✳flâneur — a map of the web's best reading

River, an embedding-based exploration interface | max-bittker

maxbittker.com · 1,859 words · saved by 1 readers

River is an exploration engine, allowing you to navigate rapidly through slices of visual culture via machine learning based connections. Try it out ↗ This post is a place for me to put down some notes on the design and technical architecture. River is in some ways my way of leveraging ML capabilities to fill negative space left unexplored by generative diffusion models. Text to image tools are exciting, but they lack the sturdiness and control required the sense of craft that I’m looking for - they still feel more like a slot machine than a pencil. I expect many of these problems are solvable, but when something is new as image diffusion, it’s not clear which limitations are incidental and which are innate, posing interesting challenges for interface design. It’s hard to make good tradeoffs when the underlying capabilities change so quickly! It makes sense that we end up with tools like Midjourney (opinionated and minimal) or Automatic1111, an evolved mess of nested tabs with infinite

River, an embedding-based exploration interface | max-bittker max bittker projects blog roll shelf River, an embedding-based exploration interface September 20, 2023 River is an exploration engine, allowing you to navigate rapidly through slices of visual culture via machine learning based connections. Try it out ↗ This post is a place for me to put down some notes on the design and technical architecture. Designing for Diffusion: River is in some ways my way of leveraging ML capabilities to fill negative space left unexplored by generative diffusion models. Text to image tools are exciting, b

Explore this link on the map →

related reading