Anima
A quiet gallery for how you feel
You choose how you feel on Plutchik’s wheel of emotions, and how strongly, and Anima opens a slow gallery walk, one artwork at a time, from the world’s museums, Indian art traditions and NASA’s cosmos.
CLIP embeddings and FAISS search, re-ranked for beauty and cultural balance; served with no GPU.
Code on GitHub- Type
- Personal project
- Year
- 2026
- Role
- Solo, design to deployment
- Model
- CLIP ViT-L/14 · FAISS
- Collection
- The Met, Cleveland Museum of Art, NASA, Indian art
Curation, not search
A vision-language retrieval system, re-ranked so every room feels like a museum.
- i
Mood match
Every artwork is a 768-d CLIP ViT-L/14 embedding; each mood is a tuned text prompt in the same space, and FAISS finds the closest works.
- ii
Aesthetic quality
A LAION aesthetic predictor scores each work, normalised within each world region, so its Western, photographic bias doesn't demote folk art, miniatures or sculpture.
- iii
Variety across cultures
Maximal Marginal Relevance penalises both look-alikes and a single tradition taking over, so a room spans the world.
- 1,600+
- works indexed
- 24
- moods · 8 emotions × 3 intensities
- $0
- to serve · no GPU
Stack: CLIP ViT-L/14, open_clip, FAISS, LAION aesthetic predictor, NumPy, Python, FastAPI, React, TypeScript, Vite, Canvas, SVG, Render