A prompt library and Claude Code Skill that captures a five-stage workflow for making AI-generated short films, based on the Zombie Scavenger production notes.

Mindmap

mindmap
  root((ai-shortfilm-prompts))
    Inputs
      Story idea
      Camera and lens choice
      Scene description
    Outputs
      Layered prompt
      Storyboard by shot
      Self-check report
    Use Cases
      Generate short film prompts
      Run Claude skill
      Adapt templates to new models
    Tech Stack
      Markdown
      Claude Code Skill
      Sora
      Kling

Things people build with this

USE CASE 1

Build a layered prompt for Sora, Kling, Veo, Jimeng, or Seedance 2.0 using the five-stage structure from this repo.

USE CASE 2

Install the included Claude Code Skill to be guided through tags, character, atmosphere, camera, and storyboard.

USE CASE 3

Adapt the Zombie Scavenger template into a personal short film with the placeholder scene field.

USE CASE 4

Strip Chinese brand references from a prompt by switching to the generalized English templates.

Tech stack

MarkdownClaude CodeSoraKlingVeo

Getting it running

Difficulty · easy Time to first run · 30min

Using the Claude Code Skill needs Claude Code installed, and generating actual video needs a paid account on Sora, Kling, Veo, Jimeng, or Seedance.

In plain English

This repository is a prompt library and methodology guide for making short AI-generated films. It collects the workflow behind a 3-minute video called Zombie Scavenger, made by a creator who goes by Mx-Shell. The film was shared by a Hollywood director named PJ Ace, who called it one of the best short films he had seen in years. After Mx-Shell explained his method on a Chinese livestream, the author of this repo turned those notes into a structured, reusable set of files. The core of the project is a five-stage prompt structure. Each prompt is built in this order: a short list of style tags, a description of the character and scene, the atmosphere and visual quality, the camera rules, and finally a storyboard broken down by second or by shot. The README explains three counter-intuitive habits that go with this structure: name real camera bodies and lenses instead of saying cinematic, describe small imperfections like worn paint or oil in joints, and leave endings quiet rather than ending on an explosion. The repo includes a single ready-to-paste prompt that sets up an anamorphic widescreen look with a simulated IMAX camera, a Panavision lens, and a handheld breath-like float, with a placeholder for the user's own scene. It also ships a Claude Code Skill that walks a user through the five stages, runs a ten-item self-check, and warns about brand names that some video generators block. Folders include methodology and FAQ documents in English and Chinese, the original Chinese-language prompt files used for Zombie Scavenger and several other shorts, generalized English templates that strip out specific brand references, and plugin metadata for installing into Claude Code. The README lists Sora, Kling, Veo, Jimeng, and Seedance 2.0 as compatible video models, with notes on quirks for each.

Copy-paste prompts

Prompt 1
Use ai-shortfilm-prompts to write a full 5-stage prompt for a 3-minute desert chase scene that targets Veo 3. Include camera body and lens.
Prompt 2
Install the Claude Code Skill from jnMetaCode/ai-shortfilm-prompts locally and walk me through generating a 60-second cyberpunk noir prompt.
Prompt 3
Critique my Sora prompt against the ai-shortfilm-prompts ten-item self-check and rewrite weak sections. My prompt is: [paste prompt].
Prompt 4
Adapt the Zombie Scavenger template from ai-shortfilm-prompts to Kling, noting which brand names to remove because Kling blocks them.
Prompt 5
Compare the English and Chinese FAQ docs in ai-shortfilm-prompts and summarize the differences in advice that a Western creator should know.
Open on GitHub → Explain another repo

Generated 2026-05-22 · Model: sonnet-4-6 · Verify against the repo before relying on details.