Hoshino Lina (星乃リナ) 🩵 3D Yuri Wedding prep!!!

Hacker & Mad Scientist VTuber!

💍 Cyan Nyan 星乃しお 🩵✨ 3D Yuri Wedding 2026 (npub1rzf…w4kn) 🩵 | We are Tokyo Hacker Girls (npub1k6r…8c3g) ✨

ママ : https://bsky.app/profile/heripiro.bsky.social | パパ : https://twitter.com/otozuki_teru
All the links: https://lina.yt/me

YouTube: https://lina.yt
Twitter: https://lina.yt/twitter
Bluesky: https://lina.yt/bsky
GitHub: https://lina.yt/gh
Sponsor Me!: https://lina.yt/sponsor
Buy me a Ko-Fi!: https://lina.yt/kofi

Public Key

npub1cxqje22r8ewnr30rt625d58cr8gpdjnw405cq5765m2qsvghnmpquz6wtz

NIP-05 Address

lina_at_vt.social@momostr.pink

Profile Code

nprofile1qqsvrqfv49pnuhf3ch34a92x6rupn5qkefh2h6vq20d2d4qgxyteassprpmhxue69uhhyetvv9ujumt0d4hhxarj9ecxjmntqy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyfqzfqs

Publishing to

relay.momostr.pink

relay.mostr.pub

nostr.wine

Author Public Key

npub1cxqje22r8ewnr30rt625d58cr8gpdjnw405cq5765m2qsvghnmpquz6wtz

Show more details

Last Notes

2026-10-30 09:44:48 UTC

- reply

It is definitely possible but you need asynchronous realtime resampling (which OBS doesn't do) and a pretty careful stack of drivers/etc.
However, that isn't actually necessary for perfect karaoke sync most of the time, the fundamental fix here is different!

2026-10-30 09:42:52 UTC

- reply

So the fun thing about budget USB stuff is sometimes it uses the USB clock to sync, so amusingly enough, if they're all on the same bus they might magically sync together!

2026-10-30 07:31:16 UTC

I'm a longtime user of the LSP audio plugin suite on Linux, and after switching to macOS I'm realizing that... there's nothing quite like it, certainly not free.
https://lsp-plug.in/?page=plugins
They have a TON of great plugins!
Should I sign up to get win/mac officially supported?

2026-10-30 06:18:15 UTC

- reply

BTW, this is not a dig at OBS. OBS was not developed for real-time audio production and synchronized mixing. It is simply the wrong tool for this job.
It's fine if you're just streaming games or zatsu, just not singing.

2026-10-30 06:14:37 UTC

- reply

BTW, practically speaking, this is largely a non-issue with PipeWire too. It handles non-locked audio I/O with internal resampling and does a damn good job at it. Whatever sync offset remains, your brain will take care of as long as you do what I said above.

2026-10-30 06:12:29 UTC

- reply

This happens even with native JACK OBS sources and even with a single audio interface.
And the problem can be solved even if you use multiple audio interfaces, because your brain is ✨magic✨!
I'll explain in the video but the TL;DR is that your brain does magical audio latency compensation (within reason) internally, so even if you have some variability from startup to startup, you'll still sing on beat as long as the total delay is not too large and (this is the important bit) you are HEARING the same mix you are STREAMING.

2026-10-30 06:06:41 UTC

- reply

To fix this, you need to mix your stream audio outside of OBS! You can either:<li>Use a hardware mixer</li><li>Use a DAW</li><li>Use dedicated streamer audio software</li><li>Use an OBS plugin that can internally mix & monitor the audio reliably (like atkAudio PluginHost2)</li>
In order to ensure reliable karaoke sync (and completely eliminate the need for manual sync adjustment), you should always MIX mic+inst first, then MONITOR that mix to your headphones live & send the SAME mix to the stream.
That will give you PERFECT karaoke sync, every time!
I will be making a video tutorial explaining this soon, so look forward to that~ ✨

2026-10-30 06:06:20 UTC

VTubers and VSingers!
⚠️ If you use OBS to mix your audio, it is messing up your vocal sync and making you sound OFF BEAT! ⚠️
You get a RANDOM offset every time you restart OBS, of 30ms or more.
There is NO WAY around this as long as you use plain OBS (without dedicated mixing plugins) for your inst/vocal mixing.
If you have separate sources in OBS for your mic/inst, you are affected.
OBS monitoring won't help you notice, because it has its own DIFFERENT random offset!
This screenshot is showing recordings made with OBS on Windows 10, with both MIC and BGM audio sources coming from VAIO virtual cables, driven by the same VoiceMeeter instance.
This is the BEST possible scenario. With different audio source types, it could be even worse!
This happens because OBS does NOT do ANY synchronization of audio coming from different sources.
It just YOLOs it, so every time you start up OBS, the audio input from each source starts at a different point in time.
You can adjust the sync, but next startup it'll change.
https://media.vt.social/media_attachments/files/117/528/347/039/400/316/original/2c5f3d490c485f67.png

2026-10-29 08:06:13 UTC

- reply

Update: She just fell asleep with her hand on my boob again.
Every now and then she squeezes in her sleep.
ヤチヨパワーすげ〜

2026-10-29 07:56:45 UTC

*Cuddling in bed*
Cyan: zzzz
Me: *Reading phone*
Cyan: *Hugs and cuddles closer*
Cyan: *Moves hand to my boob*
Me: ???
Cyan: *Sleep talking* Yachiyo~~
Me: ???????????

2026-10-29 03:50:12 UTC

- reply

Easiest PQC mitigation would be a PSK on top. Much easier than PQC asymmetric crypto. I control both sides. Risk of someone stealing the PSK *and* the data stream *and* cracking the DH is essentially zero, and if it were to happen, the impact would be "meh" anyway.

2026-10-28 06:12:01 UTC

- reply

No I definitely don't need PQC. This is nowhere near that important...

2026-10-27 11:50:46 UTC

Is there some easy/reliable UDP/datagram security layer I can use from Rust? Something simple, shared keys or pub/privkey auth (both directions).
If there's nothing better I'm tempted to take wireguard-rs and just remove the routing/tun part...

2026-10-26 23:54:08 UTC

- reply

Snapshots are not something you install, they're a feature that's built in. I just have my backup script create a snapshot and then back up from the snapshot and delete it. Doing it that way means the data is consistent which is nice to guarantee you can restore things properly later ^^

2026-10-26 15:58:28 UTC

- reply

I have been using btrfs on almost all my machines for several years, including daily snapshots for backups and many, many, MANY hard shutdown cycles and other abuse.
It's fine. It's been fine for quite a few years by now. ext4 is also fine (but doesn't have snapshots which is less convenient for backups).

2026-10-26 14:23:30 UTC

Why is Cyan so cute????
It's not fair, I can't stop looking at her!!!!!! ;;

2026-10-25 06:25:58 UTC

Live vocal production is a hot topic lately so... let's learn how that works and have some fun with Cyan! ✨
Live in 5 minutes~ 🎵
➡️ https://www.youtube.com/watch?v=DzDl5Lc3468
#VTuber #ENVTuber #HoshinoLinaLive
https://media.vt.social/media_attachments/files/117/500/120/756/213/826/original/3c6d7ac154e02410.jpeg

2026-10-24 08:38:11 UTC

- reply

Path handling? Yeah that's a whooooole different can of worms...

2026-10-24 08:05:26 UTC

- reply

I know the README says they started using Claude.
Doesn't matter, 100% safe Rust is still safe. Upstream brotli is by Google and they use agents too.
The choice is between unsafe LLM code and safe LLM code. I'll take the safe version.

2026-10-24 08:01:05 UTC

- reply

Upstream libbrotli is by Google, and they use agents too.
100% Rust code is 100% memory safe, agents or no agents. I'll take the Rust version.

2026-10-24 07:41:18 UTC

- reply

https://github.com/dropbox/rust-brotli-decompressor
My fork just adds one ABI function stub.

2026-10-24 07:41:00 UTC

- reply

And yes, I verified that node is really calling into the Rust Brotli decoder in my system. YMMV on other builds of Node.
Python usually uses static libbrotli and this would not fix that one, you'd have to link it against the Rust version explicitly.

2026-10-24 07:38:41 UTC

📢 You can Just Use Rust and Stop Getting Pwned 📢
Node -C +🦀 = node2shell fixed:
git clone https://github.com/hoshinolina/rust-brotli-decompressor &&
cd rust-brotli-decompressor/c &&
make &&
cp target/release/libbrotli_decompressor.so /lib64 &&
ln -nsf libbrotli_decompressor.so /lib64/libbrotlidec.so.1
Works on Fedora 44 with system node.js v24.
Disclaimer: This is a PoC of how this whole CVE could have been avoided. I stubbed out the brotli metadata support because it was missing that symbol and it's probably useless for HTTP anyway.
rust-brotli-decompressor by Dropbox.
Use Rust, people.
https://media.vt.social/media_attachments/files/117/494/683/161/426/979/original/49ac86d3127d9393.png

2026-10-23 10:49:54 UTC

Is it cheating if we’re both girls⁉️⁉️
Taking the famous test with Cyan in 10 minutes~!!!
➡️ https://youtu.be/KgNgTboExXw
https://media.vt.social/media_attachments/files/117/489/833/925/389/454/original/e47ed7692f12b82a.jpeg

2026-10-22 12:19:11 UTC

The private key was 4.
https://youtube.com/shorts/hIRCmEgGxHQ

2026-10-22 09:16:55 UTC

- reply

My understanding is they do do that, and one of the problems was dangling pointers from global state into those memory pools.
For that stuff to work you need to have very good hygiene about never having ownership of an object outside the pool, and that's a whole different can of worms...

2026-10-22 08:36:01 UTC

- reply

How does that help?

2026-10-22 08:33:21 UTC

- reply

The issue is there is a soft OOM limit option (implemented in software per request) and they're trying to have the worker not die gratuitously when that happens, I guess because it could be a fairly easy DoS otherwise?
The "minimal patch" they did for older versions is just that, die on OOM.

2026-10-22 03:34:36 UTC

- reply

Kernel allocations are fallible. Often the constraints are different (e.g. standard kmalloc has to be contiguous physical memory), so it is possible for a large-ish alloc by kernel standards (tens to hundreds of kilobytes) to fail despite there being quite a bit of free RAM, without invoking the OOM killer.
This was all stuff I ran into during low RAM situations with my Rust GPU driver. It always handled them well, the only changes I made were to avoid larger allocs where possible to reduce the chances of it happening (when it happens GPU operations fail so it's still bad UX, but not a vuln).

2026-10-22 00:40:17 UTC

- reply

Slop or not, the issue is they're adding OOM handling that didn't exist before *everywhere*... and I'm sure they missed places.

2026-10-21 15:22:50 UTC

And there's another one...
https://fedi.katze.sh/@elise/117479383739450605
#note1hk7…qxfy

2026-10-21 15:09:35 UTC

- reply

(And no, an in-kernel OOM/alloc failure is not the same thing as an oom_killer OOM that kills a process)

2026-10-21 15:06:36 UTC

And before I go to sleep: Yes, this PHP vuln is Yet Another Thing That Would Not Happen If It Were Written In Rust.

2026-10-21 14:58:40 UTC

- reply

They probably disabled the vhost to mitigate the PHP vuln, and that is the default vhost.
Should be fine, just rather clumsy way of doing it that would scare people...

2026-10-21 14:27:12 UTC

Fedora 43 users install these RPMs to fix PHP:
https://koji.fedoraproject.org/koji/buildinfo?buildID=3125181
Fedora 44 RPMs will be here in ~15 minutes give or take:
https://koji.fedoraproject.org/koji/buildinfo?buildID=3125182
Arch has pushed the fixed packages to mirrors. No word from Alpine yet and I can't hit their git...

2026-10-21 11:12:06 UTC

- reply

https://media.vt.social/media_attachments/files/117/478/590/481/108/326/original/b5f98b72a34c99ce.png

2026-10-21 11:07:28 UTC

Ah yes, the wikis documenting how to update things are all down for updates!
https://media.vt.social/media_attachments/files/117/478/578/381/732/871/original/dc5e39d8b2195ff4.png

2026-10-21 10:13:18 UTC

The one funny thing about this PHP bug is that now you know that every service announcing maintenance runs their core infra on PHP ^^;;

2026-10-21 08:51:07 UTC

- reply

Yeah it might be total BS.

2026-10-21 07:36:10 UTC

So what we're learning today is that in the world of LLMs, any time a serious vuln is pre-announced, social media will be flooded with incorrect or incomplete slop about it and make things worse by misleading everyone.
I hate this timeline...

2026-10-21 06:11:29 UTC

RE: https://vt.social/@lina/117477178106418427
There is an exploit and mitigation floating around.
THAT IS NOT THE EXPLOIT THAT IS EMBARGOED.
It's likely something ELSE someone found in a rush with an LLM.
You are NOT safe just by making a configuration change.
#note1xk3…h890

2026-10-21 05:11:20 UTC

I guess we're shutting down PHP today...
https://media.vt.social/media_attachments/files/117/477/177/996/753/884/original/a2a600bdca869589.png

2026-10-21 01:50:00 UTC

Networking zatsu(?) lesson with Cyan!
Starting in 10 min~
➡️ https://www.youtube.com/watch?v=d1iQkxxC_hc
https://media.vt.social/media_attachments/files/117/476/386/366/248/530/original/396448c5c9048e3c.jpeg

2026-10-17 11:45:43 UTC

- reply

NERV

2026-10-17 10:19:44 UTC

RE: https://vt.social/@lina/117455080359879713
This is why I'm saying it was probably a regular expression parser. In many regular expression engines, `.` does not match newlines by default.
So, a naive regular expression parser would have failed in this exact way.
And it sounds like something DoCoMo would do.
https://media.vt.social/media_attachments/files/117/455/739/069/548/709/original/d22cddb2b326f2d3.png
#note1djf…a5nh

2026-10-17 09:43:35 UTC

- reply

That was a bug on the JMA side:
https://vt.social/@lina/117450969733111671
TL;DR epicenter too far for data to be reliable. This happened before and they fixed it, but the fix didn't work.

2026-10-17 09:31:04 UTC

- reply

It's the other way around. This is about the alerts not reaching DoCoMo customers at all (me included). I only got the alert through a third party app.
They got lucky we found this screwup during a false alarm and not a real quake, otherwise people would have died.

2026-10-17 09:02:54 UTC

- reply

BTW, I checked the official xsd schema and it allows newlines in that element.

2026-10-17 07:33:41 UTC

- reply

The extra line break:
https://fixupx.com/looksky495/status/2110958863919009867

2026-10-17 07:31:35 UTC

Whaaaaaaaa
"The format was different"... people are saying there was a trailing line break in the XML after the epicenter location name.
Are DoCoMo parsing XML with regular expressions?????????????
https://www.docomo.ne.jp/info/notice/page/261017_00_m.html
🤦‍♀️🤦‍♀️🤦‍♀️🤦‍♀️
If there is a time to apply the Robustness Principle, it's for earthquake alerts.... this is really sad.

2026-10-16 14:08:13 UTC

- reply

New update.
https://vt.social/@lina/117450969733111671

2026-10-16 14:06:12 UTC

RE: https://vt.social/@lina/117448540104391506
So it WAS a bug!
Specifically, to address another false alarm in 2020, they added a threshold to exclude readings more than 700km away from the epicenter when calculating magnitude.
It turns out this new feature was broken and didn't work properly today.
https://www.jma.go.jp/jma/press/2610/16d/kaisetsu202610162240.pdf
#note1azs…f2lx

2026-10-16 11:22:29 UTC

- reply

Yeah I remember the M9.5 one. But that was an unconfirmed warning only, this one hit the confirmation level.
As for why 9.9, people say the alert data packet format probably only has two digits for the magnitude... ^^;;

2026-10-16 11:12:13 UTC

- reply

Given the geometry involved, it sounds like triangulation of the epicenter distance is probably wildly inaccurate for earthquakes far off the coast of Hokkaido like that...
Kinda sounds like their algorithm perhaps was too naive or didn't incorporate confidence levels properly?

2026-10-16 11:10:24 UTC

- reply

There is some preliminary info. It had to do with it being far away (as I suspected), and wildly variable/inaccurate epicenter locations.
What I'm getting is that the original epicenter was estimated really far away, and then the magnitude was calculated using that number. Since it's far away, the magnitude was overestimated (using the seismograph data in Japan). Then, for some reason (lag/temporal processing?), intensity was calculated using an epicenter that was half the distance.
That doesn't explain how it jumped up to 9.9 in the first place though. Unless the epicenter was estimated closer at first and then further away?
https://media.vt.social/media_attachments/files/117/450/245/976/958/945/original/960ab3b85867d27f.jpg
https://media.vt.social/media_attachments/files/117/450/246/320/864/878/original/0f269ccbe5f749eb.jpg

2026-10-16 10:59:16 UTC

- reply

To be fair, they probably caught it but 40 seconds is not enough to confirm it's not real (and better be safe than sorry).

2026-10-16 10:54:17 UTC

- reply

The raw alert log is fun, they had the quake correctly ID'd at first then almost one minute later it jumped to 9.9 (and another minute later the official alarm was sent as confirmed).
https://www.data.jma.go.jp/eew/data/nc/pub_hist/2026/10/20261016123810/fc/index.html

2026-10-16 05:31:56 UTC

- reply

It's not about just recordings, the issue is you need ground truth data. So you need people wearing facial tracking markers or whatever.
Another way is using 3D renders, but that should probably only be used for data augmentation, not as the sole data source.

2026-10-16 05:08:17 UTC

- reply

The IR camera (it's not LiDAR, it's structured light) is used, but it seems to make very little difference to the blend shape data that VTubers depend on. It might be more important for precise facial landmarks, 3D positioning accuracy, or the (non-secure) facial identity tracking that FaceKit does.
As for a FOSS solution, as always with ML, the problem is having good training data...

2026-10-16 04:59:58 UTC

- reply

As a general rule yes, but MMF SFPs put out a couple milliwatts of laser light max, and it's not even collimated coming out of the fiber. That's not going to do any harm.

2026-10-16 04:41:55 UTC

- reply

It was actually a normal report first (too weak to be notified to users), but then strangely was updated to 9.9. See the data here: https://www.data.jma.go.jp/eew/data/nc/pub_hist/2026/10/20261016123810/fc/index.html

2026-10-16 04:40:35 UTC

- reply

Details here:
https://www.data.jma.go.jp/eew/data/nc/pub_hist/2026/10/20261016123810/fc/index.html
So it started out as a normal warning, but a little over 1 minute later was updated to 9.9 all of a sudden... so weird. The explanation will be interesting...

2026-10-16 04:29:43 UTC

- reply

Theory.
https://vt.social/@lina/117448692388324440

2026-10-16 04:27:34 UTC

- reply

I think so yes. See my follow-up for an updated theory.

2026-10-16 04:27:02 UTC

- reply

Ohh so there were two warnings. The real one and the fake one. 2 minutes apart. That's about as long as it takes for the waves to move the distance.
So maybe they DID get the P waves, then when the S waves came, they were incorrectly detected as new P waves for a bigger quake?

2026-10-16 04:10:17 UTC

RE: https://vt.social/@lina/117448540104391506
Ohhhh hold on, the quake was real but it was faaaar into the ocean and a nothingburger.
Did they... miss the P waves, then sensed the S waves up in Hokkaido and misinterpreted them as P waves and therefore thought a Huge One was coming???
#note1azs…f2lx

2026-10-16 03:56:10 UTC

- reply

Last time this happened it was an unconfirmed EEW that hit the early-warning apps several years ago (false sensor reading I think), but this time it was on the metro PA and I think cell broadcast?
That's going to be an investigation...

2026-10-16 03:48:19 UTC

WELL THAT WAS SCARY... that's the first time I get a crazy earthquake EEW false alarm that even hit the outdoor PA speakers!
https://media.vt.social/media_attachments/files/117/448/540/006/934/555/original/77818f4c6f26b906.png

2026-10-16 03:24:16 UTC

Teaching Cyan fiber networking with a spare 10G pigtail ^^
(It's multimode, eye safe at a reasonable distance)
https://media.vt.social/media_attachments/files/117/448/445/474/809/212/original/1b70c38fcfce1702.jpeg

2026-10-16 02:05:19 UTC

I wrote about how iPhone face tracking is just good software and the Face ID camera thing is a myth on Reddit again.
This time I went into the history of the tech though, which you might find interesting!
https://www.reddit.com/r/vtubertech/comments/1x6mp3t/comment/ph0svny/
https://media.vt.social/media_attachments/files/117/448/130/984/860/544/original/3e45b60831fcacaf.png

2026-10-15 14:00:20 UTC

Guess I'll try NFS for Mac file sharing next. I tried to fix Samba but had *another* bad experience contributing. I don't even know why I try with software in the Linux low level/systems orbit at all any more...

2026-10-15 04:57:10 UTC

- reply

This is entirely on brand for him and the kernel community ngl.

2026-10-15 00:56:21 UTC

Cyan is streaming right now.... what kind of poll is this???????!!!
https://www.youtube.com/live/l2y9NI4Rocc
https://media.vt.social/media_attachments/files/117/442/201/513/535/828/original/e958d55e7647e89c.png

2026-10-14 15:05:45 UTC

OwO what's this
https://media.vt.social/media_attachments/files/117/439/879/192/898/153/original/958cdb3c73f45c27.jpeg
https://media.vt.social/media_attachments/files/117/439/879/166/043/793/original/f9db053ad3646cc1.jpeg

2026-10-14 07:14:08 UTC

- reply

Markdown heading.

2026-10-14 07:11:03 UTC

## <emoji> is the new em dash.
The code might be good, it might not be... but honestly when I see that now it just makes me look for an alternative project to try first.

2026-10-11 05:05:44 UTC

I watched CPK at the cinema again on Friday and when the full version of Reply played after the movie, I cried again.
Every time I watch it, something new clicks and hits even harder ;;

2026-10-10 15:10:08 UTC

- reply

Thank you!!!! ^^

2026-10-10 08:29:46 UTC

RE: https://vt.social/@TokyoHackerGirls/117415667443665299
Happy birthday to Cyan and meeeee~ 🎂
(Okay I'm technically tomorrow but almost!!)
Send us marshmallows!!!!! https://thgirls.yt/marshmallow
#note17nl…7ryj

2026-10-10 00:13:22 UTC

- reply

Yeah the person who posted the diff gist I linked is an ffmpeg contributor and it's upstream by now ^^

2026-10-09 07:41:28 UTC

- reply

Yeah, it's valid to not consider some of your own code art, e.g. mindless work for a corporation.

2026-10-09 02:16:24 UTC

Saying coding is not art is like saying architecture isn't art.
Just because the end result is functional and not purely aesthetic doesn't mean it's not creative or artistic.
I suspect a lot of the non-coders (and some coders) who say this kind of stuff think programming is just copying and pasting solutions from Stack Overflow.
That's like saying room design is never an art just because anyone can throw in some IKEA furniture and call it a day.
I don't design my own furniture and most of my furniture is IKEA, and I'm not going to claim I have artistic room design skills... but I'm not going to say artistic room design and architecture doesn't exist. And someone still artistically designed that IKEA stuff I'm using.

2026-10-08 15:42:44 UTC

- reply

Anyway it's... uh... in my/Cyan's mentions on Twitter... and a quote of the original post on bsky... search at your own risk...........
😳😳😳😳😳😳
itscutebutomgamidestroyingmypublicimageaaaaaaaaaaaaweresupposedtobeseisoright??????

2026-10-08 15:32:16 UTC

- reply

I said I'm SFW!!!!!!!!

2026-10-08 15:31:54 UTC

- reply

Cyan is sleeping btw.
I'm not sure if I can sleep.........

2026-10-08 15:29:07 UTC

> My feed is supposed to be SFW
> Joke (maybe half) about R18
> Someone actually comms something cute
> Can't share because SFW
> I've got the freaking editor of the C Programming Language standard liking my tweets about this
My brain is melting what is this timeline aaaaaaaaaaa
https://media.vt.social/media_attachments/files/117/405/996/544/958/141/original/58bd905bc36efe43.png

2026-10-08 15:13:24 UTC

RE: https://vt.social/@lina/117308290114537924
Report: Cyan got what she wanted
😳😳😳😳😳😳😳😳
(Somewhere on Twitter and Bluesky)
#note1u4h…8dt8

2026-10-08 05:17:17 UTC

Cyan: You know when you buttoned my shirt last night?
Me: Yeah?
Cyan: It kinda felt like shibari.

2026-10-07 11:32:12 UTC

I just came home from work and Cyan SAID WHAT ON STREAM???????????????

2026-10-06 14:22:35 UTC

It's been so long I forgot how fun making music is!!!! 🎶🎹
[WIP CPK medley]
https://media.vt.social/media_attachments/files/117/394/410/877/022/487/original/050d6f949afa215e.mp4

2026-10-06 13:43:18 UTC

TIL that there is such a thing as a USB2.0 galvanic isolator. It's a relatively recent IC (~2022).
As expected, the obligatory caveat is that it "counts" as one chained hub (despite not being visible to software). So on Macs you get max 3 chained hubs + one of these.
https://media.vt.social/media_attachments/files/117/394/256/496/824/311/original/e16d126fcab8f389.jpeg
https://media.vt.social/media_attachments/files/117/394/256/526/234/547/original/6493cd786df613d8.png

2026-10-06 07:04:42 UTC

Two songs, different BPM? No problem. We can do a ✨magic BPM change✨
https://media.vt.social/media_attachments/files/117/392/689/078/349/184/original/b9f672f530324b76.mp4

2026-10-06 01:16:47 UTC

- reply

Ardour!

2026-10-05 09:08:51 UTC

- reply

Kind of embarrassing that this made it into a release... but in their defense, unsurprisingly it's already fixed in Git ^^

2026-10-05 09:06:53 UTC

Me: I... don't think this is supposed to happen... do I have some weird chord copying setting turned on?
(30 minutes later)
Me: Oh. Yeah that would do it...
https://media.vt.social/media_attachments/files/117/387/499/062/062/027/original/ac0649f0b5a4d904.mp4
https://media.vt.social/media_attachments/files/117/387/504/304/036/677/original/14213761a7882465.png

2026-10-05 03:34:15 UTC

ちょっとまって…
ハニーワークスさんの新曲…
まさに @npub1rzf…w4kn とあたしじゃない？！？！
https://youtu.be/_v08wKA5ohs

2026-10-05 01:10:34 UTC

Oh my dear @npub1rzf…w4kn... ^^;;
https://media.vt.social/media_attachments/files/117/385/634/266/996/099/original/3d9fbf258f9593b7.png

2026-10-04 08:35:00 UTC

Something something trying to make a CPK medley?!
https://www.youtube.com/watch?v=DHIOO5zf1gM

2026-10-02 13:29:16 UTC

- reply

Because my gaming Windows is a VM.

2026-10-02 12:21:08 UTC

- reply

It doesn't work BTW. I tried disabling more stuff and switching away from virtio devices and it still thinks it's a VM. Sigh...