IPdesign-3Dprint — IP Character to 3D Print Pipeline

Overview

Turn any text description or reference image into a 3D-printable IP character figurine. Designed for Pop Mart / Skullpanda-style collectible figures.

User Prompt → [Image Generation Layer] → 3-View References
                                                ↓
User Confirms? → Yes → [Blender Modeling Layer] → Solidify + Subdiv + Color
      ↕ No (adjust prompt)                             ↓
Manual Upload →                          [Export Layer] → STL / OBJ / 3MF

Three Work Modes

The skill auto-detects which image generation method is available:

Default (no GPU, no ComfyUI Cloud): Uses Google Imagen 4.0 via GEMINI_API_KEY.

Prerequisites

Required

# Blender (any version ≥4.0)
apt-get install blender   # Linux
brew install blender      # macOS

# Verify
blender --version

For Image Generation (pick one)

# Mode A: ComfyUI Local — see comfyui skill
# Mode B: ComfyUI Cloud — export COMFY_CLOUD_API_KEY

# Mode C: API-Only (recommended for no-GPU)
export GEMINI_API_KEY="your-key-here"
# Or use any of: EVOLINK_API_KEY, OPENAI_API_KEY, BFL_API_KEY

Pipeline Stages

Stage 1: Image Generation (Three-View References)

Option A: From Prompt Generate three-view character concept images (front/side/back):

# Using Gemini/Imagen 4.0 (API-only mode)
curl -X POST "https://generativelanguage.googleapis.com/v1beta/models/imagen-4.0-generate-001:predict?key=$GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "instances": [{"prompt": "Skullpanda-style character, front view..."}],
    "parameters": {"sampleCount": 2, "aspectRatio": "1:1"}
  }'

Option B: Manual Upload User provides front/side/back view images directly.

Option C: User Confirmation Loop

1. Generate → show images to user
2. User says "OK" → proceed to Stage 2
3. User says "adjust X" → modify prompt → regenerate
4. User provides own reference images → skip generation

Stage 2: Blender Modeling

Based on the reference images, generate a Blender script that:

1. Load references as background planes (front/side/top)
2. Build base geometry:
   • Head: UV sphere → sculpt to Skullpanda proportions
   • Eye sockets: Boolean cutout (rounded triangular shape)
   • Eyes: Sclera + iris + pupil + highlight spheres
   • Cat ears: Spherical buttons (NOT pointed!)
   • Body: Short chunky cylinder (1:1 head:body ratio)
   • Arms/Legs: Short rounded cylinders
   • Tail: Curved thin tube
   • Base: Circular disc with raised rim
3. Apply 3D printing modifiers:
   • Solidify (1.2mm wall thickness)
   • Subdivision Surface (Catmull-Clark, level 1-2)
4. Color/Material:
   • Helmet: White (#F5F0EB), low roughness, slight subsurface
   • Eyeshadow: Dark charcoal (#19191E), matte
   • Eyes: Amber/brown iris with highlights
   • Bodysuit: Black (#1A1A1A), medium roughness
   • Base: Dark gray (#141414), matte
5. Export:
   • STL for 3D printing (manifold, watertight)
   • .blend for further editing

Stage 3: 3D Print Preparation

# Solidify — wall thickness
solidify = obj.modifiers.new(name="Solidify", type='SOLIDIFY')
solidify.thickness = 0.12   # 1.2mm
solidify.offset = -1.0       # inward offset

# Subdivision — smooth surface
subsurf = obj.modifiers.new(name="SubSurf", type='SUBSURF')
subsurf.levels = 1
subsurf.subdivision_type = 'CATMULL_CLARK'

# Export
bpy.ops.export_mesh.stl(filepath="figure.stl", use_selection=True)

Configuration

Default Scale

1 Blender unit = 1 cm
Total figure height: ~10.5 cm (including base)

Key Proportions (Skullpanda standard)

Verifying with Skullpanda

The skill includes a complete Skullpanda build script at scripts/skullpanda_build.py.

blender --background --python scripts/skullpanda_build.py

Expected output:

• skullpanda_base_mesh.stl (~1.6 MB, 16K verts) — 3D print ready
• skullpanda_subdiv1.stl (~6.5 MB, 67K verts) — smooth version
• skullpanda_figure.blend (~1.8 MB) — editable source

Color Palette Reference

Dependencies

Required (auto-installed by setup)

• blender ≥ 4.0

Optional (for image generation)

Common Pitfalls

1. Skullpanda ears are SPHERICAL, not pointed — this is the most common modeling mistake
2. Boolean eye sockets — must be rounded triangular shape (point down)
3. STL too large — use Decimate modifier (ratio 0.15) to reduce face count
4. Memory limit on VPS — skip subdivision rendering; export base mesh STL instead
5. Imagen 4.0 credits — Google AI Studio free tier has usage limits; check your quota
6. Multi-view consistency — generate front view first, then describe for side/back views