Ecom Video Seedance Prompt skill
The skill breaks a sales video into a storyboard and writes a prompt pack for the Seedance video model in Jimeng.
- 120 ★ on the repository
- Last commit
Extended access
Why this level
- Video frames go to the external DeepSeek API together with your key.
- Copying someone else ad video can infringe copyright and platform rules.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/ecom-video-seedance-promptDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Video frames go to the external DeepSeek API together with your key.
- Copying someone else ad video can infringe copyright and platform rules.
This is third-party code. Review the repository files before installing.
What it does
The skill works in two steps. `scripts/analyze.py` splits a video into frames, detects cuts with ffmpeg and asks a DeepSeek vision model to describe each shot in the S-A-C-S-C scheme: subject, action, context, shot and color. The result is cached in `storyboard.json` and never recomputed. `scripts/replicate.py` turns the storyboard into a pack: a prompt per shot, one full prompt, negative words in Chinese and English, a list of reference images and notes on how it differs from the original. Three styles are available: faithful copy, organic feed and paid ads. Output comes as Markdown and self-contained HTML. The README and prompts are in Chinese.
Who it is for. Producers and sellers of short ad videos who work with Seedance in Jimeng and want first-draft prompts fast.
Good fit when
- You found a strong sales video and want to see how its storyboard is built.
- A Jimeng prompt is needed per shot, plus a full version of up to 800 characters.
- The same video is reworked for organic feed style or paid ads.
- The prompt needs a list of reference images: what to shoot for the product and people.
Not a fit when
- The source video is copyrighted and the copy is meant for publication without the author consent.
- Work must be free of external APIs: frame analysis needs a DeepSeek key.
- The target model is not Seedance: the prompt format is tuned for Jimeng.
Example request
Break this shampoo video into a storyboard and write Seedance prompts in the organic feed style, the product being dry shampoo.Limitations
Frame analysis needs a DeepSeek key in `DEEPSEEK_API_KEY` or in `~/.deepseek_api_key`, so video frames go to an external API. Python with the `openai` package and ffmpeg must be present. The repository states no license. The README, schemas and examples are in Chinese. The author reports about 16 seconds of analysis for a 10-second video, a figure not independently checked. Marks such as an AI-generated watermark are noted in the storyboard and not carried into the prompt.
How to disable. Delete the skill folder from the agent skills directory, the key from `~/.deepseek_api_key` or the environment, and the `storyboard.json` cache next to the video.
FAQ
What does S-A-C-S-C mean in this skill?
Five required prompt elements: subject, action, context, shot and color. The skill fills them for every shot and merges them into the final text.
How do the three styles differ?
Faithful copy stays close to the original. Organic feed rewrites the video as natural content, and paid ads add a call to buy. Use `--style organic` or `--style paid`.
Which key is needed?
A DeepSeek key: the `DEEPSEEK_API_KEY` variable or the first line of `~/.deepseek_api_key`. No key is hardcoded.
What dependencies does it need?
Python with the `openai` package, ffmpeg and ffprobe on PATH and a DeepSeek key. Install dependencies with `pip install openai`.
Where does the output go?
In `./seedance-out/`: `<video name>_<style>_提示词.md` and `.html` files. The HTML places each prompt next to the original frame.
Discussion
No questions or reviews yet.
Related
SkillsA skill that teaches an agent to write prompts for ByteDance's Jimeng Seedance 2.0 video model: @ references, camera language, timing, templates
A dialogue that turns an idea into a storyboard and a prompt for the Seedance 2.0 video model in Jimeng, with six templates and ten examples
A skill that writes structured prompts for Seedance, Kling, Veo and Sora: a home-footage look, camera flaws, a timeline and sound
A skill that improves Seedance prompts, plans storyboards and generates video through Atlas Cloud, plus a library of ready prompts
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 turns an idea, theme or novel excerpt into a 30-second multi-shot prompt for the Seedance 2.5 video model.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Ecom Video Seedance Prompt (this card) | no | 120 | |
| Seedance Storyboard | no | 312 | a dialogue turns an idea into a storyboard; here the storyboard comes from an existing video |
| lengyi-seedance2.5-prompt | no | 139 | writes a 30-second prompt from an idea or a novel excerpt; this skill mirrors the structure of an existing clip |
README in short
The storyboard is cached once per video and prompt generation can repeat as often as needed. The repository includes a sample result for a 10-second Jimeng clip with an HTML page placing prompts beside the original frames.
README badge
Original SKILL.md
--- name: ecom-video-seedance-prompt description: 带货视频反向拆解与 Seedance（即梦）复刻提示词生成。当用户提供带货/电商/广告/商销/种草/千川视频（或提到即梦、Seedance、文生视频、复刻爆款、拆解视频、提示词），想分析镜头、复刻效果、生成视频提示词时，使用本skill。自动产出镜头级分镜表（景别/运镜/光影/动作时序/卖点）和符合 S-A-C-S-C 编导规范的完整提示词包（分镜头提示词+基础参数+负面词+@参考图清单），支持忠实复刻/自然流/付费流三种风格。 --- # Ecom Video → Seedance Prompt 看懂一条带货视频，把它翻译成可复刻的 Seedance 提示词。两步走： 1. **analyze.py 拆解**：抽帧 + 场景切换检测切分镜头 → 视觉模型逐镜头产出 S-A-C-S-C 拆解卡（景别/视角/景深/运镜/光影/色彩/主体/动作时序/物理交互/卖点）→ 全片带货全局分析（产品/框架/自然流或付费流/钩子/节奏）→ storyboard.json（缓存） 2. **replicate.py 生成**：分镜表 + 编导规则知识库 → 一次推理调用产出提示词包 （每镜头≤200字 + 完整版≤800字 + 负面词中英 + @参考图清单 + 差异说明 + 升级建议）→ markdown + 自包含 HTML（每镜头提示词配原片关键帧对照） ## 环境要求 - Python：`openai`；**ffmpeg/ffprobe 在 PATH** - API key：环境变量 `DEEPSEEK_API_KEY` 或 `~/.deepseek_api_key` 首行（代码不内置密钥） ## 工作流 ### 第 1 步：拆解（每份视频一次） ```bash python scripts/analyze.py "<带货视频路径>" ``` 输出分镜数与每个镜头的一句话摘要。常用：`--force` 重建｜`--min-shot 0.8` 最短镜头｜ `--clean` 清缓存。10 秒视频约 15-20 秒完成。 ### 第 2 步：生成复刻提示词 ```bash python scripts/replicate.py "<视频路径>" # 忠实复刻 python scripts/replicate.py "<视频路径>" --style organic # 自然流种草版 python scripts/replicate.py "<视频路径>" --style paid # 付费流千川版 python scripts/replicate.py "<视频路径>" --product "品牌名，主打XX，色号XX" # 补充产品信息（推荐） ``` 参数：`--duration 15`（4-15 秒）｜`--shot A,B` 只重出指定镜头。 产出在 `./seedance-out/`：`<视频名>_<风格>_提示词.md` 和 `.html`。 ### 第 3 步：向用户交付 1. **HTML 预览（首选）**：路径给用户双击打开——每镜头提示词旁就是原片关键帧， 可逐镜对照检查是否抓住了原片精髓 2. 把「完整版提示词」直接贴给用户（可直接粘贴即梦）；提醒按「@参考图清单」准备素材 （产品图纯色背景自然光拍摄、真人照片需打码/抠脸过审） 3. 用户不满意某镜头时，只带 `--shot X` 重出该镜头（省时省钱） 4. 需要变化时换 `--style` 重跑；产品信息以 `--product` 提供的为准 ## 回答规范 - 交付时简述：原片是什么（产品/框架/流量类型判断）→ 复刻策略 → 提示词 - 提示词是否符合规范（S-A-C-S-C 齐全、≤200字/镜头、含约束尾、负面词）由 skill 保证， 但要向用户说明：品牌 LOGO 等不可控元素已按"差异说明"处理 - 视频理解基于画面（无音频分析）；口播内容是从字幕/画面推断的，需注明 ## 注意事项 - 生成用开推理 + max_tokens=32768（reasoning 吃预算，小了会空回复） - 编导规则知识库在 `references/seedance-knowledge.md`（6 篇实战文档蒸馏）， 调整生成风格前先读它；分镜表结构见 `references/storyboard-schema.md` - 单视频拆解一次永久缓存（.cache/<sha>/storyboard.json）；视频改了会自动另建