Conversation
Messemger apps recpmpressing my already blurry 6kb pictures is giving me trust issues
2
2
3
@snacks the worst part is its self-inflicted. like if file transfers were just p2p it wouldnt be an issue but we literally cannot do that because ipv4 and isps being.... super unreliable in regards to protecting their customers
1
0
2
@snacks isps when you expect them to not leak your irl home address all over the fucking place: catNono
0
0
1

@snacks IRC (v3 http file hoster extension) yet again does nothing and wins

2
0
2
@mia i understand recompressing and downscaling large stuff. But that saturn pic has like 100 usable pixels, leave them alone reee
2
0
1
@mia but yeah, doing nothing wins here. I'd expect if bigger networks adapt that standard they're gonna start doing the same shit as every other big website tho
1
0
1
@mia actually, does 4chan still reencode images after getting pwned by that?
1
0
1
@mia didn't even resize them afaik
0
0
1

@snacks you understand? because i don’t. thumbnails sure but if i upload a picture i want that to be left alone and not subjected to incorrect processing and webpee blur

2
0
2
@mia ig you can also just set a snae upload size limit, thatcs prob the better option but then normies will take 100mpix images with their phone and not be able to upload them
0
0
2

@snacks and for data caps and low bandwidth they should use AVIF with progressive loading (if browsers fuckin implemented it) because that needs veery little data to look decent

1
0
3
@mia @snacks my biggest issue with AV1 is that the spec is truly monstrous in size.
2
0
0

@icedquinn @snacks doesn’t seem to be an issue in practice, judging by how many different encoders and decoders there are for it

1
0
1

@icedquinn @snacks AV1 is also designed to be efficient to implement in hardware. jxl isn’t designed to be efficient to implement period

0
0
1
@mia @snacks i have a feeling somewhere down the line we're going to end up with tensor codecs. there was some semi-recent R&D papers where they were basically just overfitting an autoregressor to an image *as the block codec itself* and storing residuals on the end. it was highly competitive and doesn't need nearly as much machinery to decode.

i had two of those papers but apparently i saved them stupid or i'd link them
0
0
1
@mia @snacks nostr does basically the same thing with blossom. it's basically just a boring "http put get checksum back" and "http get blob from checksum" and you toss the blossom link in the messages.
0
0
0