File: deHR_zi_00040_.png (2.88 MB, 2176x1152)

2.88 MB PNG

>>107738088
node alchemy

File: bbs-zit-2026-01-02_00088_.png (2.26 MB, 1792x1024)

2.26 MB PNG

>>107738068
yeah this makes sense.

ControlNet 0: "Module: dw_openpose_full, Model: diffusion_pytorch_model_promax, Weight: 1, Resize Mode: Crop and Resize, Processor Res: 512, Threshold A: 0.5, Threshold B: 0.5, Guidance Start: 0.0, Guidance End: 1.0, Pixel Perfect: False, Control Mode: My prompt is more important, Hr Option: Both"

so we have a sensible value for Module. also, Control Mode is populated, which i mean who the fuck knows if that matters?

i mean really at some point you may want to consider just biting the bullet and goin comfy. idk if u have used it ever, but it's... fine... now. easy to have multiple workspaces open, kind of easy to swap images around between them (if ur stuck in upscale/inpaint/whatever workflows). idk it's a huge pain in the dick but usually it's because you stepped on your own penis and not because the main codebase is complete ass

>>107738106
this is a sure sign of the end of days... illuminatianon selling comfentanyl to the kiiiiiids fuck them kids

>>107738106
i just installed forge and it works better than A1111 regarding color bleed so im going to stick with it and see if i can make the most of things, while your here do you remember an extension which prevented colors spilling into different tags? (i.e blue shirt, brown eyes, red shoes would make only the specific things they were tagged with that color) i cant remember what it was for the life of me but it worked with SDXL models

>>107726581
I use the voice chat thing as my barometer for howbgood they are because im my mind an ai you can converse with thats indistinguishable from a person that can do answer anything quickly and accurately is the end goal.

So far gpt is the best. Its latest model responds so incredibly fast when you're talking to it you kinda forget that its even looking anything up.
Its tone isn't always the best. I still like grok for its tone. Gemini is really good as well.

>>107726581
Kimi K2 Thinking

File: Screenshot_20251231-182921.png (704 KB, 1080x607)

704 KB PNG

>>107726581
Every ad in social media is always how AI replace the "grunt work" in your generic office work, where people in the fancy office cant write emails or Siri-like voices in call centers.

With the persistent difficulties people have encountered with Text-based AI (Chatgpt,Copilot,Gemini), no wonder why they cant exist in fields where constant supervision when doing anything that actually matters or where errors are a problem (Processing images of Hubble and James Webb, making blueprints for Architects, analyze data for Radiologists)
At my company (Pilots and Aircraft Mechanics) I've noticed the people has zero shit about introducing AI in this and other fields

>>107726741
keep seething, nobody likes a lowlife like you irl

>>107736782
we'll soon have AI like in Her. 5 years maybe
https://www.youtube.com/watch?v=c8zDDPP3REE

File: 1767182414016994.png (246 KB, 646x550)

246 KB PNG

The F7 is 400 for me right now for the 12G ram 512gb version. But with what you guys are saying about the bootloader bullshit I don't know...
While I prefer lineageOS I can probably live with the chinkshit ROM but root is nonnegotiable.
Unrooted android is unusable to me. You can't even screenshot most apps at this point.
Fucking fidelity doesn't let you screenshot, despite their portfolio price history tracking being total dogshit.
The day safetynet spoofing stops working is the day I switch to carrying an old goyphone for banking apps.

File: hands-on-grapheneos-relea(...).jpg (74 KB, 640x813)

74 KB JPG

>>107729530
Ebay and banking apps work. You can transfer your files. If you can't remember your passwords that's on you.

>>107730051
>Posts trans child cartoon
Staring in the mirror Anon?

>>107738161
Graphmeme doesn't pass safetynet by default right?
Does your google wallet work for TTP? that's generally the strictest app you'll encounter
Also how is it on the fold over stock? Any improvements in multitasking?

>>107737745
You don't need to unlock bootloader.
https://youtu.be/8mZzRrZf7Nw?si=dGNNb-EzDPrjis31

File: qwen_image_2512_fp8_e4m3f(...).png (1.99 MB, 1440x1120)

1.99 MB PNG

new qwen, the GPU test.
extremely good result.

>>107738302
wait it can reproduce gpus that accurately? that's cool

File: 1761645672244495.jpg (3.78 MB, 6612x4556)

3.78 MB JPG

>>107738122

File: Superhero sloppa 2.png (1.26 MB, 896x1152)

1.26 MB PNG

>>107738282
I'm only interested in doing pornographic 5 second animations at like 600-720p from pics like this. How long would it take on it?

>>107738321
it's not long nowdays, you go for a lightning lora, 4 steps, with my 3090 I can have a 5 sec 720x720 video in less than 3 minutes

File: bbs-zit-2026-01-02_00122_.png (2.82 MB, 1792x1024)

2.82 MB PNG

>>107738279
these have a high potential to "not be terrible".
B+ :)

>>107737573
>>107737618
I want high pixels per steradian, what's wrong with that? Exact integer multiples shouldn't matter, it's still dynamically rendering a 4K frame, just that the UI elements are scaled up to not be 2mm high. I'd be fine with 300%, though it is kinda goofy, but that causes the same problems. I like not being able to see individual pixels, maybe owning a macbook has poisoned my mind. I wasn't expecting to have to use 250%, but large UI elements are actually nice. You might even want more than 300% for raspberry pi projects.

My question is why my different monitors with different pixel densities have to be scaled to different levels in the first place, as if a pixel has some standard size with respect to UI elements. Instead everything should default to scaling to have pixels the same size with respect to that pixel's size on the monitor itself. There should be a slider entitled "pixels per millimetre", not some arbitrary scale percentage limited between 50-250%. Seems like a no-brainer, idk. Maybe there's a utility out there that does this, especially for people using linux on laptops with extra screens plugged in.

>>107738023
Additionally, some applications don't even respect this scale in the first place, and appear miniscule, as if set to 100% scale. If I knew I'd have this much trouble getting high-DPI monitors working I'd have bought a lower resolution monitor. I didn't even think 27" was that small. To get those 100%-only applications running at my preferred scale, I'd have to use a 1080p 32" monitor, which seems unreasonably grainy. But maybe there's a hacky workaround for scaling those uncooperative applications, or using a different window manager might do something. Suggestions welcome.

For reference, the Submit button here on 4chins is 9mm tall when set to the 250% scale. I sit between 40" and 28" from my screen.

>>107738023
>Exact integer multiples shouldn't matter
is / aught
>as if a pixel has some standard size with respect to UI elements
When you're talking about PC content it's supposed to be 1/96 ~ 1/110 linear inch square. Nobody since the early 90s steps outside this.

rocminfo shows nothing in chroot but works in the parent environment, why's that?

# rocminfo 
ROCk module is NOT loaded, possibly no GPU devices

Tried systemd-nspawn and conventional chroot but nothing.

>>107738225
did you bind the host /dev?

>>107738209
better to cut losses early then

>>107735804
It will, that bubble will pop and it will crash down the economy big time.
The stupid idiots in power had their chance but they fucked it big time by assassinating Nuno Loureiro.

File: Yates.jpg (73 KB, 640x795)

73 KB JPG

>>107735774
Never have used AI, never will use AI.

File: 1749116201033830.jpg (27 KB, 204x320)

27 KB JPG

>>107735774
>mfw luddite troons like you ACK

>>107738245
The bottleneck is energy. Without nuclear fusion AI is doomed.

>>107732565
I doubt Taiwan workforce is much cheaper than European one

>>107738242
China helped the Norks with theirs 100%. And it wasn't actually the nukes that caused this submission, it was the federal reserve bank.
Israel knew they could steal that material without any repercussions because America was already subdued, why do you think America sacrificed so many warriors fighting Germany for no personal gain whatsoever.

>>107738264
>no personal gain
Most European "countries" are American vassals, have been since

>>107738281
And that has helped America how? The country is self destructing much faster than Europe is. Jews gained control, Americans lost lives and wealth, now possible their future.

>>107737225
the will do that. taiwanese hate the chinese. the Chinese love ta>>107738242
>from stolen everything
>that they have openly brought to bargaining
their bargaining is complaining about gas chambers they made up during the eichmann trial that were mentioned once in the entire documentation of nuremburg then not until the 60s when mossad grabbed eichmann illegally out of south america and executed him.
from 1945 to 1963 nobody on planet earth thought concentration camps had gas chambers because they didn't. boomers got scammed by yids acting like they've been injured over something that didn't even happen.

Just got these from a recommendation https://www.aliexpress.com/item/1005008657385412.html
hoping they're decent for the price

File: 1753450615313692.jpg (820 KB, 3072x4096)

820 KB JPG

IEM stand arrive

>>107737918
only desktops still suck, laptops and mobile devices are fine these days (excluding ghetto poorfag trash)

ia500s coming in the mail soon.......

File: IMG_20260102_112234.jpg (3.17 MB, 4000x5333)

3.17 MB JPG

>>107737834
USB + magnets.

>>107738280
pdfs are not really meant to be editable
use latex or tikz or groff or postscript or whatever like a normal person

>>107717118
They have it already

>>107719148
Terrible

>>107732210
Don't listen to those two idiots who replied to you, holy shit. I've a Samsung TV and installed the Jellyfin app using Docker.
https://github.com/Georift/install-jellyfin-tizen
Nit really hard, just read the thing through, run that PowerShell command, and you're golden

>>107710235
nfo is still buggy.
https://github.com/jellyfin/jellyfin/issues/13197
https://github.com/jellyfin/jellyfin/issues/13655
>who cares
nfo working properly would be extremely useful for using Jellyfin with non tv/movie media including but not limited to youtube
https://github.com/tubearchivist/tubearchivist-jf-plugin/issues/30

>>107737180
>nfo working properly would be extremely useful for using Jellyfin with non tv/movie media including but not limited to youtube
don't care

File: bearlain.jpg (103 KB, 468x488)

103 KB JPG

>>107738079
Concerning OpenBSD's port tree in particular;

Almost everything in OpenBSD's ports takes full advantage of features exclusive to the OpenBSD kernel. The main two being pledge and unveil. Which ensure no running process can go looking around the file system in places it shouldn't have access to or access other running processes (or parts of the kernel its not supposed to be dealing with).

OpenBSD is the only OS shipping a port of Firefox and Chromium that are secure by default. All other OSs force you to shove these inside of a chroot, jail, or some other form of container in an attempt to prevent a rouge website from exploiting the local machine. You can never be sure because it's easy to break out of the chroot/jail/container on most every POSIX OS and obtain root anyway.

On OpenBSD anything that attempts to do that causes the application to crash and dump core. Firefox on OpenBSD can't access anything but ~/Downloads by default. You have to edit a config file to allow it to see any more of the local file system than that.

Pledge/Unveil was added to everything in the base system within weeks of it becoming a feature of the kernel. Within a few months most every thing in the ports tree gained support as well. I can't remember the last time I ran across anything that wasn't fully supported.

That's the main difference. Ports are more closely integrated with the rest of the OS in the BSDs. In Linux a package is typically something some guy tested once on his local machine with all compile time options enabled. That's the case in a distro like Arch. Where people claim it's "minimal" but in reality every last package included is bloated up to the max with a bunch of crap most systems do not need. More code = more bugs.

IF you really think you need the cancer from Linux then one should simply run FreeBSD. Seeing as most all cancer from modern Linux is fully supported right down to drivers and everything. But this means the OS is less coherently designed.

I like OpenBSD a lot, but audio over HDMI doesn't work on my laptop (or I at least don't know how to set it up), and I kinda need an alternative to Anki.
Only these 2 things are stopping me from fully switching :/

>>107738151
>Audio over HDMI
You can enable this if you're willing to build your own kernel from source. It probably works on most hardware. The relevant file is /usr/src/sys/dev/pci/azalia.c.

It's currently disabled to prevent the hdmi port from claiming /dev/audio instead of built-in speakers or the aux port.

>Anki
Last I checked this had a port. Although I avoid it because it's pretty shitty software. I know there is a good alternative emacs package for flash cards.

If you're learning Japanese I would suggest making your own flash cards by hand. Since you'd get to practice writing kanji as well. That's how we did it back in my day. Grinding Anki seems to be one of those things where a lot of people claim they're doing it and it works. But I never see them getting beyond reading manga for children.

At any rate Anki is pretty terrible software so I tend to avoid it as much as possible. I know the port was working when I checked last year. But considering the author goes out of his way to use just about everything that's a pain in the ass to build from source (Rust, Python 3.x, Qt and on and on it goes) I highly suggest you avoid it whenever possible.

Anki is one of those things where even if you can build it on AMD64 it'll never support the vast majority of the platforms OpenBSD supports due to the nature of the project. The people running it are certainly not interested in improving this situation. Then there is the fact that most of the add-ons for it want to integrate with google auth and a bunch of other cancer.


Comment too long. Click here to view the full text.

>>107738151
>>107738252
Re: Anki

I just checked the mailing list and I was correct ports still has an older version from before Rust became a dependency. It's currently working and patched again in -current. It was noticed just after the window closed to make it into release 7.8 so if you're on that it's probably broken.

relevant post from mailing list: https://marc.info/?l=openbsd-ports&m=173210715726899&w=2

You're always better off checking the mailing list (openbsd-ports) instead of seeking support outside of the mailing lists. Anything you need to know is almost always there.

Also, lol I forgot this crappy software wants node.js too. It's like someone went out of their way to make it as horrible as possible. It seems like there are currently no plans to attempt porting over current versions. The guy maintaining it plans to keep patching the older version until it stops working all together.

I would suggest trying kanji-mode for emacs instead: https://github.com/wsgac/kanji-mode

It worked well enough last I tried it.

https://isopenbsdsecu.re/

https://media.defcon.org/DEF%20CON%2025/DEF%20CON%2025%20presentations/DEF%20CON%2025%20-%20Ilja-van-Sprundel-BSD-Kern-Vulns.pdf

>>107738195
It's about speed, not smarts.

>>107738195
speculative decoding should give the same results as the big model. Its usually pure win when you can use it.
The problem is no one is making good models compatible with the practice, so the tooling is also immature garbage.

>>107738208
Okay I get it, but I mean, lets say I have 16gb of vram. Wouldn't I'd rather want to squeeze in a q4/q5 of a 32/24b model rather than drop from a q5 to a q4 just to fit in a q4+3b or whatever speculative model ontop?
Or is the idea that you make the drop from a higher bit quant "worth" it by using the smaller model to juice the speed up to compensate for the quality drop?

>>107738239
Speculative models are supposed to be small enough to make it worth the speed gain. A 1b model for speculative while the main one is 32b or higher is fine. It's just 1gb (+ context) for a potential 3 or 4x in token gen speed (and only for easily predictable tasks like coding). If you're coping with a q1xxxxxxxs_small_compacted_tiny_trimmed and struggling with the 74kb left in your gpu, speculative is not for you. As described, it's for the ones that have an extra h200 laying around for an 8b.
But even for the mortals, if there was a good implementation and compatible models, it could be worth it if you have some spare vram and a good 500m or whatever speculator model.
It'd be up to you to decide whether it's worth it or not.

>>107731243
muh trannyformers

https://www.youtube.com/shorts/4g251atrdX8

>>107737696
>wired connection
Of course, both my main pc and my NAS that I use as a pseudo seedbox are on wired. Just hope they won't be bitches with all my torrenting

>>107737687
I am aware. I'm not trying to run it. I just want to grab the text strings from the page's HTML so i can modify a bunch of them en-masse using regex and substitutions into a reference guide.
And importXML can definitely grab text strings from HMTL if you give it the xpath. "Imports data from any of various structured data types including XML, HTML, CSV, TSV, and RSS and ATOM XML feeds."

I could literally accomplish what I want to do with a bunch of copy pasting, but I'd be at it for hours.

I'm datascraping a minecraft mod.

>>107737858
not that anon but
>I'm trying to import the raw source code
>raw source code
JSON is structured data
>"Imports data from any of various structured data types including XML, HTML, CSV, TSV, and RSS and ATOM XML feeds."
>various structured data types
Sounds like this tool you're using doesn't parse txt, or JSON hence
>"content can not be parsed"

File: importxml.png (32 KB, 1397x452)

32 KB PNG

>>107737896
okay how do I get it to stop parsing it and just give it to me as it is?

Anywhere else i tell it to grab text from it can do it just fine.

File: importxml2.png (130 KB, 1920x857)

130 KB PNG

>>107738192
But right here, is for some reason a no go.