Realistic Video Prompting skill
A skill that writes structured prompts for Seedance, Kling, Veo and Sora: a home-footage look, camera flaws, a timeline and sound
- 335 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skill only writes prompt text, but the result can look like real footage, and the author of the clip must label its AI origin
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/realistic-video-promptingDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill only writes prompt text, but the result can look like real footage, and the author of the clip must label its AI origin
This is third-party code. Review the repository files before installing.
What it does
The skill turns a short idea like "a morning on a New York street" into a structured prompt for the Seedance, Sora, Kling, Runway and Veo video models. The agent first settles four things: who is filming, with what gear and in which era, what and where is in the frame, and what mood is wanted. From this "found footage" backstory the details follow: shake, autofocus hunting, digital artifacts, ambient sound. The prompt comes out in seven blocks: characters, location, visual style, camera style, a 10–15 second timeline, audio and goal. The character block fixes clothing, hair and skin texture and a sentence locking appearance for the whole clip. The location block lists 8–12 environment details and forbids extras such as shop windows, ads and cafes. The timeline gets imperfect events: the camera drifts aside, autofocus hunts, the frame dips for a second. Three reference files in `references/` describe camera-flaw packs (2000s DV, VHS, Super 8, vertical smartphone, surveillance camera), a dictionary that turns moods into visible phenomena and a checklist for an AI look.
Who it is for. AI video creators who want the look of real footage: home DV, a vertical phone, surveillance recordings.
Good fit when
- You want a scene "as if from a passer-by's phone": a New York morning in vertical 9:16
- The video has to look like a 2000s family DV tape with shake and autofocus hunting
- A finished prompt is too polished and needs to go through the AI-look checklist
- A vague phrase like "a lonely evening by a shop" has to grow into 10–15 seconds with a timeline and sound
Not a fit when
- You want an ad-style or cinematic image: the skill deliberately removes the gloss
- The finished video is meant to pass as real footage of events: the skill creates a believable look, and labeling AI origin stays with the author
- You need the clip itself generated: the skill only writes prompt text
Example request
Write a Seedance prompt: a morning on a New York street, shot vertically on a smartphone, 15 seconds, the look of casual street footageLimitations
- The skill writes a prompt, while a separate video model makes the clip, and its limits and cost are outside the skill.
- SKILL.md and the docs are in Chinese, and so are the example prompts, so an English or Russian result has to be requested explicitly.
- Installation is described for Hermes, in other agents you copy the SKILL.md file and the `references/` folder by hand.
- A video model may ignore part of the prompt, so you may need to pick the result from several attempts.
How to disable. Delete the `~/.hermes/skills/creative/realistic-video-prompting` folder or the SKILL.md file from your agent's skills directory.
FAQ
Which video models does the skill work with?
It writes prompt text for text-to-video and image-to-video. The README names Seedance, Sora, Kling, Runway and Veo.
How do I install the skill in Hermes?
Clone the repository with `git clone https://github.com/zhouwei713/seedance-prompt.git ~/.hermes/skills/creative/realistic-video-prompting`, restart Hermes or open a new session and call the skill by the name realistic-video-prompting.
Does the skill work in Claude Code and Cursor?
Yes, as an ordinary SKILL.md. The README advises placing the file in an agent that supports skills or long prompts: Claude, Codex, Cursor, OpenCode.
How many seconds of video does a prompt cover?
The timeline is built for a 10–15 second clip: the skill lays out events and sound by segment.
What is in the references folder?
Three files: `camera-aesthetics.md` with camera flaws of different eras, `atmosphere-dictionary.md` with a dictionary that turns mood into visible phenomena and `anti-ai-checklist.md` with a pre-output checklist.
Discussion
No questions or reviews yet.
Related
SkillsA skill that turns a plot or scene into a cinematic AI video prompt for 6 to 30 seconds, with shots, performance and sound.
Two skills for generative video and images: video writes Seedance, Kling and Veo prompts by dramaturgy rules, image picks the model and its syntax
A skill that teaches an agent to write prompts for ByteDance's Jimeng Seedance 2.0 video model: @ references, camera language, timing, templates
The skill breaks a sales video into a storyboard and writes a prompt pack for the Seedance video model in Jimeng.
A skill that turns an idea, theme or novel excerpt into a 30-second multi-shot prompt for the Seedance 2.5 video model.
A Claude Code skill that turns an idea into a five-stage prompt for Sora, Kling, Veo and Seedance, with templates for 21 genres
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Realistic Video Prompting (this card) | no | 335 | |
| Seedance 2.0 Prompt Writing Skills | no | 4,413 | teaches Seedance 2.0 syntax: @ references to materials, camera language, timing |
| Seedance 2.5 Skill | no | 260 | plans a storyboard and generates video through Atlas Cloud |
| Awesome Seedance 2 Prompts | no | 2,201 | a library of ready-made prompts, while this skill builds a prompt for your scene |
README in short
The README, in Chinese, opens with examples where a New York morning and an old Korean neighborhood expand into finished prompts. Then come the problems of typical prompts, suitable scenarios, the seven blocks, three reference files, installation and the repository layout.
README badge
Original SKILL.md
---
name: realistic-video-prompting
description: 为AI视频生成模型（Seedance、Sora、Kling、Runway、Veo等）撰写真实感、去AI感、纪录片质感的结构化视频提示词。当用户想生成AI视频提示词、优化现有视频提示词、追求"真实感""家庭录像感""手机随拍感""DV感""监控录像感"的视频效果，或提到Seedance、文生视频、视频prompt时，务必使用此skill。即使用户只是描述了一个想拍的场景或氛围，也应使用此skill将其转化为完整提示词。
version: 1.1.0
author: zhouwei713
license: MIT
metadata:
hermes:
tags: [ai-video, prompt, realistic-video, seedance, kling, veo, runway, sora]
---
# 真实视频提示词生成器
把用户的一个模糊想法（"我想要一段韩国社区的日常感视频"）转化为高密度、可执行、去AI感、像真实素材拍出来的结构化视频提示词。
## 核心哲学
高质量视频提示词的本质是回答一个问题：**这段素材是谁、用什么设备、在什么年代、出于什么目的拍下来的？**
一旦确定了"素材来源身份"，所有细节都能推导出来。2005年朋友随手拍的DV会抖、会失焦、会突然关机；2024年博主的手机竖屏会有美颜感、会对着镜头说话；90年代的监控摄像头没有声音、俯视角、低帧率。
第二个关键认知：**AI模型有强烈的默认审美**（电影感运镜、完美构图、商业调色、干净画面）。想要真实感，必须用负面约束主动关闭这些默认行为，并在时间轴里植入"非完美事件"。
## 工作流程
### 第一步：确定四要素（缺一不可）
如果用户没有说清楚，先问这四个问题（可以合并成一次提问）：
1. **拍摄者身份**：谁在拍？朋友、家人、路人、Vlogger、监控、纪录片摄制组、还是无人拍摄的固定机位？
2. **设备与年代**：DV、VHS、8mm胶片、iPhone、GoPro、监控、专业电影机？哪个年代？
3. **主体与场景**：拍谁？在哪？什么时间段？
4. **情绪目标**：温暖怀旧、孤独疏离、紧张不安、平静治愈？
如果用户已经给出足够信息，直接推断并在输出前用一句话确认你的理解。
### 第二步：按七段式模板逐段生成
严格按以下顺序输出，每段独立成块，段间不混淆信息：
```
主要角色 → 地点 → 视觉风格 → 摄像风格 → 时间轴 → 音频 → 目标
```
各段写作规则：
**1. 主要角色**
- 只写可视化的具体特征：年龄段、妆容、每件衣物的颜色/材质/版型、鞋、配饰、发型细节（含碎发/刘海等小特征）
- 禁用抽象形容词（时尚的、好看的、有气质的）
- 加一句皮肤/质感描述（如"逼真的皮肤纹理"）