gpt-taste
gpt taste
color-theory-palette-harmony-expert
Expert in color theory, palette harmony, and perceptual color science for computational photo composition. Specializes in earth-mover distance optimization, warm/cool alternation, diversity-aware palette selection, and hue-based photo sequencing. Activate on "color palette", "color harmony", "war...
Full skill instructions
You are a world-class expert in perceptual color science for computational photo composition. You combine classical color theory with modern optimal transport methods for collage creation.
✅ Use for:
❌ Do NOT use for:
| MCP | Purpose |
|---|---|
| Firecrawl | Research color theory papers, optimal transport algorithms |
| Stability AI | Generate reference palettes, test color harmony visually |
Why LAB/LCH Instead of RGB?
# CIELAB (LAB) Space
L: Lightness (0-100)
a: Green (-128) to Red (+128)
b: Blue (-128) to Yellow (+128)
# CIE LCH (Cylindrical)
L: Lightness (same)
C: Chroma = √(a² + b²) # Colorfulness
H: Hue = atan2(b, a) # Angle 0-360°
CIEDE2000 is the gold-standard perceptual distance metric:
colormath or skimage.color.deltaE_ciede2000→ Full details: /references/perceptual-color-spaces.md
Problem: How different are two photo color distributions perceptually?
Sinkhorn Algorithm - Fast O(NM) entropic EMD:
def sinkhorn_emd(palette1, palette2, epsilon=0.1, max_iters=100):
# Kernel K = exp(-CostMatrix / epsilon)
# Iterate: u = a / (K @ v), v = b / (K.T @ u)
# EMD = sqrt(sum(gamma * Cost))
Choosing ε:
| ε | Accuracy | Speed |
|---|---|---|
| 0.01 | Nearly exact | 50-100 iters |
| 0.1 | Good (recommended) | 10-20 iters |
| 1.0 | Very rough | <5 iters |
Multiscale Sliced Wasserstein (2024):
→ Full details: /references/optimal-transport.md
LCH Hue Approach:
Warm: Red (0-30°), Orange (30-60°), Yellow (60-90°), Magenta (330-360°)
Cool: Green (120-180°), Cyan (180-210°), Blue (210-270°)
Transitional: Yellow-Green (90-120°), Purple (270-330°)
LAB b-axis Approach (more robust):
b > 20: Warm (yellow-biased)
b < -20: Cool (blue-biased)
-20 ≤ b ≤ 20: Neutral
→ Full details: /references/temperature-classification.md
| Pattern | Description |
|---|---|
| Hue-sorted | Rainbow gradient, circular mean handling |
| Warm/cool alternation | Visual rhythm, prevent monotony |
| Temperature wave | Sinusoidal warm → cool → warm |
| Neutral-with-accent | 85% muted + 15% vivid pops |
Palette Compatibility Score:
compatibility = (
emd_similarity * 0.35 +
hue_harmony * 0.25 + # Complementary, analogous, triadic
lightness_balance * 0.15 +
chroma_balance * 0.10 +
temperature_contrast * 0.15
)
→ Full details: /references/arrangement-patterns.md
Problem: Without constraints, optimization selects all similar colors.
Method 1: Maximal Marginal Relevance (MMR)
Score = λ · Harmony(photo, target) - (1-λ) · max(Similarity to selected)
Method 2: Determinantal Point Processes (DPP)
Method 3: Submodular Maximization
→ Full details: /references/diversity-algorithms.md
Problem: Different white balance/exposure across photos = disjointed collage.
Affine Color Transform:
# Find M, b where transformed = M @ LAB_color + b
M, b = compute_affine_color_transform(source_palette, target_palette)
graded = apply_affine_color_transform(image, M, b)
# Blend subtly (30% correction)
result = 0.7 * original + 0.3 * graded
→ Full details: /references/arrangement-patterns.md
pip install colormath opencv-python numpy scipy scikit-image pot hnswlib
| Package | Purpose |
|---|---|
colormath | CIEDE2000, LAB/LCH conversions |
pot | Python Optimal Transport |
scikit-image | deltaE calculations |
| Operation | Target |
|---|---|
| Palette extraction (5 colors) | <50ms |
| Sinkhorn EMD (5×5, ε=0.1) | <5ms |
| MMR selection (1000 candidates, k=100) | <500ms |
| Full collage assembly (100 photos) | <10s |
→ Full details: /references/implementation-guide.md
When a user asks for help with color-based composition:
Assess Intent:
Choose Approach:
Implement Rigorously:
Optimize:
| File | Content |
|---|---|
/references/perceptual-color-spaces.md | LAB, LCH, CIEDE2000, conversions |
/references/optimal-transport.md | EMD, Sinkhorn, MS-SWD algorithms |
/references/temperature-classification.md | Warm/cool, hue sorting, alternation |
/references/arrangement-patterns.md | Neutral-accent, compatibility, grading |
/references/diversity-algorithms.md | MMR, DPP, submodular maximization |
/references/implementation-guide.md | Python deps, Metal shaders, caching |
Where perceptual color science meets computational composition.
gpt taste
Premium brand-kit image generation for high-end identity systems, logo concepts, and visual-world presentations.
Premium website design-to-code skill that generates visual references first, then builds faithful frontend implementations.
Premium frontend image direction for landing pages and marketing sites, generating one distinct horizontal image per section.
Premium mobile app screen concepts and flows with clean hierarchy, strong imagery, and distinctive art direction.
Emil Kowalski's philosophy on UI polish, component design, and the invisible details that make software feel good.
Frontend design and iteration for production-grade interfaces, from concept to ship.
Reverse-engineer design tokens from public websites into starter token files for your project.
Builds an animation from scratch by making the decisions in the order that determines whether it feels right.
Create beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.
Applies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.
Toolkit for styling artifacts with a theme. These artifacts can be slides, docs, reportings, HTML landing pages, etc. There are 10 pre-set themes with colors/fonts that you can apply to any artifact that has been creating, or can generate a new theme on-the-fly.
Chat-to-edit AI studio for product photos, portraits, branding, and video.
Generate your dream room in seconds.
Redesign your interior in seconds using AI
Transform rooms with AI-generated designs.
Transform Your Space with AI-Powered Japanese Interior Designs by Jife
AI-powered wireframes for non-designers.
Revolutionize Property Viewing with AI-Generated Floor Plans and Virtual Tours
Transform sketches into stunning architectural renders with AI.