blender-modeling
Skill0 starsChanged 22 days agoMITView on GitHub
用本机 Blender(headless bpy 脚本)把自然语言提示词做成真实可交付的 3D 模型:解析需求→结构拆解→写 bpy 建模脚本→无头多角度渲染→看图自检迭代→导出 .blend/.glb/.stl 与预览图;也支持打开已有 .blend 修改后重导出。用户说「用 Blender 建模」「做个 3D 模型」「建个模」「低多边形××」「3D 场景」「改模型」「导出 glb/stl/blend」等时调用。完整规范在 AGENTS.md,动手前必须整篇读完。
6 other projects use a version of this.
Adopted by
Most recently taken up by
- robgallardof/blender-forgetheir own version, first seen Nov 1, 2026
- TraX22/HydraOps-Skillstheir own version, first seen Oct 6, 2026
- sixtysevenlf/dsh-skill-blender-modelingtheir own version, first seen Oct 4, 2026
- haw2fregel-lab/claude-in-blendertheir own version, first seen Aug 24, 2026
- poorvith-mp/skills-gamedevtheir own version, first seen Aug 19, 2026
- keysforthewin/thisofficetheir own version, first seen Jul 24, 2026
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- sixtysevenlf/dsh-skill-blender-modeling★ 21same name
- poorvith-mp/skills-gamedev★ 4same name
- haw2fregel-lab/claude-in-blender★ 2same name
- TraX22/HydraOps-Skills★ 1same name
- keysforthewin/thisoffice★ 0same name
- robgallardof/blender-forge★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
Files
Blender 建模(blender-modeling)
把用户的一句提示词变成真实、可交付、可复现的 3D 模型:几何 + 材质 + 灯光相机 + 多角度渲染图 + 导出文件,全部由本机 Blender 以 headless bpy 脚本完成。
用户输入的提示词就是建模目标。如果本技能被调用时没有拿到具体提示词,先向用户要一句描述(主体 + 风格 + 用途),不要凭空猜。
四条铁律(违反任一条视为任务失败)
- 先读规范再动手。 本文件只是入口,完整流程、命令与验收标准在
AGENTS.md。动手前必须整篇读完,并严格按其「阶段 0 → 阶段 9」执行。 - 必须真看图再交付。 每轮渲染后用
read_image打开renders/view-*.png逐张检查;禁止只看日志就宣称模型没问题。 - 必须过闸门。
scripts/check_output.py0 ERROR(含视觉指标闸门与迭代记录校验)且scripts/doctor.py无 FAIL,才允许向用户交付。 - SAFE_MODE 边界。
model.py禁止subprocess/os.system/eval/exec/网络库/__import__/shutil拷贝;写文件只允许白名单目录或单段report.json/brief.md,禁止绝对路径、盘符相对路径(C:evil.txt)、..与无法静态确定的变量路径(含Path.open/write_text);交付前python scripts/safe_check.py必须 0 ERROR。
最小工作流(细节以 AGENTS.md 为准)
- 环境体检:
python scripts/doctor.py(有 FAIL 先按提示修环境,再开工;纯文件级视觉探针可加--vision,零依赖验证渲染图真的可解码/尺寸对/非空白) - 解析提示词 → 写建模简报
<输出目录>/brief.md(默认不问用户,把假设写清楚) - 复制
templates/model_template.py为<输出目录>/model.py,用lib/blender_kit.py的 API 实现建模 - build 自检:
python scripts/run_blender.py --script <输出目录>/model.py --out <输出目录> --stage build - 渲染预览后看图自检:read_image 打开
<输出目录>/renders/view-*.png,对照提示词找问题;量化值看report.json.vision(主体占比/四边留白/底部留白/亮度) - 迭代修正(最多 4 轮,每轮只改 1-2 类问题;迭代记录写清「看了哪张图 + 量化值」,可选
python scripts/snapshot.py --out <输出目录>打快照;--rollback只还原 model.py/renders/report.json,不还原 exports/,回滚后重跑--stage full)旧项目没有
report.json.vision时可python scripts/vision_metrics.py --out <输出目录> --write补写(标记backfill,只校验图片引用;下次--stage full后严格校验生效)。 - 终渲导出:
--stage full→ blend/glb + 最终渲染图 +report.json - 交付闸门:
python scripts/check_output.py --out <输出目录>+python scripts/safe_check.py <输出目录>/model.py - 向用户交付:文件路径、怎么看预览、还能怎么改
改闸门/想证明闸门有效时:
python scripts/gate_bench.py对examples/两个合格产物做 17 种确定性变异(共 34 条),逐条交给check_output/vision_metrics/vision_probe判定,输出「变异 → 是否被拦 → 命中错误码」与抓取率;有逃逸 exit 1。本机实测 34/34 拦截、100%。单张渲染图存疑时用python scripts/vision_probe.py --dir <输出目录>/renders零依赖定位到具体文件与原因。
改已有模型(模式 B):不是从第 1 步开始——用
templates/edit_model_template.py复制出model.py,通过--blend <源.blend>打开现成模型,只填EDITS字典,然后同样走 build → render → full → 闸门。详见AGENTS.md第 10.5 节。
路径约定
<技能目录>:本文件所在目录(skill 工具会返回 base directory;scripts/、lib/、templates/都在这里)<工作区>:用户当前项目根目录(cwd)<输出目录>:<工作区>/blender-models/<slug>/(slug 全小写连字符,如lowpoly-mug)- 产物禁止写进技能目录;
model.py里禁止硬编码 Blender 路径(执行器自动定位)
参考文档(按需读,不要一次全读)
| 文档 | 什么时候读 |
AGENTS.md | 动手前必读:完整流程、每阶段命令与验收条件 |
references/prompt-to-model.md | 阶段 1-2:提示词怎么翻译成建模简报与部件表 |
references/modeling-recipes.md | 阶段 3:常见题材的部件拆解与代码骨架 |
references/architecture-modeling.md 等 4 篇 | 阶段 2-3:建筑 / 人物 / 动物 / 植物题材专用——结构拆解、真实比例尺寸表、技巧库、配方与实测记录(architecture-modeling.md、character-modeling.md、animal-modeling.md、plant-modeling.md) |
references/bpy-cookbook.md | 阶段 3-4:bpy 4.0 实测 API 与坑 |
references/render-and-check.md | 阶段 5-6:相机灯光参数与看图自检清单;含 vision_probe.py / gate_bench.py 的「可机器验证」用法 |
references/troubleshooting.md | 报错时查:症状 / 原因 / 解决 |
references/export-and-integration.md | 导出和接引擎:格式怎么选、单位与坐标系、Three.js/Unity/UE/3D 打印 |
templates/edit_model_template.py | 模式 B:改已有 .blend 的模板与 EDITS 写法 |
examples/ | 需要 few-shot 参考时看两个完整实测样例 |
交付时告诉用户
模型是什么、文件在哪(.blend / .glb / 渲染图)、怎么在 Blender 里打开或导入游戏引擎、哪些是简化处理、下一轮可以怎么改。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

