Seedance-prompt is a Hermes/Agent Skill that generates realistic video prompts by identifying shooter, device, era, and in-scene details to mimic authentic footage. It provides structured prompts and references for device flaws, atmosphere, and anti-AI checks.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (2 so far).
What it is
A Skill designed to produce realistic AI video prompts for Seedance and text-to-video models. It focuses on identifying who shot the video, what equipment was used, the era, and embedding device flaws, environmental sounds, and timeline events into prompts.
How it works
The project divides a video idea into seven parts:
- 主要角色, 2) 地点, 3) 视觉风格, 4) 摄像风格, 5) 时间轴, 6) 音频, 7) 目标. It includes three reference libraries:
- 设备美学缺陷包 with device-specific flaws (e.g., 2000 年代消费级 DV, VHS, Super 8, 智能手机竖屏, 监控摄像头)
- 氛围翻译表 translating abstract terms into observable phenomena (e.g., 温暖, 孤独, 紧张, 生活气息)
- 去 AI 感自检 checklist (consistency, defects, timeline events, audio-visual alignment, avoidance of ad-style shots)
Getting started
Installation has three options:
- 方式一：安装到 Hermes using:
git clone https://github.com/zhouwei713/seedance-prompt.git \
~/.hermes/skills/creative/realistic-video-prompting
Then restart Hermes and use:
realistic-video-prompting
- 方式二：作为通用 Skill 使用 by placing SKILL.md in compatible Agent environments (Claude, Codex, Cursor, OpenCode, etc.).
- 方式三：作为提示词方法论参考 by reading SKILL.md and references files.
Recent releases
There are no releases listed (latest 0).
Traction
Not provided in the README beyond repository stars. The repository shows 145 stars.
Behind the repo
No linked startup/company information is provided in the README.
Caveats
License is MIT. Created 2026-07-07; last_push 2026-07-07. No age-related warnings beyond creation/date.