zhouwei713/seedance-prompt — explained in plain English
Analysis updated 2026-05-18
Turn a short scene idea into a detailed, realistic prompt for AI video tools like Seedance or Sora.
Add device specific camera flaws, like VHS artifacts or smartphone shake, to make generated video feel less polished and more authentic.
Use the built in checklist to catch overly clean, AI looking output before generating a video.
Install the skill into Hermes, Claude, Codex, or Cursor to reuse the prompting method across projects.
| zhouwei713/seedance-prompt | easylink-ai-open/tiri-agent | kitlangton/skills | |
|---|---|---|---|
| Stars | 257 | 257 | 257 |
| Language | — | TypeScript | — |
| Setup difficulty | easy | moderate | easy |
| Complexity | 1/5 | 4/5 | 1/5 |
| Audience | vibe coder | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Reference material and examples are written in Chinese.
This project is an Agent Skill, a reusable instruction file, for writing more realistic prompts when generating AI video with tools like Seedance, Sora, Kling, Runway, or Veo. Its main idea is that most AI video prompts describe a pretty picture but skip a key question: who supposedly filmed this footage, on what device, and in what era. The README explains that answering those questions first, then writing in the resulting camera flaws, natural mistakes, ambient sound, and a timeline, is what makes generated video feel like it actually happened rather than looking like a polished AI sample. The README gives two worked examples in Chinese. One turns the phrase morning on a New York street into a prompt describing a smartphone video supposedly shot vertically by a passerby, including a consistent character description, a timed shot list, and camera quirks like exposure shifts and slight shake. The other turns a short description of an old Korean neighborhood into a prompt styled after a 2000s consumer camcorder, complete with compression artifacts, faded color, and small unplanned moments like a cat wandering into frame. The skill breaks a video idea into seven parts: main character, location, visual style, camera style, timeline, audio, and overall goal, and it includes built in reference material covering device specific visual flaws for things like VHS camcorders, film cameras, smartphones, and security cameras, a dictionary that translates moods like warmth or loneliness into concrete filmable details, and a checklist for catching leftover AI looking polish before finishing a prompt. You install it by cloning the repository into a Hermes skills folder, or by dropping its SKILL.md file into other AI agent tools such as Claude, Codex, Cursor, or OpenCode that support skills or long prompts, or you can simply read the included reference files directly as a prompting guide. The project is released under the MIT license, which allows free use, modification, and commercial use as long as you keep the copyright notice.
An AI agent skill that writes realistic AI video prompts by first deciding who filmed the footage, on what device, and what flaws that would leave behind.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly vibe coder.
This repo across BitVibe Labs
Verify against the repo before relying on details.