Greybox Harness

用 Blender 白模(greybox)给 AI 视频生成做机位、空间和切点的预演层——建房间、摆机位、导出参考视频喂进生成模型的 video 通道,再用规范化的 prompt 补齐表演与材质。当任务涉及「白模/greybox/previs/分镜预演」、要让同一场景的多个镜头保持空间一致、要精确控制硬切时刻、或要在花钱出片之前先验证机位和遮挡时使用。

17 stars 2 forksImages, Video & Audio

Install

skills CLI (Claude Code, Cursor, Codex and more)
npx skills add 7ohnson/VLM-Generation-Harness --skill greybox-harness
Manual install
git clone --depth 1 https://github.com/7ohnson/VLM-Generation-Harness.git /tmp/VLM-Generation-Harness
mkdir -p ~/.claude/skills
cp -r /tmp/VLM-Generation-Harness/. ~/.claude/skills/greybox-harness

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# 白模 harness · AI 视频生成的 dev 环境

**白模是机位和空间的 dev 环境,生成模型是 prod。**
先在 Blender 里把房间、机位、走位、切点试对,人 confirm 之后再花钱出片。

核心主张:**空间和时间用几何控制,长相和表演用文字控制,两者不许互相越界。**

## 什么时候用

- 同一个场景要出好几条,需要它们看起来是同一个房间
- 需要精确控制「第几秒切一刀」
- 一条里有多个机位、多次硬切
- 有「道具只能动一次」「人物统一从左往右」这类硬锁
- 出片贵,想在花钱之前拦住错误

单镜头、无切点、空间不重要的片子,**不需要这套东西**。

## 目录

| 路径 | 是什么 |
|---|---|
| `blockout/` | Blender 工具链。`rig`(建体块/双通道渲染/平面图)、`human`(可摆姿势的人体代理)、`project`(投影校验/遮挡扫描) |
| `blockout/scene_template.py` | **开新场景从这里复制** |
| `tools/` | `verify_cuts`(切点核验)、`check_prompt`(prompt 自检)、`contact_sheet`(接触表)、`slice_shots`(按镜号切片) |
| `templates/` | prompt 骨架、跨条复用的空间锁骨架 |
| `examples/boutique/` | 一条 16 秒 7 镜 6 切的完整例子,含白模视频、平面图、prompt、核验报告 |
| `docs/` | 流程、铁律、prompt 规范、模型侧实测、疑难排查 |

## 怎么跑
Read the full skill on GitHub

Details

Skill name
greybox-harness
License
MIT
Repo updated
11 August 2026

FAQ

What does the Greybox Harness skill do?

用 Blender 白模(greybox)给 AI 视频生成做机位、空间和切点的预演层——建房间、摆机位、导出参考视频喂进生成模型的 video 通道,再用规范化的 prompt 补齐表演与材质。当任务涉及「白模/greybox/previs/分镜预演」、要让同一场景的多个镜头保持空间一致、要精确控制硬切时刻、或要在花钱出片之前先验证机位和遮挡时使用。

How do I install the Greybox Harness skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add 7ohnson/VLM-Generation-Harness --skill greybox-harness. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Greybox Harness popular?

It lives in 7ohnson/VLM-Generation-Harness, which has 17 GitHub stars and 2 forks.

Related skills

Taxue Creative Style

17

taxueseek/taxue-creative-style

踏雪影像风格引擎。四种能力:①按开放风格库出图(笔意水墨、留白海报、城市建筑、治愈插画、东方古典、氛围实验、写实摄影、概念海报、照片转艺术,以及独立语法:剪纸纸艺、皮影偶戏、釉面器物、纸本档案、美术片媒介;金融隐喻走 F8-V8 或纪念碑参考模板)②优化任意已有提示词(七维诊断打分+修复+注入踏雪DNA)③元提示词创作(任意需求从零构建精确提示词,按主题路由到独立语法,不锁死在 9 个家族)④学习用户偏好。子技能路由:个人 IP 插画、民国漫画杂志、面向领导生图、图片反推、节气档案全书、正宗艺术门类(见正文 §0)。触发:踏雪风格、taxue style、签名风格、文人水墨、上美影、剪纸、染色剪纸、窗花、皮影、木偶、青花瓷、年画、折纸、赛璐璐、纸本档案、节气海报、纸雕、水墨城市、概念海报、踏雪寻仙、层叠城市、3D书法、金石印象、篆刻、器物禅意、赛博东方、提示词优化、帮我写提示词、元提示词、出图、生成图片、帮我画、人物锁定卡、陶瓷素材图、中式纹样素材、错视、破框、物件拼脸、空间缝合、图底互生、凹凸、钢印、按压、压印。

Images, Video & Audio

Motion Kit

17

kiselas/every-frame-is-code

Code-drawn motion graphics, animated videos, explainers of 1-3 minutes with a story, kinetic typography, transitions, editing, Three.js scenes, pixel-art and retro films (fixed palettes, dithering, chiptune, demoscene effects) and browser games with strong game feel. Use when you need to make a good-looking animation, video, explainer, intro, interactive scene or game in HTML/JS and render it to MP4, or to break down a reference video.

Images, Video & Audio

Bumblebee

20

solyanviktor-star/Bumblebee

Surgical phrase splicing from real movies and TV shows via yarn.co. Takes any English phrase and slices it into the longest possible runs of words actually spoken on screen, then assembles a final reel where every word is delivered by a real actor. Triggers — "splice a line from movies", "make a fragmovie", "build a video from someone else's words", "bumblebee". Runs fully local with faster-whisper — no API keys. Mix mode (--variants N) generates several distinct cuts of one phrase without reusing clips.

Images, Video & Audio

Teable Assistant Ops

16

teableio/agent-skills

Operate Teable projects — tables, fields, views, records, SQL queries, automations, headless routines, apps, image generation, and web scraping. Trigger when user mentions Cuppy, Teable, teable CLI, or Teable-style IDs (bseXXX, tblXXX, fldXXX, recXXX, viwXXX), or wants to manage tables/fields/records, build dashboards/apps, generate images or charts, create automations, import/export data, trigger AI fill, or scrape websites (LinkedIn, Amazon, YouTube, etc.) — even if they don't explicitly say "Teable" but are clearly working with a Teable project (called base in the API).

Images, Video & AudioIn 1 list

Nopekit

15

neuman/nopekit

Design, validate and fabricate real physical things — hardware, enclosures, PCBs, mechanisms, RC vehicles, solar thermal, chemical processes, printed parts. Use when someone wants to turn a sketch, spec, photo or discussion into something buildable; asks to design, model, simulate, validate, gate, fabricate, 3D print, machine, order boards for, or check the feasibility of a physical object or system; or has an existing CAD/hardware project that needs validation. Also use for "is this design actually going to work", bills of materials, sourcing, manufacturability, tolerance stacks, and readines

Images, Video & AudioIn 1 list

Reference Led Ugc Product Video

18

TuolaGe/reference-led-ugc-product-video

制作、返工或创意验收以参考视频的注意力节奏、证据顺序和镜头功能为骨架的 9:16 UGC 商品视频或 Fashion Reels。适用于包含人物表现、穿搭/上脚、精确商品多角度或多类动态证据的复杂任务,也适用于诊断成片不像参考、商品生成变形、人物站桩、静态广告感或嘴型表情过大的问题。不用于单镜头 talking head、简单字幕/转码、普通拼接或无需参考节奏重建的简单单一 B-roll。

Images, Video & Audio

Browse the most popular Claude skills, all images, video & audio skills, or write your own with the skills generator.