The open, community-driven leaderboard for AI avatar generation.
Avatar Arena is the LMArena for avatars — a public benchmark where AI avatar platforms are ranked through blind, side-by-side human evaluation. The goal is to become the definitive reference for avatar quality rankings, giving enterprises, creators, and developers an unbiased source of truth when choosing an avatar platform.
The AI avatar market is growing fast, but there's no standardized, public benchmark for quality. Marketing demos are cherry-picked. Review sites are pay-to-play. Avatar Arena fixes this by letting the community decide — with transparent methodology and open data.
- Blind Battles — Two AI avatar videos (generated from the same script) are shown side by side. The viewer doesn't know which platform made which video.
- Community Voting — Users vote on which avatar looks and sounds better (or declare a tie).
- Elo Rankings — Votes feed into a Bradley-Terry / Elo rating system (the same approach used in chess and LMArena) to produce a live, converging leaderboard.
- HeyGen
- Synthesia
- Google Vids
More platforms will be added over time (D-ID, Tavus, Colossyan, Hour One, etc.).
- Next.js 16 (App Router, TypeScript)
- Tailwind CSS for styling
- Elo rating system with 95% confidence intervals
- In-memory store (database integration coming soon)
npm install
npm run devOpen http://localhost:3000.
src/
├── app/
│ ├── page.tsx # Landing page
│ ├── battle/page.tsx # Blind voting UI
│ ├── leaderboard/page.tsx # Live rankings
│ └── api/
│ ├── battle/route.ts # Serve random video pairs
│ ├── vote/route.ts # Record votes, update Elo
│ └── leaderboard/route.ts # Return ranked competitors
├── components/
│ ├── Navbar.tsx
│ ├── VideoPlayer.tsx
│ └── LeaderboardTable.tsx
└── lib/
├── types.ts # TypeScript interfaces
├── elo.ts # Rating math + confidence intervals
└── store.ts # Data store
Avatar Arena is open source. Contributions welcome — especially help with:
- Generating standardized test videos across platforms
- Adding new competitor integrations
- Building out the database layer
- Improving anti-gaming measures
MIT