MAX-786/claude-3d-harness / library/cc/text-to-blender
text-to-blender
VerifiedDrive Blender from natural language. Converts plain-English requests ("model a sword and render it with cinematic lighting", "make this glass look frosted", "set up three-point lighting", "export this scene as glTF for the web") into Blender Python code executed via the Blender MCP server. Acts as the orchestrator that picks and chain-loads specialised sub-skills (blender-modeling, blender-materials, blender-lighting, blender-cameras, blender-rendering, blender-animation, blender-export, wireframe-to-3d, blender-pro-workflow, blender-skill-harmonizer, quality-refinement-autoloop). Use this skill whenever the user wants Claude to do anything in Blender, including creating geometry, applying materials, lighting a scene, framing a camera, rendering, animating, or exporting. Make sure to invoke this skill even if the user does not say "Blender" — also covers requests like "create a 3D model of...", "render this...", "make a glTF from...", "set up a scene with...", or any 3D-creation task. Requires the Blender MCP addon (ahujasid/blender-mcp) running on port 9876.
assets/v0.6.0-round2-validation/README.md
View on GitHub# Round 2 Validation Renders — Sword Scene: Failure Modes → Final
Seven renders documenting the full investigation arc from "recipes verbatim" through every recipe gap surfaced and patched. Each iteration is the raw output Blender produced — none are cherry-picked.
## The investigation arc
| File | Recipe state | Outcome | What it revealed |
|------|-------------|---------|------------------|
| `M_sword_attempt1_broken_world.webp` | Recipes verbatim | Magenta floods scene | World tree leftover from earlier C2 HDRI test; Environment Texture had `image=None` |
| `M_sword_attempt2_clean_world.webp` | After manual world reset | Sword mostly in shadow | Lights at fixed coords with hardcoded rotations; no subject-aware aim |
| `M_sword_attempt3_aimed_lights.webp` | + aimed lights + reframed | Parts visible but blade clips top | Camera lacks bbox-aware framing |
| `M_sword_attempt4_correct_dimensions_wrong_orientation.webp` | + proper sword dimensions (78×4.5×0.8cm) | Looks like a "thin pole" | Camera viewing edge of blade (0.8cm thick), not broad face (4.5cm) |
| `M_sword_attempt5_rotated_no_taper.webp` | + Z-rotated 90° to face broad side | Visible blade but blunt tip | Top vertices scaled to 30%/50% leaves a chiseled flat |
| `M_sword_attempt6_with_pointed_tip.webp` | + merged top vertices to a point | Recognizable sword | Working — needed proper merge after collapse |
| **`M_sword_FINAL_v0.6.0.webp`** | All v0.6.0 patches applied; built fresh | Credible sword | Final result with all patches integrated |
## Six patches that came out of this
1. **`reset_world()` helper in orchestrator** — prevents broken Environment Texture state from corrupting renders
2. **`references/common-object-dimensions.md`** — real-world reference proportions (sword: 78×4.5×0.8cm blade, not "guess")
3. **`aim_at(light, target)` helper in `blender-lighting`** + Recipe 0 (subject-aware three-point)
4. **Bbox-aware camera framing** in `blender-cameras` Recipe 0 — fits subject to ~80% of vertical frame at the chosen focal length
5. **Axis-orientation guidance** in `blender-modeling` for elongated objects (broad face vs thin edge; rotate so broad axis faces camera)
6. **Proper tapering recipe** in `blender-modeling` — collapse top verts AND `remove_doubles` (without merge, the four collapsed verts stay distinct → degenerate "fake point")
## Meta-lesson committed to the orchestrator
The orchestrator's workflow now includes a **mandatory visual validation checkpoint** between rendering and reporting success. Numerical checks (object count, file size, no API errors) passed on every iteration above — but iterations 1, 2, 3, 4, 5 all looked broken. The user catching this is what drove the patches. The orchestrator must look at the picture before declaring done.
See `test_round2.md` at repo root for the full investigation, root causes, and patches.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |