Blender 资产生成

Generates ecommerce/game/teaching 3D assets via Blender scripts (Blender 4.2+ including 5.x). Supports cloud Windows and local machines; reference-image guid...

Security Scan

The skill's files, instructions, and requirements are consistent with a Blender-based 3D asset generator and do not request unrelated credentials or perform unexpected network/activity.

Purpose & Capability

Name/description (Blender asset generation) aligns with the included artifacts: Blender Python scripts (build/render/export/donut_build), run scripts for macOS/Windows, spec.json/schema, and reference docs. There are no unrelated environment variables, binaries, or config paths requested.

Instruction Scope

SKILL.md and the referenced docs limit actions to normal asset-generation tasks: normalize inputs to spec.json, produce Blender scripts, run Blender in background mode, generate renders and exports, and produce logs. The documents mention optional 'vision-assisted' steps but do not include any code that accesses camera hardware, system secrets, or external endpoints; vision use is conditional on a vision-capable agent/environment.

Install Mechanism

There is no install spec and no network downloads; the skill ships a project skeleton and Blender scripts. All actions write local project files and call the Blender binary — no external installers or extract-from-URL steps are present.

Credentials

The skill requests no credentials or special environment variables. It reads/writes files within the project folder (spec.json, scene.blend, exports, renders, run-log.txt) which is expected for this purpose. No secret-named env vars or unrelated service tokens are requested.

Persistence & Privilege

always is false and autonomous invocation is not disabled (the default); this is normal for skills. The skill does not attempt to modify other skills or global agent config; its footprint is limited to project files it creates/updates.

Assessment

This package appears coherent for generating Blender 3D assets. Before running: (1) ensure you have a compatible Blender binary installed and point BLENDER_BIN if needed; (2) review the included Python scripts yourself (they are plain bpy scripts that read spec.json and write files) and run them in a controlled environment; (3) if you plan to run this on a cloud PC, confirm the cloud runtime permits executing Blender and storing outputs there; (4) note that 'vision-assisted' behavior depends on a vision-capable agent/runtime — the skill does not itself access cameras or external servers. If any of those are unacceptable, do not run the scripts.

Blender Asset Generator (Blender 4.2+ / 5.x)

This Skill generates a reproducible asset project that can be executed in:

• Cloud PC (Windows) as the primary runtime
• Local computer (Windows or macOS) as an alternative runtime

The Skill outputs:

• A structured spec (spec.json) describing the asset
• Executable Blender Python scripts (scripts/*.py) using bpy
• Run scripts for Windows/macOS
• Export artifacts: GLB (required), FBX (optional), USDC (optional), plus preview renders

Quick Start

1. Choose a mode:

   • Text-only Mode (no image or no vision): fast and deterministic
   • Vision-assisted Mode (reference images + multimodal model): extract structure/dimensions from images, then confirm with user

2. Provide input:

   • Required: object name + use case (ecommerce / game / teaching) + desired output formats (GLB required, FBX optional)
   • Optional: target size (at least one anchor dimension), poly budget, material style, reference images

3. The Skill will:

   • Normalize input into spec.json
   • Generate scripts/build.py, scripts/render.py, scripts/export.py (or update them)
   • Provide Cloud (Windows) and Local (Windows/macOS) run commands
   • Generate a QA checklist and a final evaluation

Input Template

Blender Version: 4.2+ (supports 4.2 and 5.x)
Primary Runtime: Cloud PC (Windows)
Output Priority: GLB required; FBX optional
USD: USDC optional (via file extension)

Use Case: ecommerce | game | teaching
Asset Name:
Asset Category: (e.g., bottle / lunch box / hanger / simple prop / teaching diagram)

Text Description (required):
- What it is:
- Key parts/components:
- Material style (optional): plastic/metal/glass/wood/fabric
- Target realism: low/medium/high

Dimensions (recommended):
- Provide at least ONE anchor dimension, e.g. height=240mm
- If unknown: provide category-typical size OR accept default size

Constraints (optional):
- Poly budget (triangles):
- UV: required | optional | no
- Texture placeholders: yes | no
- Naming convention: (optional)

Reference Images (optional):
- Attach images (front/side/top) if available
- If images are provided, say whether there is a scale reference in the image:
  - e.g. A4 paper / coin / ruler / hand / known product standard size

References

Read these as needed:

• Workflow: references/workflow.md
• Ecommerce reference-image template: references/ecommerce-reference-guided-template.md
• Degradation strategy (no vision available): references/degradation-strategy.md
• QA checklist: references/qa-checklist.md
• Presets (common defaults): references/presets.md

Output Contract

Always output in this structure:

1. Project Package
   • File tree (what to create/update)
   • spec.json (final)
2. Run Instructions
   • Cloud PC (Windows) command(s)
   • Local (Windows) command(s)
   • Local (macOS) command(s)
3. Expected Artifacts
   • scene.blend
   • exports/*.glb (required)
   • exports/*.fbx (optional)
   • renders/*.png (required: at least 2 views)
   • run-log.txt (required)
4. QA + Evaluation
   • Checklist results
   • Verdict: Achieved | Partially Achieved | Not Achieved
   • Top fixes

Behavior Rules

• Determinism: prefer parametric geometry and explicit settings; avoid random seeds unless user asks.
• Evidence: do not invent certifications/performance claims; in 3D context, do not assume hidden dimensions from a single photo.
• Reference images:
  • If a multimodal model is available, extract a draft structure card and dimension hypotheses, then ask for confirmation of at least one anchor dimension.
  • If vision is NOT available, never block the workflow. Use Text-only Mode + ask for a minimal clarification form (see references/degradation-strategy.md).
• Units: default to millimeters in the spec; enforce unit conversion in scripts.
• Exports: always export GLB; export FBX only when requested.

Assets (Project Skeleton)

Use the project skeleton to generate a runnable folder:

• assets/project-skeleton/spec.schema.json
• assets/project-skeleton/spec.example.json
• assets/project-skeleton/scripts/build.py
• assets/project-skeleton/scripts/render.py
• assets/project-skeleton/scripts/export.py
• assets/project-skeleton/run_local_windows.ps1
• assets/project-skeleton/run_local_macos.sh
• assets/project-skeleton/run_cloud_windows.ps1