🍥｜anime

[wispy canopy]

lol, revenge of the melonmancer xD

[ocean reef]

watercolor style looks good on your model

dont ask me what it is tho

[wispy canopy]

just prompting watercolor style?

[ocean reef]

a watercolor painting

[wet elm]

im seeing a lot of helltaker-style chars in tiktok these days, but idk how to recreate those

like it looks so good

[wispy canopy]

it should be versatile though, i've specifically remixed the v1.1 so that it should have pretty decent knowledge to do a lot of things

[fallen sundial]

@wispy canopy how to install a lora and use it?

[wispy canopy]

copy/move it to your lora folder (models/lora) and then refresh it in the bookmarks (the red button below generate)

lol, starbucks tarot 😄

[ocean reef]

😏

time for coffee brake 😄

[wispy canopy]

no seriously, why is it THAT good at the starbucks logo 😮

[ocean reef]

indeed

[trim current]

alright pow time

[ocean reef]

so goood!

[wispy canopy]

all that pizza went to her breasts?

[trim current]

thank you 🙂

[steady grail]

ehy guys, there is some new fantasy model out there?

[wispy canopy]

every day a couple

[fallen sundial]

@wispy canopy it didn't work

[steady grail]

something good with bright colors

[fallen sundial]

I put it and it doesn't show lora

[trim current]

Colorbox

Idk

Theres allot

[steady grail]

ok, let's start with a new model hunting

[wispy canopy]

you have to refresh

[fallen sundial]

I did

[ocean reef]

but it has wrong format

delete that thing after safesensor

[supple raptor]

This looks like a failed download. Delete and download again.

[fallen sundial]

no it's not failked

failed

[steady grail]

ok 50 more g to spend in models

[fallen sundial]

like this?

[steady grail]

download at the end means the download got interrupted

you should delete and redownload the file

for who is interested, there are new god slime girl loras

there is a database where i can see the recipe of the models?

[trim current]

a quick test

[supple raptor]

No, the best place to look would be the CivitAI and/or hugging face posting for the model. Sometimes the author documents their recipe. Unfortunately a lot or people have started keeping their recipes secret, especially since those sites that were paying for exclusive model rights started popping up.

[steady grail]

that makes hart to select models

i'm looking for a good anime model with not booru based prompts

[supple raptor]

If you mean without booru tagging, then there aren't really any IMO. Booru tagging, and especially the WaifuDiffusion tagging tool just works way better for training the CLIP text encoder so even plenty of the non-anime models use booru tagging.
If you mean without using booru image datasets, then it's basically impossible to know because very few people share their datasets.

[steady grail]

i'm just very bored of using syntetic booru tokens

[fallen sundial]

@supple raptor hello could you help me with an issue ?

[steady grail]

every new model is only of 2 gb. 😦

[supple raptor]

You don't have to just use booru tags for models that were trained with them. Often it works even better to work the booru tags into natural language so your prompt ends up being a combination of natural language and booru tags.

[supple raptor]

That's just because they've been converted to fp16 and had EMA pruned out of them.

[supple raptor]

What's the issue?

[steady grail]

many of the models i use with good fantasy graphic, gives boring results, with a story telling prompt

oh

[supple raptor]

CLIP is just not that good at interpreting language. It's pretty surprising that we can get it to work as well as it does for us.

[gritty patrol]

guys, i was searching for models and found this example image
https://civitai.com/images/749005?period=AllTime&periodMode=published&sort=Newest&view=categories&modelVersionId=65092&modelId=51057&postId=203444

[steady grail]

i'm looking for something holy mana, i didn't expect this result at all.
if i use a very synthetic prompt, it's just like "meh, just like i predicted"

[gritty patrol]

check the negative

[trim current]

I plan on manually tagging my next dataset

But i wont worry about that for awhile

[fallen sundial]

stable diffusion refues to make good image

what the ??

[steady grail]

the problem is the hand, right?

[supple raptor]

If you drag and drop the image file from your outputs folder I'll take a look at your parameters and prompt to see if I have any suggestions.

[fallen sundial]

yes and face

[supple raptor]

The problem with manually tagging is consistency. It's a lot more work, and often gives worse results. It's usually best to let WD tagger make the tags for you, then manually edit those.

[fallen sundial]

it's mistoon anime

I don't think it's high quality image..

[gritty patrol]

i wanna have my take on it, can i have the original image?

[trim current]

oh fr hmm okayu

[supple raptor]

Is this your input image that you're using for img2img?

[fallen sundial]

yes

[supple raptor]

The first 2 things I would recommend experimenting with are your negative prompt and the model that you're using. It tried to put a realistic nose and lips on the guy which indicates to me that the model might not be very good at anime in the first place. To be honest, it would be best to start with doing some regular txt2img generations so that you can learn how the model responds to prompts before you try to add the extra complication of img2img on top of that.

[fallen sundial]

but its name was ( mistoon anime )

now it gave me good result

but with img2img not

[gritty patrol]

best i can do

[fallen sundial]

look how good it's, not as img2img

[fallen sundial]

Okay but I think the character get destroyed

[gritty patrol]

lower denoising is just bad

the problem is that he doesn't initially have a face

[fallen sundial]

to be honest I will remove the face

but my problem with hands and the drawing itself

[supple raptor]

It may be really overfitted on the dataset it was trained on. That would allow it to spit out pretty good images that don't really follow your prompts in txt2img but full flat on it's face when you try to force it to generate something different with img2img.

[steady grail]

god, i forgot to add nude!!! in negative and the model gave me a naked woman, using that image lol

[fallen sundial]

._.

[steady grail]

and the prompt young boy in fighting pose

i which part i asked for a naked woman?!!!!!!

[supple raptor]

lmao, which model are you using?

[steady grail]

midnight maple

[supple raptor]

Hmm. It wouldn't have surprised me if it was something AOM based, but I don't think that DrBob used AOM in that one.

[gritty patrol]

guys help, i keep "downgrading"

i went from using kl f8 anime2 blessed 0.9 vae to my old ft mse

and now i'm doing better using kuki than aurora

[supple raptor]

nothing wrong with mixing things up.

[gritty patrol]

but i feel like it's wrong and i'm missing out on improvements

[steady grail]

and now, why he is trying to use the spirit cannon?

XD

forgot to set the riight resolution

that's looks very good

[near elbow]

is this anime or portrait? 🙂

[gritty patrol]

portrait

[near elbow]

ok, but this is clearly anime 🙂

[gritty patrol]

people that know better than me, what model do you guys think this is?

i'm thinking counterfeit but idk

[steady grail]

it may even be altered by a lora

like this? @fallen sundial this is when you use controlnet

still need some polishing, but the results looks already better, in my opinion

[gritty patrol]

the negative

noclip

[gloomy scroll]

So you are saying that I should go look for my image's novel I mean negative

[gritty patrol]

i was saying check the negative of the image i posted

[gloomy scroll]

This is my image's negative

[gritty patrol]

WTF

[gloomy scroll]

I believe that's some old chinese negative

[gritty patrol]

why do people do this

[gloomy scroll]

It worked

[steady grail]

all that stranger numbers are chinese text?

[nova remnant]

What extensions do you recommend as essential?

[steady grail]

essential?

[gloomy scroll]

https://github.com/KohakuBlueleaf/a1111-sd-webui-lycoris

Not sure if webui is updated to support lycoris by default

[gloomy scroll]

https://github.com/Mikubill/sd-webui-controlnet.git

[gritty patrol]

welp rip civit again

back already lol

weird

[ocean reef]

like a wreaking ball!

[terse ore]

What kind of anime models show different facial features of different ethnicities, rather than merely changing skin tone?

[supple raptor]

A Lunar Dream is the only one I know of that's had any effort put into training multiple ethnicities into it. Anime in general isn't exactly known for having good racial diversity representation outside of fantasy races.

[stable crane]

❤️

[small pecan]

ribit

[subtle crown]

love the frog

[small pecan]

thanks i do too

[subtle crown]

been like this all day, on and off

[supple raptor]

Civitai developers when they built their site on postgres without knowing how to use postgres:

[stable crane]

thx clyde

[steady grail]

i love control net and tilrd diffusion so much. i upsized to 2k and pushed denois to 9

the results looks very promising

[brittle jay]

Can someone pls train Lora on this https://www.pixiv.net/en/users/20973893 ?

I just want this because their name in Hebrew means fagi_pooppoop

[steady grail]

what? o.O'

[brittle jay]

Btw how tf can I remove embeds of messages from phone?

[steady grail]

that's a good thing? lol

eh?

[supple raptor]

Edit the message and add triangle brackets around the link like <your.link>

[stable crane]

I need to give tiled diffusion another shot

[steady grail]

it's possible to control sb from phone?

i'll show the result of my experiment in a hour or 2

[brittle jay]

Nice thanks

[strong thunder]

3

[steady grail]

resolution?

[supple raptor]

Yep, if your phone will be on the same network, all you have to do is add --listen to your commandline args then you can access it from your phone with your PC's local IP address followed by the port number it should be something like http://192.168.1.121:7860/. I'd recommend setting a static IP address for your PC so that you don't have to check the current local IP for it each time.

[steady grail]

that's damn cool

[steady grail]

i don't like the result of my upscaling

[stable crane]

I find it really hit or miss

I end up just doing hires fix

usually works out best with the pure txt2img

[lament osprey]

hey guys, for negative embeddings, i literally just put the file name in negative prompt?

[compact warren]

Bro
U gotta stop making her so cute wtf

[stable crane]

yes

[lament osprey]

thanks

[compact warren]

https://tenor.com/view/glitchart-failure-meltdown-404-fuck-gif-5011516

[compact warren]

Jellyfish waifu

Looks like she could be from kantai collection

[subtle totem]

i'm struggling using control map for some reason it's not working as expected

can someone make something great with my sketch

[clever shore]

is she an oc?

[gloomy scroll]

Yeah

[clever shore]

kawaii

[gloomy scroll]

Sleepy time

[terse ore]

How can you edit and still misspell Hawaii

[compact warren]

im incredibly jealous

I wanna take her so bad 😭

[gloomy scroll]

This is her non chibi form

[subtle totem]

omg very cool man

how did u achieve that

[keen hollow]

evenin'

[subtle totem]

and also how did u color it very perfect like this

[icy lake]

is it outdated to use underscores like danbooru tags in new models like aurora lol

[stable crane]

I use underscores still

[keen hollow]

quiet tonight

[lucid ferry]

sd seems to have died down in hype, might pick up if something comes along

[compact warren]

Oh dam

😳

[wispy canopy]

dang, such clean lines ❤️

[cunning gyro]

Family in my story. Fyriel, her husband Caelum, and their children Aria and Asher.

[cunning gyro]

Also, Rena, a slime girl OC

[torpid plover]

Hey could you see what did you put on info? Cause me i put alot of characteristic and it still changing and don't want to put what i wanted

And i see you it's not changing

[fallen sundial]

really good but I want anime style lineart

[solid ruin]

can i get my picture into anime?

[wispy canopy]

i don't know. can you? if you have generated pictures, and they're not nsfw, you can post them here

[solid ruin]

if i knew i wouldnt ask

[wispy canopy]

did you check the #1072220168534642768 and #1072229020520947753 ?

besically, most, if not all of use, use a program called A1111 to generate these images

[solid ruin]

ok thank you

[wispy canopy]

@supple raptor

[supple raptor]

I'm pretty hot I guess

[wispy canopy]

Never knew you were a fertility jungle waifu 😮

[supple raptor]

I didn't either. Maybe it's time I rethink my beard

[wispy canopy]

lol XD

[solid ruin]

is it paid?

[wispy canopy]

it's free

https://github.com/AUTOMATIC1111/stable-diffusion-webui

[solid ruin]

tell me how do i change my picture into anime

[wispy canopy]

you'd have to img2img that

first you'd have to install the softare and get some models loaded

[solid ruin]

tell me which softwares xd

[wispy canopy]

i just posted the link

[gloomy scroll]

I used ControlNet with scribble preprocessor

[ocean reef]

@wispy canopy do you even work?! xD

[wispy canopy]

shhhh 😛

[ocean reef]

🤣

[wispy canopy]

everytime i run a command in any of my consoles, i can throw in a new try wherever in SD

[ocean reef]

pros working from home 😄

[wispy canopy]

(consoles = linux terminal consoles, not the things you play games on)

[ocean reef]

glad you explained, I'm not a pro when it comes to IT 😄

[frozen olive]

https://cdn.discordapp.com/attachments/752489365967601664/1106151700131229727/Botanic_Student.png

[pale moss]

6 hours of work (blending many images, editing etc)

[ocean reef]

worth the effort! Looks amazing!

[pale moss]

thank you !

[ocean reef]

You should submit sth for PoW! I can already image how great your work would be 😍

[pale moss]

what is PoW ?

[ocean reef]

#⭐｜pow-info check it out!

[wispy canopy]

dude(ette) 😮 amazing

[brittle jay]

this is amazing

[pale moss]

thank you

[steel arrow]

good evening

[torpid plover]

How did you make stable character it always change me

[torpid plover]

Show what you put to take example :_;

[gloomy scroll]

BREAK 1girl, solo, (open hair, long hair, straight hair, white hair, hime cut, blunt bangs, azure eyes:1.3), pointy ears, dark, plain choker, fangs, (realistic:0.8),

[torpid plover]

Break?

[gloomy scroll]

BREAK starts new 75 token block

[torpid plover]

What is it

I try to put middle part it won't do

[gloomy scroll]

In webui you can see token count in top right corner of prompt input and every time you go over 75 tokens you get new token block added and words at the beginning of these blocks have higher weight so using BREAK lets you easily see where new 75 token block starts

[torpid plover]

Ohhhhhhhh okayyyyy

I'm new this is why

[mental island]

Omg my mind has been blown, hell yeah

[frozen olive]

https://cdn.discordapp.com/attachments/1066951281110941707/1106165584955195452/Sisters_of_Fox_Kingdom.png

[quick hill]

Let me check

There is AbyssOrangeMix.

[steel arrow]

shes back

[cyan burrow]

Not entirely happy but beginning to get the hang of making cards

[steel arrow]

i got my first actual word by accident (it says damn in the back)

[subtle totem]

just that?

[wispy canopy]

rejects for my pow submission

more rejects

[wispy canopy]

https://tenor.com/view/damn-damn-damn-damn-damn-right-dame-time-gif-18029769

one more gen of 8, and then, if no more suitable images come up, it's image fixing time

[distant gale]

why does it always think the eyebrows are hair, it makes the hair and eyebrow connected

[steady grail]

sweet, how did you do it?

it's so damn charming 🥰

[gloomy scroll]

Hoshino Ai Lora and (chibi:1.5)

[steady grail]

so chibi works 😌

[weak moss]

hello, can anyone help me with prompts to describe these clothes?

[trim current]

Upscaling and submitting my pow submission, doubt ill win it was fun

[cyan burrow]

rejecting this, I wanted it to be a little different and less anime, maybe I can photoshop and then img2img it

[steady grail]

maybe leading on a more gray and golden scale of colors?

golden rad and golden blue

[tropic wraith]

FAQ: What is Stability AI?

Our vibrant communities consist of experts, leaders and partners across the globe. They are developing cutting-edge open AI models for Image, Language, Audio, Video, 3D and Biology.. AI by the people, for the people. Learn more here

[fallen sundial]

how can I know if a model has virus or not?

[steady grail]

usually, on civitai, are all clean

safetensors should be safe to use anyway

[fallen sundial]

so I shouldn't be worried ?

[stable crane]

not by safetensors

[fallen sundial]

So any model has this name I shouldn't be worried about it ?

[stable crane]

file extension, ya

[fallen sundial]

you mean yes ?

[steady grail]

yep

[fallen sundial]

okay

thank you

[stable crane]

yup

[fallen sundial]

anything on civitai is clearn ? like VAE or LoRA ?

[steady grail]

usually

[stable crane]

don't think there's ever been a documented case of someone getting infected by a SD model

safetensors are safe

[cyan burrow]

nah I've decided I want them standing back to back and affecting the environment

[steady grail]

ohhhh

[sullen hollow]

rate my inpaint hand pls

[stone onyx]

needs thicker lines to match the overall style

other than that it's perfect

what model is that btw?

[sullen hollow]

ahh youre so right. now i see it too.👍 this is ghostmix model

[stone onyx]

Very cool looking model, imma try it out

[sullen hollow]

yeah its awesome i work with it all the time because its very versatile for me

[cyan burrow]

hmmm if I sell my 3070 ti I can count on getting 500 dollars, a used 3090 will cost around 900 dollars. Not sure if worth it, I don't want to go that hardcore into AI generating and I guess I could just rent servers if I need them. Or sell 3070 ti and buy 4070 new (700 dollars) but then doubling the vram would only be 200 and that's ridiculous difference in value...Plus 4000 series are power monsters

[steady grail]

suggestions on the prompt to get a more glossy anime result?

[terse juniper]

(((masterpiece))), (((best quality))), ((ultra-detailed)), (cinematic lighting)

[trim current]

Aaaaand pow done 🙂

[sullen hollow]

if you dont want to go super hardcore in ai maybe think if you really need the double vram. i mean if you have it thats great but are you gonna use it in the end

[cyan burrow]

I might not use it for AI, but I do enjoy gaming at high settings

but for the 400 difference I could upgrade more important things in the home

like a new couch, some stuff for the kitchen, some film, primer etc to refurbish my desk

[steady grail]

like a big laser beam to vaporize your neighbors that disturbs you lol

[sullen hollow]

i think the real life couch upgrade sounds awesome 😄

[trim current]

Ai gives me an excuse to 4k game

[cyan burrow]

yeah mine has been....too seated

they've become more flat :p

[steady grail]

or a motorized couch with wheals !

[trim current]

I got a new bed and secretlab before getting my gpu. If only i waited a year for the tech to get mainstream

I wouldve opted for nvidia

But how would i of known lol

[cyan burrow]

I dunno what secretlab is

I already have a new bed

memory foam and cold foam with some of that cool gel foam on the top

well sometimes it does

I mean I guess it's a nice image

yeah

[sullen hollow]

ice finger

[steady grail]

we should invent it lol

[sullen hollow]

you know this is the stuff i generate all the time. i dont know what my negative prompts think they are doing but it surely isnt working

[steady grail]

they are fixing weird fingers, using ice fingers, much better

🤣

man, i was trying to do a man in armor and that was the result!

[sullen hollow]

(interlocked fingers:1.2), Ugly Fingers, (extra digit and hands and fingers and legs and arms:1.4), (deformed fingers:1.2), (long fingers:1.2), (bad hands), (missing fingers),extra hands,extra fingers,too many fingers,fused fingers,(bad and mutated hands:1.3),

you cant tell me none of these are working

oh wait i forgot to add (ice fingers)

[steady grail]

i use the TI bad-hands-5 and works pretty well

[cyan burrow]

most of them are probably useless since it doesn't really tag its mistakes

the extra finger etc probably works

[sullen hollow]

hmmm

needs more testing then

and emphasis on that

[primal panther]

Anyone know what are good prompts to get generalyl older looking characters?

[ocean reef]

(old:1.6) 😄

[molten shadow]

((((((((((((((((old)))))))))))))))))

[ocean reef]

xD

[steady grail]

OMG you are so old!!!!!!!!!!!!!!!

guys, where i can't find prompts to change the graphic style of the image?

[cyan burrow]

https://promptomania.com/stable-diffusion-prompt-builder/ https://zele.st/NovelAI/ have fun

@steady grail

even if its called novelai that doesn't matter, it still has valuable info

[stone onyx]

may I have the prompt for this?

I'm trying to get the same outline/style

[steady grail]

wow, it's like a puzzle

i like it

[steady grail]

this site is very cool, i just need to upload the image and add the effects, to test the difference

[steady grail]

someone can interrogate this image for me? the interrogator online is down

[trim current]

So pretty

[brittle jay]

multiple boys, school uniform, baseball bat, serafuku, yellow eyes, blonde hair, red hair, cigarette, black hair, blue hair, blue eyes, 4boys, 1girl, night, jewelry, multiple girls, sky, smoking, short hair

wd 1.4 swinv2 with threshold of 0.35

[cyan burrow]

ah posted my submission, I'm not too happy with it but it's what I could do with the little energy I have atm. Still recouperating

[ocean reef]

health is more important so don't worry ❤️ get well soon!

[brittle jay]

amazing submission

I like

[cyan burrow]

really? It's not flashy or anything but I suppose that's the style and aesthetic of it

bah corner text

there, better

got myself a real nice prompt it seems, well on my model anyway

[compact warren]

Anyone here use dadapt to train

Besides kazoo

[steady grail]

the fire need more smoke and cinder, would be really perfect

[cyan burrow]

depends on what's burning no? 😛 But smoke yes

RuntimeError: shape '[2, 100, 160, 320]' is invalid for input of size 10485760 this is an annoying error that just came outa nowhere no matter what I alter

[torpid plover]

whaaat give text prompt

[trim current]

Love the little bunny in the hair :3

[brittle jay]

😭 why besides me?

[compact warren]

I wanted to know how widespread its use is

[brittle jay]

oh

ask on projectai

[compact warren]

But since ur here
I will ask if u lv trained it on styles or anything

[supple raptor]

I've seen some people on toohoo using it.

[cyan burrow]

ok now it works....seems an extension was auto enabled, that's annoying

[compact warren]

I have, only got one or two peeps to answer

[compact warren]

Where?

[brittle jay]

I trained only on styles

[supple raptor]

ProjectAI

[brittle jay]

though I heard it also works great for characters

[compact warren]

Good, cuz that what i need good settings for

[brittle jay]

toohoo = projectai

[supple raptor]

just a dumb spelling of touhou.

[compact warren]

Ive never heard it called that but nice

[brittle jay]

dumb?

much smart

[supple raptor]

[stone onyx]

It's her hair tie thingy

have you watched oshi no ko?

[cyan burrow]

wtf, far cry 6 already down to 15 euro on steam

[trim current]

I havent no o;
Still allot i need to catch-up/watch
Adhd is a bitch on commiting to watching things

[stone onyx]

:x
you're definitely gonna like the first episode
I get that, I have tons of animes that I haven't finished for no reason lmao

[torpid plover]

BREAK 1man, solo, male, teen (middle part hair, short hair, messy hair, brown hair, middle part cut,center part,green eyes:1.3), depressif, emo , dark, <lora:Helltaker:0.6>,

i wanted a middle part not a side part

[trim current]

I added to my list and hopefully wont forget about it
I really need to watch more anime >_<

[merry tendon]

moin👋

[supple raptor]

https://cdn.discordapp.com/emojis/999416119045656596.gif?size=64

[brittle jay]

BREAK in beginning of prompt?

[torpid plover]

What i understand is for putting more word

[brittle jay]

well...

[cyan burrow]

what does BREAK do?

[brittle jay]

not really

[cyan burrow]

is it a sort of cutoff?

to loop it?

[brittle jay]

you know how every 75 tokens it starts from the beginning again? (76 token = 1 token in weight and stuff)

it just does that

but like you don't have to fill in 75 tokens

bwm here to make me embarrassed from how good his explanation is compared to mine

[merry tendon]

new "anime in our world" lora

[cyan burrow]

makes sense

[supple raptor]

SD only understands prompts that are 75 tokens long. Automatic1111 added a feature that allows it to split longer prompts into blocks that are 75 tokens long, then average those together. BREAK simply fills the current block of 75 tokens with blank ones so that the next token will be the start of the next block of 75. This is useful because the closer the tokens are to the beginning of a block of 75 tokens are considered more important by the AI, and also because sometimes a single word might ocuppy both token 75 and 76.

[brittle jay]

so many of these

[brittle jay]

can you do one where she is further away

[merry tendon]

no, only seen like 2 and 1 is from me(my old one)

[merry tendon]

yep

[gloomy scroll]

My waifu is real

[merry tendon]

still in testing phase tho

[brittle jay]

where you get dataset?

[merry tendon]

my old one + bunch of gathering in random places

[torpid plover]

Oh okayy thanks

[torpid plover]

I still don't know to not do side part but middle part i tried alot to find the good word

[brittle jay]

go to danbooru / safebooru and look up characters that have the hair you want and look at the tags

[torpid plover]

Okay

[supple raptor]

try parted bangs in your prompt and curtained hair in the negative

[torpid plover]

Oh so smart thanks im french too it's a bit more difficult for me

[brittle jay]

hair between eyes perhaps?

[torpid plover]

Will try when I cna

[brittle jay]

also bangs

[supple raptor]

Fair enough. Even as a native English speaker, the danbooru tags aren't always the most intuitive.

[torpid plover]

I try to find the site of danbooru i can't find it

[torpid plover]

Yup

[cyan burrow]

swept bangs, parted bangs

[merry tendon]

why?

[brittle jay]

I thought about making such a lora myself

[merry tendon]

wanna have my dataset?

[brittle jay]

yes

[supple raptor]

That is the saddest throne I've ever seen.

[merry tendon]

😂

[compact warren]

@brittle jay what are the reasons for running into a cuda error with dadapt when all my settings are fine

[brittle jay]

idk

[compact warren]

And ive trained with this much space before

[merry tendon]

@compact warren didnt you help me with my dataset some time ago(the anime in RL one)?

[brittle jay]

cuda error is usually when you don't have enough vram

show settings maybe?

[compact warren]

I vaguely remember seeing pics of that in here but idr if i did or not

[compact warren]

Ill dm

[stone onyx]

using @supple raptor's VAEs

[brittle jay]

what is clearvae?

[supple raptor]

The flower in the top left shows perfectly why I don't like any of the NAI based VAEs.

[brittle jay]

looks best imo

actually nvm

I sense that it is based on nai vae

[merry tendon]

going off👋

[brittle jay]

so it bad now

[stone onyx]

https://civitai.com/models/22354/clearvae

[merry tendon]

gn👋

[supple raptor]

They claim it's a merged VAE. I'm not convinced it's anything more than NAI with the contrast cranked up to like 1.2 or 1.3. It fell on it's face with the denim texture and tried to turn it into blue reptile skin or something, and it has the same bad texture on the foliage.

[stone onyx]

YUP, I never liked anime2 ever since I started using orangemix vae, it's a little washed but it's easily fixable with a bit of editing

[brittle jay]

just used blessed version

[merry tendon]

kl-f8-anime2 is best

[keen hollow]

yeah it's a good model, been having a lot of fun with it. one of those toon models.

[merry tendon]

gn guys👋

[supple raptor]

orange is the NAI vae, just renamed. look closely at the flower and you'll see the repeating texture that I'm talking about. It usually does really good on the character, but looks worse on backgrounds than kl-f8-anime2. If you haven't tried it yet, you should give the reduced contrast versions of kl-f8-anime2 a try.

[cyan burrow]

I need to photoshop it right tho

[brittle jay]

guys https://bard.google.com/

[keen hollow]

Anyone fartin' with the new animation stuff?

[cyan burrow]

there, better, that's the right version

[merry tendon]

💀

[supple raptor]

@stone onyx @merry tendon Lower contrast kl-f8-anime2 here: https://huggingface.co/nubby/kl-f8-anime2-blessed

[merry tendon]

why would i want lower contrast

[stone onyx]

I'm testing out 8 and 9 atm, not sure which one I like better yet

[supple raptor]

It can be a bit blown out at times on the regular one. I really like the 0.9 version.

[merry tendon]

ok going to sleep now fr👋

[stone onyx]

but I think I'll probably like 8 with some photoshop color correction

[supple raptor]

I might experiment with adjusting brightness a bit too since that's the other variable that the Blessup tool gives. Just need to come up with a good varied set of prompts to test them on for a fair comparison.

[stone onyx]

imma have my eyes peeled for that 👀

still trying to find a VAE that I actually like

[keen hollow]

messin' around with Bard now

let's see how it goes

[supple raptor]

Yeah, but is it actually decent yet? Last I heard, it was really bad.

[brittle jay]

idk

for now it seems worse than chatgpt even

[keen hollow]

so far it is better than ANY of the local models I've run

but I just started kicking the tires

[steady grail]

i would prefer loras to lower colors. would be cool to set colors like in a tv lol

[cyan burrow]

Oh nice, I managed to prompt exactly what I wanted first try

[supple raptor]

Last time I saw anyone talking about Bard, the general consensus was that it was significantly worse in every way than even GPT-3 was when it first came out but it's been a while since then.

[keen hollow]

They track every word you type into it, so .. there's that.

[cyan burrow]

I can't show it though I think it might have too much blood

[steady grail]

talk in code with chat gpt lol

[cyan burrow]

or maybe not

It has pooled blood on the floor and dripping from a characters mouth

[steady grail]

measure the pressure, or add a bottle of ketchup in the image

[brittle jay]

bruh

wizard lm good

and stable vicuna

[keen hollow]

I found them to be rather awful.

[keen hollow]

I mean.. relatively. 😛

[supple raptor]

WizardLM was quite good, especially for a 7B model. It even impressed me after being used to 13B models.

[brittle jay]

no model can do math bruh

[gloomy scroll]

Default is best

[brittle jay]

me: solve for x. 7-3x=-5
much trial and error
bard: there is no solution to the equation.

[cyan burrow]

Skeletor took some nudes to slide into DMs with

the joke here is not sexual btw, bot

[brittle jay]

ah yes "nudes" != sexual

[stone onyx]

sometimes I do like the default, though other times it kinda deepfries the image

[cyan burrow]

not for skeletor

that being the joke

[brittle jay]

https://cdn.discordapp.com/attachments/807809192537882647/1091905253739331584/attachment.gif

[supple raptor]

Using LLMs to do math seems like a dumb use of LLMs anyways. The better use would be to teach the LLM how to input the math problem into a calculator app so that the math is actually solved, rather than being predicted.

[brittle jay]

ofc

unless you are talking about academic level math

but that is only for like gpt4+

[supple raptor]

Right, but that's still just predicting answers. That's what LLMs do. They look at the text input and predict what should come next.

[spring kestrel]

Bnw

[brittle jay]

I think that models like gpt4 have already transcended that. ofc at the core it is just predicting likelyhoods of tokens, but from that it get other capabilities

[supple raptor]

It's kind of like a kid who can write all of the correct answers on their math test but fails anyways because they didn't show their work. (yes I know that GPT-4 can sometimes show it's work so it isn't a perfect analogy.)

[cyan burrow]

teach an LLM to teach other LLMs, make an LLM a jack of all trades teacher, create an LLM taskmaster that provides a task to the assigner, that calculates what experience is needed to reach the goal. It then creates that number of LLMs and tells the teacher LLM to teach each a specific thing. After which they will be instructed to research and learn more on their own

[supple raptor]

?

[brittle jay]

I meant that it gets actual other capabilities. stuff you would think you need to have thoughts for

like did you see the paper on it?

[spring kestrel]

@supple raptor devil imagine

[supple raptor]

Yeah, it's pretty amazing what it can do. I'm just saying that using it to try to guess answers to a math problem is a pointless usecase when we already have calculators that can actually solve the math problem for us while spending far less compute power to do it.

[brittle jay]

depends on the problem

geometry for example might be harder

[supple raptor]

I'm not sure I understand. Are you trying to generate an image?

[brittle jay]

you are devil image

[supple raptor]

Right, which is why I said that we should teach the LLM how to input the math problem into a calculator app instead of trying to teach it to actually do math. Solving a complex word problem by condensing it into an equation that can be solved with a calculator would be both more reliable and more efficient.

[brittle jay]

wait wtf google collabed with character.ai???

[compact warren]

[supple raptor]

Asks for fountain of life
Get's deep puddle in a cemetary

[stone onyx]

I think I'll settle with bless9 for now, seems like a good middle ground

[stone onyx]

swamp butterfly

[supple raptor]

That was my favorite too. I don't want to have to change VAEs and 08 is sometimes too washed out.

[stone onyx]

It IS really annoying having to swap

I'll just keep 9 and try to forget about this whole VAE thing lmao;

[supple raptor]

Not exactly the fairy wings I asked for, but pretty none-the-less

[ocean reef]

evil fairy

[mild tree]

I'm tripping...
WHAAAAAAAA

[supple raptor]

Whatever she told me to do, I'm pretty sure the correct answer is "Yes, Mommy"

[ocean reef]

or death by snu snu?

[supple raptor]

in that case, I amend my answer to "No"

https://cdn.discordapp.com/emojis/1093687562071179284.gif?size=64

[ocean reef]

😏

[distant gale]

bit of a script i wrote that makes a video out of sampling steps

idk how to make it full res yet tho

[supple raptor]

There's an extension that does this. You might be able to find something to improve your method from this one: https://github.com/vladmandic/sd-extension-steps-animation

[distant gale]

oh cool thank you

[distant gale]

remove detail LoRA

[ocean reef]

I use detail:-0.6 now, effect is rly nice 😄

especially on models which like details

[brittle jay]

how are you managing to make images so small??

bruh

[distant gale]

wdym 512 by 512

[brittle jay]

no

368x368

[distant gale]

oh right it is 368 by 368

[supple raptor]

Might be grabbing the Approx NN or Approx cheap versions rather than full res.

[distant gale]

i used full -1 in that image

[brittle jay]

oh

is does seem like it

[supple raptor]

Oh, you're talking about the image. I missed that and thought it was about the video.

[distant gale]

i dont really want to run it in higher res than 512 by 512 on a 4 year old crusty laptop

[brittle jay]

well both

[distant gale]

i love this effect

[distant gale]

it is approx nn and i dont know how to make it full res

its set to full but it still doesn't output full res

[supple raptor]

Hmm... I'm not sure either

[wispy canopy]

30b LLM makes characters feel much more alive, 13b model feels smarter

[wispy canopy]

me in my younger years... wait i never looked that cool XD

[brittle jay]

which 30b LLM?

[wispy canopy]

https://huggingface.co/TheBloke/OpenAssistant-SFT-7-Llama-30B-GPTQ fits just barely on a 3090

[ocean reef]

It's not for you to decide!

[wispy canopy]

if i looked cool or not? 😛

[ocean reef]

yup!

[distant gale]

geez

[ocean reef]

coz for ourselves we always look bad

so shush!

[distant gale]

me vibing with an mx 250 2gb

[wispy canopy]

didn't have the bangs like that 😛

[wispy canopy]

tbf, i got my 3090 just so i could run 3090 models

[ocean reef]

Then it's even better 😄

no complains!

[wispy canopy]

now i just need a wizard trained 30b model

the openassistant models are pretty cool though, they don't seem to be censored at all, at least, that specific model

[supple raptor]

Does it have enough space to avoid OoM or swapping to RAM once a conversation hits 2048 tokens?

[wispy canopy]

it's been made with groups 1024, so i think it's swapping, i see my vram filled out. still 'decent' speed though

[distant gale]

how much time does the model take to respond?

[wispy canopy]

pretty fast actually, when not swapping local LLM's generate about 15 tokens(words)/sec

[distant gale]

cool

[wispy canopy]

i've been using them to make my image prompts, but in my own weird style lol

[nova remnant]

what language model are you using for prompts?

[wispy canopy]

switching between gpt4-x-vicuna 13b, wizard-vicuna-13 13b and Openassisant sft-7 30b

i have to say, using the openassistant in combination with tavernAI gets the characters REAL lifelike

it's conceptual awareness is amazing, it's deductive capabilities a bit less

[nova remnant]

interesting, I'll check those out

[wispy canopy]

be sure to get the 4bit gptq versions

[nova remnant]

I only started playing with LLMs a few weeks ago, thought about using it for prompts but the story models I've tried can barely take inference from the prompt, much less confident in their ability to carry the format for SD prompts lol

[wispy canopy]

the 7b models? pretty much useless in that regard

honestly, i think stable-vicuna is the best 13b model, however it was seemingly trained on top of vicuna 1.0 and has annoying sideeffect that it starts talking to itself

[nova remnant]

it's not an open source LLM if it doesn't start talking to itself sometimes 😅

[wispy canopy]

to run 13b models you need 12gb at least... 10gb is just not enough 😦

you can fit the model in memory, but you are left with too little vram to run inference with effectively

[nova remnant]

I have a 3060 12GB which is decent enough but I ran one on a 3090 with runpod and it has me wanting to upgrade

[wispy canopy]

your 3060 would beat my (ex) 3080 -> you'd not get the slowdowns

[supple raptor]

Yep, my system uses about 11.75GB VRAM during inference on 13b models once I hit 2048 tokens.

[nova remnant]

that's why I got it tbh, I'm still a 1080p gamer, just want the extra VRAM for AI stuff

[ocean reef]

G'night man for the end of the day 😴

with no hands 🤣

[wispy canopy]

no homo, but that's a sweet lookin dude ❤️

not those buff-cakes i've been baking

[ocean reef]

hahah true 😄

[wispy canopy]

it's hard getting proper husbando's tho

you either get angsty teens or buff-cakes

[ocean reef]

ineed

[wispy canopy]

the in between seems kinda... hard

[ocean reef]

try to add mature to prompts but not always help

[ocean reef]

and If I add old they are rly old...

[wispy canopy]

tends to get you really old dudes

yeah. that

well-aged? 😄

it worked when i tried doing architecture in a forest

made stuff look rustic, but well maintained

[brittle jay]

is that an existent character?

[gloomy scroll]

No idea stole Salt's prompt

[keen hollow]

Finally something different. 😛

[wispy canopy]

whaddaya mean different? 😮

You ever caught me being consistent, or just tired of generic waifus?

[keen hollow]

the latter lol

[brittle jay]

what is it?

[gloomy scroll]

BREAK
(exceptional, best aesthetic, new, newest, best quality, masterpiece, extremely detailed, anime:1.2), walking through village, town, neighborhood```

[wispy canopy]

my llm took a shot xD

seems it missed the sailor collar xD

even after 5 tries... no mention of it... does it even know what a sailor collar is ?:O

[steady grail]

i don't know either lol

try sailor ribbot

sailor neck ribbon

sailor neck handkerchief

[wispy canopy]

uh... my llm+sd made a bimbo xD

lol

[gloomy scroll]

llm?

[wispy canopy]

my LLM bot

[gloomy scroll]

Oh

[wispy canopy]

the one that writes my prmopts

[cyan burrow]

what are you after?

[steady grail]

forget about eface, the true genius is his bot lol

[keen hollow]

Is it a samey-bimbo?

Or something different?

[cyan burrow]

yes and our true genius is stable diffusion

[steady grail]

true <.<

[wispy canopy]

nothing in particular XD

[cyan burrow]

sailors collar?

or whatever

[wispy canopy]

getting that right in a natural language prompt is hard

[cyan burrow]

is it? I dunno what you want to get right

[wispy canopy]

meh, i'll just BREAK the og prompt in there

[steady grail]

are you talking about this, righ? https://external-content.duckduckgo.com/iu/?u=https%3A%2F%2Ftenshishop.com%2Fwp-content%2Fuploads%2F2017%2F01%2F739.jpg&f=1&nofb=1&ipt=a89fb728ee71aeea46e0d7f8841f50573487c09973029e21e1555deb28c9d1c6&ipo=images

[wispy canopy]

chefs kiss i like this ^^

[steady grail]

how about interrogating this image?

[gloomy scroll]

Does your model know what serafuku is?

[wispy canopy]

it's a hentai model. IT'D BETTER

[wispy canopy]

girl, serafuku

[gloomy scroll]

There's your sailor collar

[steady grail]

lol

[wispy canopy]

i like the floral sailor collar better

[steady grail]

spring serafuku?

sakura serafuku?

[keen hollow]

the rendering on that is sort of interesting

what does that word mean?

[steady grail]

sailor uniform

[steady grail]

lol

at least it's cute

[cyan burrow]

@wispy canopy I want to rty too but I dunno what you're prompting for

[wispy canopy]

nothing, just throwing in random stuff i stole here

😛

[gloomy scroll]

Doesn't rly look like tears of joy

[steady grail]

try to add (nekomimi:0.2)

[keen hollow]

ah I see ... what's the deal with that anyways?

is sailing like really big in Japan or something?

everyone owns a boat?

[wispy canopy]

sailor moon.

aight, imma go work on my pow now

[cyan burrow]

there you go, sailor dress!

from my prompt to your screen, weeb out, good night

[steady grail]

nighty night

[slow needle]

hello first time posting here

[sage kiln]

Great initial gens. You didn't inpaint anything right?

[slow needle]

yup

[supple raptor]

I sensed 2B posting

[sullen hollow]

great pictures

[sage kiln]

This was the initial gen btw, inpainting is great

[sullen hollow]

inpaint is what gigachads use

[hearty heart]

you fixed hand. nice

[slow needle]

also tried to generate a lunar tear but its not quite there for me

[sage kiln]

First one is already really good no?

[slow needle]

a lunar tear have 5 petals but i guess i can fix it with inpainting

[sage kiln]

Did you specify that in the prompt?

[slow needle]

yeah i did

[untold glacier]

From the looks of the images, it seems like a low CFG gen--try increasing the CFG for better defined gens and a greater prompt influence. Though I will say that we haven't quite reached the ability to define numbers and have a reliable associated output with SD 1.x.

It may be best to find a lunar tear image, photoshop it in, and then i2i from there.

[slow needle]

yeah here the thing... this is a sd 2.x model

[untold glacier]

That's beyond my knowledge then, though I haven't heard of a greater numeral reliability with 2.x.

My initial recommendations should still be helpful though, process-wise.

[sullen hollow]

or generate a huge batch of low res images and pick out the best seed that come close to 5 individual petals

[small pecan]

https://media.discordapp.net/attachments/1038609371367747624/1106356740406124644/Cimg7a04b842f85b2e49eb7b.png

[sullen hollow]

blub blub

what a view wow

i can see my home from here

[small pecan]

Hi I am a fool who cannot open webm on mobile

There’s probably a way but I’m dumb

[hoary hawk]

"Are you there God? It's-a-me, Mar-ga-ret!"

[empty coyote]

Lot of trail and error to finally get a decent ultra wide image out of SD. The tiling controlnet really helped.

[sage kiln]

Damn is this composed of multiple individual images or how did you achieve that

[sage kiln]

Can you link the controlnet?

[empty coyote]

I started at a 1280 x 360 generation. Then scaled up using https://github.com/Mikubill/sd-webui-controlnet control_v11f1e_sd15_tile preprocessor and model along with the https://github.com/Coyote-A/ultimate-upscale-for-automatic1111 Script for upscaling. Did some in paints and some manual updates using Gimp along the way. kept feeding results back into the control net for the next tiling upscale until I got to the resolution I wanted.

[sage kiln]

Thanks. I already have ultimate upscale. I use R-ESRGAN 4x+ for upscaling, what did you use?

[empty coyote]

4x-UltraSharp which I got from here: https://mega.nz/folder/qZRBmaIY#nIG8KyWFcGNTuMX_XNbJ_g/folder/3JZDSYZJ after watching this video tutorial: https://www.youtube.com/watch?v=EmA0RwWv-os

[sage kiln]

Thanks 👍

[keen hollow]

I need to watch one of those tiles tutorials.

just can't stand that they're always 45 minutes of useless information and 2 minutes of useful info

Bard's legit.

[empty coyote]

I find Sebastian Kamph's videos to usually be very helpful and to the point.

[keen hollow]

Right on

[wispy canopy]

finally, done with pow

now i can back to random posting. finally

[keen hollow]

nice

imma go see

nice, looks neat 🙂

[wispy canopy]

dont think it can win tho

oh well ^^

i think it came out pretty

[tawny helm]

Well done Eface.

[wispy canopy]

i... think i spent more time on the description, working with my llm getting a good concept down

[tawny helm]

I decided to try out Adobe Lightroom to start adding a few touches to my generated images, so far it seems pretty useful.

[wispy canopy]

not quite a pixie lol, it got the mischievous look down tho ^^

Music Pixie!!!

@tawny helm

i took the liberty of removing the sword -> i actually want workable images lol

[tawny helm]

Oh my Christ you turned me into an Elden Ring character....or...better yet, maybe Final Fantasy deity. Haha

[wispy canopy]

it keeps coming back to nature/wolf(or dragon) themes

BONUS ROUND. THIS DUDE SLAYS

[tawny helm]

Welp, first job in Lightroom completed. First image is the original and second is the Lightroom work. Either image looks fine in my opinion. And to Eface, yes he does slay. lol

[wispy canopy]

is it me or is there nearly 0 difference?

[tawny helm]

Oh....no.

Give me a minute. Haha

[wispy canopy]

lol 😄

[tawny helm]

Lightroom exported the wrong image.

[wispy canopy]

oofff. imma be simping these dudes ^^ I don't think they'll be giving up their loot easily

[tawny helm]

Okay....okay....I fixed it.....first image once again is original....second is Lightroom. Ugh. lol

[wispy canopy]

100x better contrast

[tawny helm]

The first one in my opinion while it looked fine, it seemed...washed out.

Just from a stylistic perspective.

[wispy canopy]

it's the redacted vae?

[tawny helm]

Actually I haven't used that VAE in eons. Haha

[wispy canopy]

(or any of its descendants)

[tawny helm]

I've been running either the WD 1.4 one or the standard SD 1.5 one.

[wispy canopy]

yeah, those are pretty much the best

[tawny helm]

That new Blessed VAE that came out is like a nice hybrid between the two as well on color saturation.

[wispy canopy]

@dull jackal -> you've been turned into a prompt

[wispy canopy]

aight, imma go to sleep, byebye ^^

[dull jackal]

Could have been my male counterpart

If you figure out which one...

[sage kiln]

Man controlnets are sick

[pearl rivet]

Trying to redraw my gen with mouse and SAI paint

[keen hollow]

evenin'

anyone still up?

[terse ore]

First time I made a generation from blender that includes a character, props, and HDR background. Rendered image input:

1Used Controlnet Canny and depth. 1920x1920 upscaled with esrgan anime4x and then turned back into 1920x1920.

Models tend to make skin tone lighter. I should have prompted it to keep him tan.

I'd need to manually change the car's side mirror color, and some cloud artifacts. And perhaps manually adding some lineart ( I did use the thicker lines lora, and the add detail lora too). But it seems like this method can save some time. Perhaps this is the workflow I can use to make panels consistently. I got a 15.6" monitor drawing tablet for $100 (GAOMON) a month ago but haven't opened it yet, but its time to use it for manual work now. Never made a comic/manga/webtoon before but this one will be a comedy. Hopefully it will go well

[trim current]

Wheres me

[keen hollow]

I should really be working through my video game backlog instead of generating images for no purpose. 😛

[terse ore]

Do games have a purpose...?

[compact warren]

????

[terse ore]

?

[compact warren]

Whats the point of such an obvious question

Or am I missing something

[terse ore]

Well, games and generating stuff both have the same purpose, which is no purpose

[strong thunder]

Final try made with Parseq

[compact warren]

What?

[terse ore]

yes

[compact warren]

No way you’re serious

[terse ore]

Well in that specific case yes becasue the guy specifically said "for no purpose"

[compact warren]

Who did

Oh omnia

I see

[tawny helm]

Okay, I do not regret working with Adobe Lightroom on my images after generation one bit. If you know what you're doing you can really bring out some nice details. This one is only minor though, just modified the sky, trees, and masked over the water. First image is original and second is post-Lightroom.

[trim current]

Is this extension difficult to use?

[trim current]

jeez

[strong thunder]

A tuto here. https://youtu.be/MXRjTOE2v64

And here
https://youtu.be/n4zj1lrbIEM

[wispy canopy]

the ever elusive milk-goblina

[wispy canopy]

Ok, one genderbender sunny coming up 😄 (I think think the LLM+SD did somethign interesting though, even if they missed the assignments)

[wispy canopy]

you shouldn't have shown your face. now i'll be turning you into a prompt next!

[acoustic agate]

[wispy canopy]

look at this dude having his own emote xD

[acoustic agate]

Whatever you turn me into, just don't forget to include the SENPAI hat

[wispy canopy]

you're asking a LLM model to tell a SD model to add a senpai hat? we're not in 2200 yet

[acoustic agate]

[wispy canopy]

oh well, i do cheat with BREAK on other user generations -> so i suppose i could try to add it manually

you don't have a senpai hat lora by any chance? ^^

[acoustic agate]

Nope. But I got a senpai hat.

[wispy canopy]

did... did you just nina miku?!

[acoustic agate]

No, King Pendragon did back in december. I'm scrolling through the archives and reposting stuff.

[wispy canopy]

@acoustic agate you can't make this shit up 😄

[trim current]

I love it 🙂

[acoustic agate]

waow

[wispy canopy]

i had to cheat a little and BREAK joy, else she looked too sullen

[acoustic agate]

Most of the time folks make fun of my username with T-Rex Delete.

Fantasy Dinosaur is a new one.

sick

[wispy canopy]

the Openassistant 30b is really damn smart when it comes to this

@acoustic agate damn, this is a cool character concept!

[acoustic agate]

30b 🫠

you and your damn a100s

[wispy canopy]

dude, this on my 3090 😄

you can run it too

[acoustic agate]

4bit?

[wispy canopy]

yeah

lemme search the model

somebody quantized the sft7 model

[acoustic agate]

imagine using quantized models instead of their full versions

[wispy canopy]

ah, of course it's thebloke https://huggingface.co/TheBloke/OpenAssistant-SFT-7-Llama-30B-GPTQ

[acoustic agate]

I want to run 30b native

[wispy canopy]

dude, even the a100's 80gbs have trouble doing that (solo)

lol

[acoustic agate]

17gb model lmfao

les gooo

So what I really need is an array of 6 H100s on a local server

[wispy canopy]

ez, just buy SSD for AI only

[acoustic agate]

taken care of, I've got at least 10TB of nvme ssds. Some raided.

[wispy canopy]

please don't say raid 0, please don't say raid 0

[acoustic agate]

raid0, and backed up onto raided sata ssds

[wispy canopy]

you and both buddy ❤️

[wispy canopy]

my sysadmin heart dies everytim

[acoustic agate]

kekw

[acoustic agate]

my C drive is a single gen4 ssd

[wispy canopy]

i don't do raiding anymore

[acoustic agate]

the raided drives are primarily for bulk game storage

[wispy canopy]

i like my pc as a frankenstein/theseus machine lol

[acoustic agate]

an H100 80GB board due in the second half of the year is set to cost around $33,000