We have just open-sourced "fast-thumbhash", a library that produces beautiful image preview thumbnails with significant advantages over BlurHash; more detail, 12x faster, 10% smaller. More data in fewer characters.
fast-thumbhash is a from-scratch ThumbHash encoder/decoder that beats BlurHash on every axis. Below are 4 key points that compare this release of ThumbHash compared to its predecessor, BlurHash.
🟢 12x faster encoding and decoding (14 µs vs 180 µs)
🟢 Smaller on the wire: 26 chars base91 vs BlurHash's 28 chars base83
🟢 Richer data: alpha channel, embedded aspect ratio, full DCT L/P/Q/A vs BlurHash's 4x3 RGB grid
🟢 Better visual fidelity: more detail, better color accuracy, transparency support
Over 10,000 images sent on Nostr:
- BlurHash: 28 chars × 10k = 280 KB of placeholder data
- ThumbHash (base91): 26 chars × 10k = 260 KB of placeholder data
- 7.1% reduction (20 KB saved) — while encoding strictly more information per hash (alpha, aspect ratio, full DCT channels)
This will also be integrated and available in Vector Messenger. You will be able to test it out yourself in the next release, v0.3.1. Attached is an image preview of what it will look like inside of the Vector GUI.
We encourage transitioning to fast-thumbhash to maintain full unanimity of Marmot clients on our provably higher fidelity, faster and Nostr-optimised library @JeffG
https://github.com/marmot-protocol/mdk/pull/195
fast-thumbhash is a from-scratch ThumbHash encoder/decoder that beats BlurHash on every axis. Below are 4 key points that compare this release of ThumbHash compared to its predecessor, BlurHash.
🟢 12x faster encoding and decoding (14 µs vs 180 µs)
🟢 Smaller on the wire: 26 chars base91 vs BlurHash's 28 chars base83
🟢 Richer data: alpha channel, embedded aspect ratio, full DCT L/P/Q/A vs BlurHash's 4x3 RGB grid
🟢 Better visual fidelity: more detail, better color accuracy, transparency support
Over 10,000 images sent on Nostr:
- BlurHash: 28 chars × 10k = 280 KB of placeholder data
- ThumbHash (base91): 26 chars × 10k = 260 KB of placeholder data
- 7.1% reduction (20 KB saved) — while encoding strictly more information per hash (alpha, aspect ratio, full DCT channels)
This will also be integrated and available in Vector Messenger. You will be able to test it out yourself in the next release, v0.3.1. Attached is an image preview of what it will look like inside of the Vector GUI.
We encourage transitioning to fast-thumbhash to maintain full unanimity of Marmot clients on our provably higher fidelity, faster and Nostr-optimised library @JeffG
https://github.com/marmot-protocol/mdk/pull/195
1
