shrinks99

8 Nvidia B200s. Crazy from an individual perspective, costly for a small business!

The Library of Congress runs Webrecorder's Python Wayback for their web archive replay and has an extensive collection of over 35,000 web archives each comprising multiple pages: https://blogs.loc.gov/thesignal/2025/01/beta-release-of-libr...

Pretty cool! I've also been playing around with GLM 5.2 this week and was equally impressed. At work we're running it locally on some crazy expensive hardware as a test before starting another project so it's great to see people taking this massive FOSS model release and running it on an average machine, even if it's not terribly practical at this point.

Nice work!

I prefer CoMaps' tile styling, they did a bunch of updates to it early on to differentiate it from OrganicMaps, IMO it's much nicer.

CoMaps also has a better system for distributing map data now which isn't bound to app updates. Not sure if this has made its way back to the other app, but it's quite nice.

I've been using GLM 5.2 a lot this past week, it's been replacing Opus 4.8. I mostly do front-end web development and haven't noticed much of a quality difference.

Sure, "it's just frontend", but that's actual use enough for me to take it seriously.

Yes, CoMaps lets users submit basic data that updates OSM for addresses and POIs. It's a harder problem than you may initially think too, quite often the delivery format of OSM data differs in schema from the DB format. In the case of OrganicMaps and CoMaps, they both generate offline-optimized map formats for distribution.

Navidrome is so sick! I've been a full album listener for years, but switching to it was a great excuse to make sure my music library was FLAC only and tagged with good MusicBrainz metadata. If you're a self-hosting / NAS person already and you have a music library of digital files it's a no brainer.

Brand guidelines and web design pretty much don't exist any more as far as I can tell. Gotta get it out yesteday, and the only way to do that is vibe coding, styling be damned.

tangled.org writes issues as atproto data that lives in a user's PDS which is one neat idea.

There's absolutely no precident for Blender Foundation sponsorships leading to such things... So no, they probably won't do that.

It compiles almost instantly which makes this visual design by Claude instruction viable. That alone is pretty neat.

It can't render all your math in exactly the same way. If you need it to do that, it may not be for you.

Noticed that Jellyfin had inched out Plex when sorting by popularity on the TrueNAS app catalogue the other day (45,178 installs vs Plex's 42,225). The existance of this project seems to confirm that the dev ecosystem around it is getting stronger!

"Why should I change? He's the one who sucks!"

I got Claude to make me the exact same graph a few weeks ago! I had hypothesized that we'd see a sharp drop off, instead what I found (as this project also shows) is a rather messy average trend of outages that has been going on for some time.

The graph being all nice before the Microsoft acquisition is a fun narrative, until you realize that some products (like actions, announced on October 16th, 2018) didn't exist and therefore had no outages. Easy to correct for by setting up start dates, but not done here. For the rest that did exist (API requests, Git ops, pages, etc) I figured they could just as easily be explained with GitHub improving their observability.

I own one of these, it is amazingly well manufactured. Not cheap, but (with the exception of the power button) well made.

Yes, but you can get LLMFit to recommend hardware requirements with `llmfit plan --context <TOKENS> <MODEL>`.

I'm going to guess this is because the image to depth data, while good, is not perfectly accurate and therefore cannot be a shared ground truth between multiple images. At that point what you want is a more traditional structure from motion workflow, which already exists and does a decent job.

I think SideFX was the first to do that with Houdini. It's one of my favourite micro-UX features of high-end graphics software, coming in at a close second to Nuke's use of both linear and non-linear scales for slider values.

Yes, and when working with footage shot with anamorphic lenses one will have to render the footage as non-square pixels, mapped to the square pixels of our screens, to view it at its intended aspect ratio. This process is done either at the beginning (conforming the footage before sending to editorial / VFX) or end (conforming to square pixels as a final step) of the post-production workflow depending on the show.

What a bummer. It seems like what they're asking for here (a written agreement that users will be able to access 3rd party app stores) would be a win win win for Core Devices, Rebble, and users. Core Devices gets to look like a super good guy (ideally driving interest in the product), Rebble gets to look like a huge winner maintaining something for the community (as they are), and users get an open ecosystem.

There's still a chance for a win here, but looks like the door is closing.

Pixar also ships Renderman support for Blender https://rmanwiki-26.pixar.com/space/RFB26

You can see everything in your field of vision, but the area DIRECTLY in the centre has the highest level of detail. This image has high frequency animated details that are not cognisized equally by your entire FOV. The animated bit right in the middle at any given time is where your brain processes the most detail and also where you are looking.

Generally yes, but we're still working on it all these years later! This article by Chris Brejon offers a very in-depth look into the differences brought about by different display transforms: https://chrisbrejon.com/articles/ocio-display-transforms-and...

The "best" right now, in my opinion, is AgX, which at this point has various "flavours" that operate slightly differently. You can find a nice comparison of OCIO configs here: https://liamcollod.xyz/picture-lab-lxm/CAlc-D8T-dragon

Most of Blender's icons are actually made in Penpot which is also what the Blender foundation uses for UI prototyping. The brush icons are made in Blender though!

https://penpot.app/penpothub/libraries-templates/blender-con...

https://code.blender.org/2024/11/new-brush-thumbnails/

Serif (I guess Canva now) maintains their own which uses the Lensfun database.

Yeah, but Lensfun (the library they use for this) doesn't have anywhere to donate.

I'd buy some of these explinations, except the depth estimation, colorization, and super-resolution ML models they use in the app DO run locally and are still subscription-gated.

Apple has been doing on-device machine learning for portrait blurs and depth estimation for years now, though based on the UI, this might use cloud inference as well.

Granted, these aren't the super heavy ones like generative fill / editing, and I understand that cloud inference isn't cheap. A subscription for cloud-based ML features is something I'd find acceptable, and today that's what has launched... The real question is what they plan to do with this in 2-5 years. Will more non-"AI" features make their way into the pro tier? Only time will tell!

Cole (the author of didder) also has a GUI version called Dithertime: https://makew0rld.itch.io/dithertime

A friend of mine wrote one for Go with the goal of creating the best and most complete dithering library out there and I think he did a decent job. Worth a look if you're looking for reference implementations!

Here's his: https://github.com/makew0rld/dither

RIP a ton of older audio plugins.