design-judge

A/B test two designs with 10 AI judge personas. Each judge evaluates both images from a distinct expert lens and picks a winner. A synthesizer tallies votes and gives a final verdict with confidence and risks.

npx design-judge imageA.png imageB.png --context "landing page hero, goal: signups"

Install

npm install -g design-judge
# or just use npx

Usage

export ANTHROPIC_API_KEY=your_key_here

# Local files
npx design-judge old.png new.png --context "mobile checkout flow"

# URLs
npx design-judge https://example.com/v1.png https://example.com/v2.png \
  --context "landing page hero, goal: signups" \
  --label-a "Current" --label-b "Redesign"

# Fewer judges, smarter model
npx design-judge a.png b.png --judges 5 --model claude-sonnet-4-5

# Fast judges, smart synthesis
npx design-judge a.png b.png --model claude-haiku-4-5 --synth-model claude-sonnet-4-5

Options

| Flag | Description | Default | |------|-------------|---------| | --context, -c | What the design is for | "A digital product" | | --label-a | Name for Design A | "Design A" | | --label-b | Name for Design B | "Design B" | | --judges, -j | Number of judge personas 1–10 | 10 | | --model, -m | Model for judges | claude-haiku-4-5 | | --synth-model | Model for synthesis | same as --model |

Output

• Per-judge scores (A vs B) with specific visual observations
• Vote tally
• 🏆 Winner with reasoning + confidence
• ⚠️ Things to watch in the winning design
• 🥈 Losing design's biggest strength

Judge personas

| Persona | Focus | |---------|-------| | UX Designer | Hierarchy, flow, clarity | | End User | First impression, instinct | | Conversion Expert | CTR, funnel, action | | Brand Strategist | Consistency, tone, positioning | | Accessibility Advocate | Contrast, readability, tap targets | | Developer | Responsiveness, edge cases | | Skeptic | What breaks or confuses | | Marketer | Which design sells harder | | Mobile-First Judge | 375px screen, thumb reach | | Contrarian | Non-obvious strengths |

Accepts local file paths (png, jpg, gif, webp) or URLs. Requires ANTHROPIC_API_KEY.

Requirements

• Node.js 18+
• ANTHROPIC_API_KEY environment variable

License

MIT