🏞｜general-with-images

[south heath]

how to use vae on anything v4.5 or other than v3

[storm sundial]

settings

search VAE

after clicking on show all

[south heath]

i use v4.5 and showed this in settings

[storm sundial]

you need to put the vae in the VAE folder within your models folder

then you can hit refresh

and select

[south heath]

okay i will try

[vague oasis]

looks cool i will try it when it comes out

[dry crow]

for most anime models a vae is needed, OrangeMix has its own vae, same as for anything v3, v4 and 4.5 uses the v4 vae too.
The files end with .vae.pt These go into the models/stable-diffusion folder and should be renamed to match the model name there are for:
Example.ckpt
Example.vae.pt
Then select Automatic for VAE in the webui settings. Then SD will choose the right vae for each model and you dont have to switch

[tranquil kindle]

I've noticed why I can't even load these images and idk why.

[wispy nest]

@dry crow What it has done is change the model of the car perfectly fine

[dry crow]

@wispy nest Try blue, Ford Focus 2006, and also delete not blue from negative

[wispy nest]

ok

Will have to try later, gotta head out now but I will let you know if works when I get chance.

[dry crow]

Okay 👍

[strange cliff]

perfect hand

great for playing piano

[wispy nest]

https://tenor.com/view/facepalm-really-stressed-mad-angry-gif-16109475

Ok so now I have a silver one with blue mirrors.

I think I will give up, it obviously cant change from orange to blue

[dry crow]

Did you tried (blue:1.3)

?

[wispy nest]

I am trying ((blue)) at the moment but will try your suggestion later

after I mean not later

((blue)) did give it some blue doors like its had an accident and the owner couldnt get the right colour replacements

Could it be the seed that it's using as its not currently set to -1. I didnt set it to the number it is on but it matches what it was in txt2img?

[dry crow]

i dont think so

can you share me the original car?, would like to try it too 😄

[wispy nest]

Your blue is the same result as mine

[wispy nest]

Seems to only add blue to the mirrors.

[dry crow]

metallic xD

[dry crow]

got it!!

[wispy nest]

How?

Though thats on the original and not the 2006 version but still .... how ?

[kindred willow]

There's a spy in our ranks

[dry crow]

((blue ford focus))

the trick is Denois Strenght at: 1

((blue ford focus 2006))

[wispy nest]

Hmm ok, thanks 😄

[kindred willow]

Any thoughts on this?

[wispy nest]

Kill them.

[kindred willow]

Who could it be though 🤔
He could be any one of us
He could be you, he could be me!
He could even be-

[wispy nest]

Not me cus I cant grow a beard

[kindred willow]

Well, can't be me, I shaved my beard yesterday
looks at viewer

[wispy nest]

Ive tried growing a beard but it just doesnt happen. Crap genes from my father lol

[kindred willow]

You don't need it, sometimes it gets in the way lol

[kindred willow]

A still frame from the live action sci-fi spy thriller movie "Among Us", the scene where Daniel Wroughton Craig proclaims "There is one Impostor Among Us, and you're looking pretty sus" to his clone aboard the Starship Skeld

[dry crow]

@plucky star
here you have a black pyramid with golden accents, looks a bit like mekka xD

[plucky star]

MAAAN I LOVE YOU

[dry crow]

np

Tags: one big Black pyramide in sahara, (black with golden accents:1.3), intricate details,

[plucky star]

❤️

Can you create one with a more dark atmosphere? Or with a storm?

[dry crow]

sure give me a sec

[plucky star]

thx

[split timber]

how can i make it more precise?

[plucky star]

Do you wanna marry with me?

[dry crow]

haha no thx 😄

[dry crow]

precise in what?

[split timber]

Like make the mountains in a better quality. Cuz you can see sort of pixels idk how to say it
The grain maybe?

[kindred willow]

A still frame from the live action sci-fi spy thriller movie "Among Us", the scene where the actual imposter makes himself known on the escape pod, "Sir, you were the sussy imposter the whole time? How could you betray us!?", looking at each other

[fervent rune]

I trained a LoRA on the style actually, could be better but it turned out pretty good

[dry crow]

Higher resolution, clear sky and other Quality tags and Upscale can help

[spring wave]

oooh

[spring wave]

that looks good. so is there anyway to improve it?

like train more?

[fervent rune]

Yeah I think it needs further training, it was only 2000 steps over 2 epochs (1000 per epoch)

[spring wave]

how long did it took you?

[fervent rune]

40 minutes

[spring wave]

you train on your won machine?

[dry crow]

What GPU ?

[fervent rune]

3090 using Kohya_SS trainer

[dry crow]

Ahhh okay

[fervent rune]

yep!

[spring wave]

oh nice

[dense haven]

Is there a trick to make stable diffusion generate a portrait with neutral lighting?

[strange cliff]

[kindred willow]

Also, does the guy on the left look familiar to anyone? Dude looks like a real actor but I can't place the name

[kindred willow]

CPU moment

[deep valley]

Silver (╯°□°)╯︵ ┻━┻

[split timber]

Ok ty

[peak forum]

Someone knew why I can’t write in the chat?

[analog goblet]

hi guys 🙂

whats new ? 🙂 2.2 out yet ?

[vague oasis]

#📣｜announcements

[wispy nest]

Káiser

[hasty nova]

Sneak peak from my next embedding 👀

(And this is only eualer a, 20 steps, 512x)

And yet another...

[hasty nova]

cover image I think

[frank thistle]

SpongeBob

SpongeBob

[hardy juniper]

Create a caricature of Lionel Messi lifting the champion's cup. Features the phrase 'Messi: The Tireless Champion' and a colorful, vibrant style that draws attention. Ideal for a limited edition notebook cover ¯_(ツ)_/¯

[distant idol]

training my own custom model, pretty happy with the results :)

[hasty nova]

Embedding released! https://discord.com/channels/1002292111942635562/1071271601699553320

[spring wave]

btw if you ever do train it more, please share the result with me. those looks very similar like vagabond.

[fervent rune]

Will do! Feel free to add me so I can get it to you easier

[spring wave]

oh ok, i'll add you

i was thinking of training a lora using google colab or renting a gpu

bcs i dont have a powerful gpu that i can train with

and renting a gpu is way cheaper

[fervent rune]

If you want some instructions for good results so you can get the most of your time spent on the gpu, I can help you with the setup and configuration of the training

[spring wave]

i'd love get some advice on it

i'll try to train a lora in 3 hour

with google colab

just to get the idea on how to do it

[sullen pecan]

https://cdn.discordapp.com/attachments/1026788486818320408/1071309777650585620/grid-0020.png

first attempt at training on Undone (Tv Series)

https://tenor.com/view/alita-battle-angel-undone-rosa-salazar-ripples-gif-15575848

[grave schooner]

Lovely! 💜 Will u share the embeddings?

[spare lantern]

db, finetune, lora or embedding?

[sullen pecan]

Lora, using Kohya-SS's webui.

I'm on an 8gb card, so far Loras have been the only models ive been able to train. takes bout 1 hour 30 mins for me, if i had a 4090 ti, it'd be within 5 mins 😄

[strange cliff]

hand holding VS 4000 words negative prompt who will win?

damn

[tranquil kindle]

I like this 3d art, and your name 🙂

[near bough]

ahmad logo design

anyone tell me how to use this for create logo

[exotic stirrup]

Show me vector wine glass with ice with the warm weather.

[grizzled sage]

First know logo design and all the language that goes into designing a logo and what goals you'd want to accomplish and what concepts you want to communicate. Then ask stable diffusion to create something for you with that knowledge.

just typing a prompt and using that as a logo, i think you're going to have a very bad an unneffective logo that blends in with 1000 others. i don't see stable diffusion creating something iconic in it's current state

the number one rule i follow with logo design is that the logo should scale across resolutions and color depths. while you might get a cool looking "logo" thats got photoshop filters and 3d effects, it won't be a good logo

[untold jasper]

POV: Casually getting a new model

[strange cliff]

when you put in 3 model does it merge all 3 at once or do you have to merge 2 at a time?

cuz C wasn't on the formula

[dry crow]

Weighted Sum is for merging A + B, C only is used with Add Difference and not get merged in

[strange cliff]

ah

[dry crow]

No, OpenjourneyV2 is out for months

Depends on what you like, but OpenjourneyV2, Dreamlike, Protogen are realy good models

Hmm okay you can also inpaint stuff that you want to add to an image

[weary stratus]

@open sun here are some generations i made that i liked

[open sun]

https://civitai.com/models/1116/rpg would be the best imho

for this

[weary stratus]

ic

but like

[open sun]

no there isn't "overally" best model out there 😛

[weary stratus]

i just want the style and colors of these

i like that

the empowering feeling

but not necessarily how good the angel looks or whatever, would actually be fine if it was kind of blown out lol

[open sun]

there are popular models but not the best, those images you can achieve with any model but with RPG it might be easiest.

[weary stratus]

aight thanks

i like how the angel looks here

where its more of a silhouette

probably just gotta work on my prompt for that lol

yeah it seems like rpg kinda adds too much detail to the character for my liking

this one looks cool tho

[spiral umbra]

Realistic Vision V1.3 Model

Prompt - Realistic portrait photography, A attractive white female model, showcases the model's athleticism and strength, featuring her in a gym or sports field setting. The focus will be on her physical abilities and the surroundings will enhance that aspect, providing a dynamic and energetic atmosphere, highly detailed, detailed skin, detailed face, high detail eyes, close-up, canon eos r3

Negative Prompt - out of frame, duplicate, watermark, signature, text, ugly, morbid, mutated, deformed, blurry, bad anatomy, bad proportions, cloned face, disfigured, fused fingers, fused limbs, too many fingers, long neck, nude, naked

Step 250 + GFPGAN / CFG 11

[acoustic eagle]

tfw "Hello World" for ai art ==

[distant idol]

haha thank you! 😊

some of the new gens from my re dreamboothing the model

[tranquil kindle]

Good 3d art, what are the prompts that you use?

I also forgot to mention your name is like "Saphina" or something like that.

[distant idol]

haha totally xD. It's short for saphiresurf, a username I made when I was like twelve xD

Also kinda works for sapphic haha

Also prompts wise, I used positive and negative prompts

The negative is mostly bare basic filler but with some specifics

And some of the keywords are specific to the model I trained

[tranquil kindle]

So cool.

[distant idol]

Positive: "a blue drink in a glass against an orange backround, light refractions, stylized realism, vector illustration vctor, colorful flat design, dau-al-set, isometric, hard normal shadows, caustic shadows, solid background, cubo-futurism, gradients, professional, clean, geometric"

Negative: "multiple objects, jpeg quality, low quality, artifacts, jpeg artifacts, paint, water color, text, watermarks, titles, multiple objects, language, disfigured, kitsch, ugly, oversaturated, grain, low-res, Deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb, blurry, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, long body, ugly, disgusting, poorly drawn, childish, mutilated, , mangled, old, surreal, patterns, grids, speckles"

Thank you! :)

[tranquil kindle]

I've tried your prompts just then. It's not really as good as yours, but yeah.

[distant idol]

Yeah the finetuned model helps quite a bit xD. It's good to know that it's actually doing something beyond what the normal model can do lolol

[tranquil kindle]

True

[distant idol]

the reflections on the inside of the glass on some of these though

refraction type thing

always get me 😭

[cunning geode]

im confused as to how to use automatic1111 inpainting/sketch to add an arm to this image:
can anyone help?

[dry crow]

you can try inpaint sketch, then say arm as prompt

denois has to be changed for better results i think

[cunning geode]

yeah and thats what im confused by because i tried that but it didnt work?

[dry crow]

you need to select the 1.5 inpaint model first

[cunning geode]

what were your denoise settings?

[dry crow]

0,78

but it was just to test

the rest was default

[shrewd wadi]

Hi guys good morning share your prompts for today! happy weekend everyone!using usp.ai

[cunning geode]

after changing from "only masked" to "whole picture" and "original" to "latent space" and adjusting the prompt its working
https://cdn.discordapp.com/attachments/1023277529424986165/1071622457280184451/image.png

[sudden steeple]

https://i.redd.it/yt7mwkveg8ga1.jpg

Does this look diffusion generated to you guys? Artist is claiming it as original work but it looks sus to me

Some other examples:

https://artstn.co/m/z27WN

[grizzled sage]

i see a lot of standard conceptualization techniques. photobashing, brushing, layering, smudging, stuff that you'd learn with gnomon workshop dvds or other cool digi art courses

can't spot anything i'd call an obvious ai error

[oblique owl]

Hey All! So, I am a developer and have deployed some of the popular models on a server. Currently the server structure I use means I can't use a VAE, so it rules out some models which require one, or have one baked in.

My goal is to create super high quality and detailed images of sci-fi landscapes. And I know Midjourney is probably ahead in many ways, but I am trying to improve and determine if my issue is mostly my prompt or model selection.

I'll give you an example below, without any major prompt engineering, actually the following prompt is created by GPT. And I'd be interested to know your thoughts / experience / ideas.

Example: given the prompt:

In a #dunedust city, the dark & eerie cyberpunk-style environment is illuminated by dim #neon lights. A true sci-fi classic!

Using the model:

scheduler= "DPMSolverMultistepScheduler"
        model = "darkstorm2150/Protogen_x3.4_Official_Release"

I get this image:

Midjourney image

What do you think, should I work on my prompt or find better models without a VAE?

[mossy agate]

/drea

[strange cliff]

is it possible to merge VAE with a model?

[dry crow]

yes its called inbake an is in the checkpoint merge tab, you have to select Interpolation and then model + vae and inbake

[strange cliff]

ah thx

[waxen bramble]

Autos has a script for it I think. BUt you can just replace the VAE on diffusers and the pack that to CKPT.

CKPT is just a fancy zipfile.

You can open it with 7zip.

'

[tardy latch]

caminante indigena ¯_(ツ)_/¯

[dull trail]

a bird sitting on andrew Tate's head, hq, 4k

[vague oasis]

a photo of a bald man with overbite, canon camera footage, 4k, muscle, sunglasses, beard, instagram post

[hasty nova]

A sneak peek of my next embedding :)

[wispy nest]

The Tate escaped prison

[hasty nova]

Tis done :) https://civitai.com/models/6841/adventure-diffusion

https://discord.com/channels/1002292111942635562/1071827026991915049

[minor osprey]

https://media.discordapp.net/attachments/1004159122335354970/1071829203802148874/WhatsApp_Image_2023-02-05_at_11.47.58.jpeg?width=663&height=663 ┬─┬ノ( º _ ºノ)

[wispy nest]

that looks epic mate

I'm using two embeddings here and a custom mix off a new base that what I was using previous

[vapid timber]

any ideas how to create something like this?

which model is capable to create imgs in this style?

[prime leaf]

hello guys, where can i find the "dreambot" to make the pictures? pls

[wispy nest]

the discord bot is retired

[wispy nest]

SD 1.5 based checkpoint data with a textual inversion embedding trained off something similar

or prompting / mixing checkpoints / both until you get it consistent

[prime leaf]

):

[prime leaf]

so it is no longer possible to make the images by stabble diffusion?

[wispy nest]

no, the discord bot is offline you can still use any offline UI or online but that bot is not online, it was a tempoary service, might be back in some form who knows

[prime leaf]

I understood, only midjourney is working right?

[prime leaf]

i understood, only midjourney is working right?

[wispy nest]

nah theres loads of stuff out there dream ai, etc etc, if you want stable diffusion best way is to download a checkpoint data like SD 1.5 or 2.1 with a UI like AUTOMATIC1111

[cunning geode]

i tried a whole day to fix the arm but i am unable to do it better than this 😦
https://cdn.discordapp.com/attachments/991315789695172698/1071852343802462408/00150-941651376-female_humanoid_dragon_wearing_suit.png
https://cdn.discordapp.com/attachments/991315789695172698/1071852344351920148/00205-719412879-hand_in_pocket_three_buttons_character_art_of_female_humanoid_dragon_wearing_red_suit.png

[prime leaf]

show, the midjourney too is amazing

[prime leaf]

show, the midjourney too is amazing

[left sparrow]

do any of you have any idea of how to get a witch's hat made of fire? maybe img2img?

[left sparrow]

😦 inpaint sketching sort of works

[dry crow]

hey nice one!

[wispy nest]

now I'm making the overall mix style custom trying to just test the versatility and theme of the mix, added in a little "anime ish" and heres with a hypernetwork!

[dry crow]

oh wow

[wispy nest]

its getting there still kinda messy but yeah ty ;'D

no hypernetwork or embeddings

( diff seed )

with embedding

[dry crow]

both nice

eyes of the first are top!

[wispy nest]

and i'll show with the hypernetwork one sec ;'D

[dry crow]

trying different styles rn

[wispy nest]

thats dope

[dry crow]

the low poly firefox mascot 😄

[wispy nest]

I like how the ahir and tail has a lower poly look

and here it is with the embedding and hypernetwork both added

[onyx forum]

Is this a gmail account login or something else?

[dry crow]

you need credentials from the host, who made the sd public for you

[onyx forum]

where do i get it

[dry crow]

very nice, the face got more details it seems

is the hn a special pose or an clothing?

[dry crow]

from the person who gave you the link

[wispy nest]

sxzBloom_sxzBloom.pt

[dry crow]

ahh nice

[wispy nest]

"trained on some Midjourney generations so it makes anime generations slightly more semi-realistic and adds some bloom and blurred background with some soft light."

yea thats quite nuts now

[dry crow]

i like the green spacesuit like hoodie

[wispy nest]

deffo very midjourney-like lol

[hasty nova]

It looks like

[wispy nest]

what I think is crazy, just taking out the positive prompt for the hypernetwork and the embedding exact same as before on settings gives

oh, thats with 0.25 dosmix added too

hang on

damn it look way worse with dosmix added

here's without:

ehh idk actually its different

actually the anime biased one i recon is better

[dry crow]

yea the one above is better 😄

[wispy nest]

think im gunna runa few tests and then stick with that tbh

[hasty nova]

one word prompt

[dry crow]

got inspired by the dark green spacesuit hoodie

[vague oasis]

it is 🌝

[wispy nest]

niice`1

[dry crow]

this one got more prio to spacesuit 😄

[cunning geode]

@dry crow what did you train that model on?

[dry crow]

@cunning geode on nothing, its just a merge model and normal prompt without any embed, lora, etc

[static tusk]

I will give your newest embed a spin

[hasty nova]

awesome :)

it's a little strong, so you might want to tone it down with prompt weights. I didn't in the above images, but it is definitely a really powerful style.

[static tusk]

ok no problem

Ive used many that are super overpowering so fairly used to that

[wispy nest]

awesome

[sterile kiln]

Arwen

[wispy nest]

@hasty nova @dry crow

[hasty nova]

Looks great! Advntr embed I assume?

[wispy nest]

bloom hyper and nebmgk embed !

[hasty nova]

Oh

Why was I pinged? Just curious

[wispy nest]

I havent got around to trying out your emeds on it yet but I shall be very soon 😄 also just cos you two i talk to most lol

[waxen bramble]

"Don't mind me, just shitting a shoe..."
"Yeah... I know I should eat more fibre"

[hasty nova]

Ah I see

[waxen bramble]

I laughed harder at this than I probably should.

[wispy nest]

actually running that through adventure is a good idea

I'ma do it now ;'D#

[hasty nova]

Make sure to put it at the end of the prompt, it's quite a powerful embed

[wispy nest]

gotcha, before hypernetwork ?!

I'll try both again lol

[hasty nova]

Idrk how hypernetworks work so idk

[wispy nest]

had to makea few renders just cos the hands, usual stuff and seems a slight bias to making a ball of light with this prompt and then ...

epicness

[static tusk]

got caught up with some things, thats why its taken a while.

[wispy nest]

• removed hypernetwork in this case to see just the embed*

[hasty nova]

Woah @wispy nest and @static tusk , those are all great! Thank you for sharing

[dry crow]

Is this your new embed?

[wispy nest]

welcome as always

[hasty nova]

yes, one of them

[dry crow]

Wow good work!

[hasty nova]

Thank you!

[dry crow]

Realy nice

[static tusk]

yeah looks great dude. nice work

[hasty nova]

I tried making it more for cool landscapes/settings but included some characters to make adventurers for your worlds

[wispy nest]

no embed but just showing off this anime style mix I've done, random seed, same everything else

[hasty nova]

oo nice work!

[static tusk]

@hasty nova and for something completely different

[hasty nova]

Very spooky

What embed did you use?

I really like the last one

[static tusk]

just yours, custom model and Giger style

[hasty nova]

advntr did that?

[static tusk]

yeah

[hasty nova]

Dang

Didn't expect that

[static tusk]

sorta lara croft

back to safety, away from the horrors

[hasty nova]

oo pretty

[silent pagoda]

Pizza dancing in a 1970s disco

[frozen nebula]

What prompts are you using to get the scale that epic, I can do wide views but never quite that epic

[static tusk]

"amber_atmosphere, high contrast, mad landscape, (advntr:0.9)". its the merged model that is doing most of the heavy lifting

[daring iron]

A cartoon cat soaring in a balloon

[frozen nebula]

What is it a merge of? I'm guessing the landscape one and something more smooth / anime looking ?

[daring iron]

I couldn't see this for makoto and so i created my own version

Mitsuha and taki from Your name makoto shinkai movie kissing, trending on art Station, (makoto shinkai anime movie style) , anime style,your name sky night for background, background full color, without hands in the image, improve lips, improve lips definition, pixiv design, modern art design, fan art design

[static tusk]

inkpunk and woopwoop

[frozen nebula]

Cool thanks, not heard of woopwoop before

[hasty nova]

was this also using advntr? I could link that here if @frozen nebula you are still looking for some cool "epic scale" images with just an embedding

[static tusk]

yeah I pasted the prompt above

[hasty nova]

oh my eyes glossed over it because of the weighting around the token 😆

[hasty nova]

https://discord.com/channels/1002292111942635562/1071827026991915049 Try this embedding I made, it can get some pretty epic shots I think

[static tusk]

another from that same batch

[hasty nova]

Oh very pretty

I like the autumn colors

[frozen nebula]

Just downloaded it from civitAI and gave you a follow for your fine work 🙂

Started off using MJ only just got into SD had patchy results at first, but once you get past the MJ bad habits it produces amazing stuff with the trained checkpoints and embeddings

[hasty nova]

tysm!

[frozen nebula]

Messed around with your embedding and some upscaling for a bit, it works really well, doing dark future landscapes was tricky previously (for me at least)

[hasty nova]

I'm glad it works! And that looks really good

[pallid ruin]

Some models have some sampling methods broken U.U

[quaint heart]

anyone knows how to fix this?

[dry crow]

Webui wont work for AMD GPUs on Windows

[sick cosmos]

anyone know how to not get so dull images?

so when it's rendering, the colors are pretty warm and nice but once it finishes, it just dulls the colors

[dry crow]

Your missing a vae file for that Model

[sick cosmos]

no I have it and I think it's the vae file that's doing this

bcoz it's just with this prompt and any image with 2b in it

[dry crow]

Which model and VAE?

Can you send an other image i could compare

[sick cosmos]

Anyhentai and it's vae
https://civitai.com/models/5706/anyhentai

[dry crow]

Yea definitly the vae is not aplied

[quaint heart]

oh thanks for answering, i guess can't use automatic 1111 XD

[sick cosmos]

oh

how to apply it then

[dry crow]

It only works on Linux, but for Windows have a look at Shark from Nod.ai

[dry crow]

Make sure the vae.pt file and the model share the same name:
Example.ckpt
Example.vae.pt

In the Webui settings under stable Diffusion vae should be on Automatic

[sick cosmos]

ohhh

and in the vae folder?

[dry crow]

No

[sick cosmos]

not the vae-approx folder, right?

[dry crow]

Both in the models/stable-diffusion

[sick cosmos]

ohh

is that right?

@dry crow

[dry crow]

Was the vae a vae.safetensor ?

If so then yes it looks right

[sick cosmos]

I guess it was

bcoz the command line looks to be logging fine

[dry crow]

Looks good 👍

[sick cosmos]

:D

thanks

[dry crow]

Np you will see the big Difference it makes

[wispy nest]

https://civitai.com/models/6984/czeromixv6 just released my first checkpoint data ❤️

[sick cosmos]

oh yes absolutely

its so beautiful now

[willow chasm]

Put monochrome or similar in negative prompt

[sick cosmos]

nah nvm, sorted it out, I was missing the vae

[proven lava]

Hello I am constantly getting blurry faces on my results any idea, blog where I could search some information on what could be happening?

[shut sinew]

usually bad faces come with low res, sometimes face restoration helps, i prefer to get closer images of faces for better results

[dry crow]

try using portrait as tag and change the resolution to 512x768 also activate face fix

[livid ermine]

Hey! can y'all help me dress up my great-grandmother in something more fashionable? Like some red carpet-like party dress from Net-a-porter or Victoria's Secret.

I did some restoration at Replicate but this is as far as I was able to take it.

Maybe loosen up her hair and give her some basic makeup too?

(will pay ofcourse)

[steel yarrow]

Heisemberg fingers

[rain gazelle]

yo

good gravy, things got complicated in here

[trim shoal]

Mahatma Gandhi

[livid ermine]

https://cdn.discordapp.com/attachments/1023277529424986165/1072265900771786834/photo_2.jpg getting there!

[steel yarrow]

did they just wipe out the themed image channels or what

[dry crow]

hey my images get always a pale look with any vae i try, with your model 😮

ah its more like 50/50 with wd1.4 vae

[wispy nest]

It does have a bias towards that, however it's not all the time , it';s not only able to gen white people either!

[dry crow]

okay ill try more

[wispy nest]

literally just inputted those two words

[wispy nest]

Looks dope to me 😄 have you tried 2.1 yet? I am going to start trying to work on embeddings and stuff for it

[dry crow]

not much, just a bit, waiting for good models based on it 😄

[wispy nest]

Once this headache goes away I'm going to start looking to mix and train mfor it

[pseudo aspen]

Some fantasy themed diffusions

[round kelp]

Playing with different models and prompts

[velvet hemlock]

Is there a good tutorial out there for how to embed or whatever a stable charicter?

Like, I want to create a character and then have sd reliably generate that person, in different art styles, situations, locations, etc

Not really sure where to start with that

[tawdry geode]

For a created character with a good seed and face, there is no way (engine limit). What you can do is called dreamboothing its a model that you will generate using 15ish or more images of your character. Dreambooth is the best way to get detailled images with good proportions and skin while keeping the face, the only downside is it takes an already existing subject and you need a few images. So if you need a custom character you would need to create it in a 3d editor or use some face swap or face animation to generate images for the different poses and then dreambooth it.

[mint token]

i generated this but this is not amo g us

[green mountain]

What... is going on in the first picture...

[hasty nova]

What a nice spot on the civitai leaderboard :)

https://civitai.com/leaderboard?position=69

[lime raft]

hey my images get always a pale look with any vae i try, with your model 😮

¡ generated pikachu

[sterile kiln]

Hello guys, is there a way to use several models on a X/Y/Z plot script?

[dry crow]

Hey, yes you can select checkpoint for that

[woeful pewter]

Anyone else feel that illustration and anime (even with retro/vintage) results are oversaturated? Even when adding desaturated and a negative prompt with weight oversaturated ? The render previews are great but then the final image it boosts the saturation to like 140%

[dry crow]

You can use an other vae like AnythingV3 to get it desaturrated

[woeful pewter]

Will give that a look. Thanks

[rain raptor]

Can anyone offer any advice on what model/settings the creator of this photo might be using to get this result?

[sterile kiln]

You could use "Preprocess images", in train Tab.
Its basic but could help

[sterile kiln]

A simple test, but works 🙂

[dry crow]

You may need to add vae files for orange and Elysium to get better colors

[sterile kiln]

Is it possible to add it only on certain models in X/Y/Z Plot script?

[dry crow]

Yes if you rename the vae file to match the model name and then use the vae setting on Automatic it will pick the right vae for the right model:

Example.ckpt
Example.vae.pt

In x/y/z you cant specify on which model they get used just on all

[midnight fossil]

can someone tell me if there's a sd model that look like this one here? i tried a couple and did not get the same style, i find this one more defined and detailed. thanks

[strange cliff]

Do you know the prompt for it?

Easier to test that way

[mint moth]

Hi

[strange cliff]

Howdy

[rich crown]

Logo

[strange cliff]

Lego

[midnight fossil]

yeah i had been trying last night, i got pretty close but the "anime styles" i got did look so sharp. i guess that's just a midjiourney thing there. if you search for "anime landscape" on instagram there are lots of images that look the same.

[wispy nest]

maybe try this one https://huggingface.co/andite/pastel-mix

[midnight fossil]

i don't think this one is similar, but thank you 🙂

[plucky star]

hey

i need a wallpaper

send me plz

[spare tiger]

Can you explain to me what the differences are between image to image and image to image alternative?

[willow plank]

https://giphy.com/gifs/barbaderespeito-logo-beard-w6KELKs8hQm44acK7O

[drowsy plover]

🤝

[toxic rover]

I love AI lol

[sinful urchin]

Weird purple spots showing up on images recently. notice the top left of the skull. Model: Anything 4.5

[grizzled sage]

those are artifacts from over training, poorly trained models. anything 4.5 is made by a couple of pranksters who don't like how popular anything 3 is, put out an over merged and broken model cand called it v4. their goal is to troll and make people not like av3 anymore.

just use av3 it's better.

[sinful urchin]

Oh, that's unfortunate. 4.5 looks really good except for those weird purple spots.

[real grotto]

how to animate?????

[vernal prawn]

How many images are required to train a custom model?

[latent remnant]

#🏞｜general-with-images himachal traditional family in the mountain surrounded with trees and snow

[exotic cedar]

Y

[strange cliff]

You’re saying that but other than artifacts it look real good

[dry crow]

You have to use a vae, choose the one from AnythingV4 and rename it to match the 4.5 model

[distant idol]

Around 4000 for a full finetune in my experience

[plucky plinth]

Realistic photography in high definition, with very high detail, of a woman in a flamenco dress from the city of Seville, Spain, with a floral background and colorful photorealistic style and a bright sky with some clouds and sun, all with maximum photorealistic detail in 4k

[exotic socket]

Does anyone have any prompt suggestions/ideas for getting an image that looks like this in terms of art style?

[sinful urchin]

Am I supposed to see some vae setting here?

[real grotto]

Video game vector art, video game cartoon art, etc.

[dry crow]

Go into the settings, under User Interface, Quicksettings and add , sd_vae, then apply and restart

[exotic socket]

Thank you!

[sinful urchin]

I added it to quicksettings, but nothing seems to be appearing still

[dry crow]

It now should be on your Main Window

Besides the model dropdown

[smoky oak]

Just started training my very own LoRAs, and they are amazing

First one was focused on Taylor Lautners face, cause its something that protogen struggled with

Protogen Taylor Lautner

Vs mine

It was just something to practice on cause the images were easy to find

then I trained a LoRA on how to do freshwater shrimp, cause I like freshwater shrimp

Protgen

My LoRA

Safe to say the improvement is noticable haha

[wispy nest]

So hi, I am a noob to the whole AI arts. But I made an anime cover for the book I am reading, hope you guy like it 🙏

[sinful urchin]

hmm I just can't understand how to do this vae thing. I still have no setting appearing.

[dry crow]

Did you run the latest Version of the webui?

[sinful urchin]

I don't know if I have the latest version. Does this update, if it's not the latest?

[dry crow]

You have to edit your webui-user.bat and add git pull to it

Then it will update everytime you launch

[sinful urchin]

okay, I'll do that thanks

[winter pilot]

Hi Everyone!
With a friend we just release a new App where you can create Stable Diffusion Images :).
I really like to have some feedback so if u can test it i will be very happy!
https://play.google.com/store/apps/details?id=com.dreamai.app

For now is only available for Android because we are a team of 2 (dev and designer).

thanks!!

[sinful urchin]

So what do I set GIT equal to here?

[sinful urchin]

Oh, okay

alright, it's installing new requirements

[sterile kiln]

Hey guys, where can i change default 7860 port?
I'm trying to use Koyha_ss Git and its on the same port. I would change the Automatic1111 port

[strange cliff]

wait what if you merge 2 model with VAE already merge in them which VAE persist? or do they both got remove?

[sinful urchin]

Thanks CS1o

[sinful urchin]

VAE error

it won't make images... any suggestions?

its error completing request

wait, but now it suddenly works again..., okay

[dry crow]

Edit your webui-user.bat and add following behind CommandlineARGS= --port xxxx to make the server listen on a specific port, xxxx being the wanted port. Remember that all ports below 1024 need root/admin rights, for this reason it is advised to use a port above 1024. Defaults to port 7860 if available.

[sterile kiln]

@dry crow my man. Thanks.

Trying this : https://github.com/bmaltais/kohya_ss, *Aitrepreneur *(a youtuber), talked about 6 min of training... Mad

LORA is unbelievable

6-8 minutes instead of 4 hours...

[dry crow]

what gpu do you have? 😄

[sterile kiln]

2070 Super

[dry crow]

hmm maybe i should try it then with my 1080

[sterile kiln]

My friend have a 1080 Ti, 1080 Ti is a bit more powerfull than 2070 Super

Do you have a Ti or not?

[dry crow]

nope

[sterile kiln]

sad

https://www.youtube.com/watch?v=70H03cv57-o

[dry crow]

yea, i tried embedding training but it said it takes 30 hours xD

[sterile kiln]

Try this

[dry crow]

yea ik the video

[sterile kiln]

6:32 minutes 😮

[dry crow]

thats insane

[sterile kiln]

Hope its a good model

[dry crow]

how much images?

[sterile kiln]

10

I will set my face on Protogen2.2 :OOOO

[dry crow]

nice 😄

[sterile kiln]

@dry crow Lowest default settings : 20 steps, 7 cfg scale, 512x512

[dry crow]

wow, can you try yourself as a knight?

[sterile kiln]

(my real face)

[sterile kiln]

i will try

[dry crow]

works pretty good! would go on 30 steps

[sterile kiln]

i will try a X/Y plot with differents CFG and steps settings

@dry crow

jeez

from a 6 minute trainning....

[dry crow]

very nice, i need to try it this weekend

[sterile kiln]

it's really me, it's crazy

[sharp beacon]

this is with what dream booth ?

[sterile kiln]

Lora

[sharp beacon]

oh. How do I install that?

in the webui

[sterile kiln]

See the video, scroll

[sharp beacon]

okay thanks. i see it

[real wigeon]

hello, question, stable doesn't have a prompt book yet right? or a channel here like midjourney?

thanks in advance

[wispy nest]

Hi all, would you be so kind to tell me a cool prompt to create this style?

[real grotto]

You look like Lucifer from the hit tv show Lucifer (that's a compliment)

[real grotto]

Crayon, kids drawing,

Those are some of the things I'd think of

[cunning geode]

https://cdn.discordapp.com/attachments/1044638177140412446/1072998896017625178/grid-0052.png
https://cdn.discordapp.com/attachments/1026990051185401917/1073000854115848262/00001-157000110.png
https://cdn.discordapp.com/attachments/1026990051185401917/1073000672909340702/grid-0044.png
https://cdn.discordapp.com/attachments/1026990051185401917/1073000762445156453/grid-0030.jpg

https://cdn.discordapp.com/attachments/1026990051185401917/1073002255604777000/grid-0054.jpg
https://cdn.discordapp.com/attachments/1026990051185401917/1073002255843868773/grid-0055.jpg

https://cdn.discordapp.com/attachments/1023293330601287711/1073003113323180092/grid-0056.jpg
https://cdn.discordapp.com/attachments/1026990051185401917/1073004298415394826/grid-0057.jpg
https://cdn.discordapp.com/attachments/1023293330601287711/1073015003151601794/grid-0071.jpg

new screenshots from the newest version of my experimental model

tails, wings, capes, dragons and transformations still dont work as intended unfortunately. but i am working on it.

[sterile kiln]

😄 thanks

[restive magnet]

Do any of you know the settings/prompts to turn a flat colored piece of art into having volumetric (or at least decent looking) lighting and shadows?

[real grotto]

cell shading perhaps? Or use pix2pix and say please add (insert type of art shading, such as cell)

[wispy nest]

not sure but i love powergirl

[restive magnet]

Hell yeah brother I wanted to make shading my comic easier

[restive magnet]

is it an extension, I tried using img2img but it gave me bad results.

[real grotto]

Well yes, but technically no. Basically you want to go here:
https://huggingface.co/timbrooks/instruct-pix2pix
and then download the model. Then using Image2Image and the model just use it and make sure you have the referance image

basically what pix2pix is is stable diffusion mixed with gpt2

so you wont be able to use it unless you have the model

im not entirely sure what the diffusor is for tho and weather or not to use it

[restive magnet]

alright thanks I appreciate that. I'll try to mess around with it.

[vague oasis]

this is my attempt with pix2pix, not very good: the prompt is "make her a unreal engine render"

i used this btw: https://huggingface.co/spaces/timbrooks/instruct-pix2pix

[restive magnet]

yeah it might be because it is dark already, oh well I was hoping to save hours of time lol.

[wispy nest]

gave it some cool depth

[sterile kiln]

For my french (or not), patriots :

[shadow jolt]

So I'm guessing that moving all the channels for specific generations to "archived communities" means this is no longer a place for that, right?

[jovial sage]

https://images-ext-1.discordapp.net/external/3M918jj1L9muJ09DhTZ_Oi9Qvj7XWq7z8KPb99iU9nI/https/s.mj.run/_zjKrEai2N0, retrofuturism style, A highly detailed and hyper realistic painting of a steampunk armour trending on artstation, sharp focus, studio photo, intricate details, highly detailed, by H.R.Giger, Tvera and wlop and artgerm and greg rutkowski, insane detail, oil on canvas, moody, eerie, mystery, artstation, sharp focus, cinematic lighting, studio photo, intricate details, highly detailed, concept art, trending on artstation, sharp focus, studio photo, intricate details, highly detailed

[smoky oak]

Thats how I just started a few hours ago, already trained some of my own in less than 30 minutes

insanely cool

I trained a LoRA on freshwater shrimp, cause it sucks at them, and I love them lmao

Standard protogen blue shrimp

Mine, trained off of just 15 images

I am currently training it on 30 images this time to see if I can get it even better

it can even generate shrimp its never seen, such as purple shrimp (which are impossible)

its insane how fast these LoRAs can be trained, with such insanely good results

[static tusk]

I cant wait till SD can do words and letters. making fonts would be amazing

[glad galleon]

Maybe you guys can help me. So recently I used a different AI to interpolate about a a bunch of real faces into a single face for two male and female characters, the result was the images shown below. I know a single image isn't enough to do an embedding, do any of you know of something that can take the single image and show the faces at different angles and create enough images, or a movie where i can screencap the frames and make an embedding with these characters?

[wispy nest]

anyone interested in a 2.1 mix that produces far better images than stock for anatomy but a bias towards scifi and fantasy

just running some examples and then I'll upload it

[smoky oak]

Does anybody know how to right this error?

I have been training my own LoRA's, and the first 2 work just fine, but the other 3 all give that same error

[violet tulip]

thought you arnt allow to make ai art of real ppl

isn't that a general "No-No"?

i mean if its cool, its cool

I'll make some brittany murphy art 😋

i miss brit murphy

she was a good actress

forgot to tick on hi-res fix

d'oh

ugh doesn't really look like her, but at least the AI knew she was a blonde without me saying anything

that's cool

best of 5

bleeeh

used no negative prompts

that was abyssOrangeMix

trying babes

mmm hey yeah, that kinda has her face

lol

trippy

[dry crow]

Hey for orange mix you need a vae file to get better colors and Details,
also try to set the resolution to 512x768 for good Portraits

[violet tulip]

rip brittany murphy

models/ldm/stable-diffusion-v1/vae-ft-mse-840000-ema-pruned.ckpt

like this one?

cool thnx

[dry crow]

Yes this should work

[violet tulip]

cool

i'll retry it

i dont know where to find any other VAE, i think that is the only one

no one makes them besides just the one

[dry crow]

For OrangeMix there is a vae on their Website

[violet tulip]

https://civitai.com/

?

no there isn't

if you mean this place

if they have an actual website

then idk

[dry crow]

https://huggingface.co/WarriorMama777/OrangeMixs/tree/main/VAEs

[violet tulip]

oh on hugging face

which is a github like site

but not a website for orange

[dry crow]

Yea i mean the HF site

[violet tulip]

wasn't orangemix made with vae in it

https://civitai.com/models/4451/abyssorangemix2-hardcore

huh i guess not

[dry crow]

You can see there is no used vae in this picture

[violet tulip]

in wut picture

i dont even know what a VAE is

[dry crow]

In the picture you send first

[violet tulip]

yea because it wasn't added

i just added it

[dry crow]

Vaes are just like finetuned additions to a model, to give more color and Details

Most anime models require a vae

[violet tulip]

huh, then they should make better models lol

or I should just make my own

[dry crow]

Most models have a vae integrated

[violet tulip]

other ppl dont have to do shit i say

she kinda thicc

doesn't look like britany murphy at all tho

[dry crow]

😄

[violet tulip]

just looks like a generic kinda fat blonde chick

and with big tits like that if don't work out she's just gonna get fatter

and don't get mad at me ladies...

she's not real

I can womanize my art all I want.

lol

[dry crow]

xD yea but keep the discord rules in mind

[violet tulip]

what do the rules say about expressing your feelings about your art

[dry crow]

Nothing, just an Information

[violet tulip]

i just ask because i don't read the rules

[dense tapir]

AI art of real people is allowed just as a fyi

[violet tulip]

cool

i dont have to read them when others will tell me

[dense tapir]

Yeah, sad what ended up happening to Britany.

[violet tulip]

ye

she was a good actress

[dense tapir]

Yeah, had a long acting career ahead of her for sure.

[violet tulip]

I liked her in the movie Spun

yeah, hollywood might actually be making good movies if she were around

now all hollywood wants to do is make white ppl evil and chopping your dick off cool

[dense tapir]

Naw, Hollyweird woke would still be there and she would either be woke or find work as a waitress again.

[violet tulip]

oh right

brittany would be evil cause she is white

[dense tapir]

Yep

[violet tulip]

sorry for making the modern world ppl ppffff

so dumb

[dense tapir]

Thankfully there are indies now and I so much prefer them since 2015 to Hollyweird nonsense. Use the force, Luke has a whole new meaning now.

[violet tulip]

prob has something to do with a dildo

as if

[dense tapir]

They retconned him to be gay

[violet tulip]

it was one thing bk in the 90's when they fought to get married, like fine, now you can deakl with devorce too happy?

but noooooooooo

now they want drag queens reading to our kids

afk gotta wiz

[dense tapir]

You should see the list of retconned into the alphabet community they did. I don't mind new stuff but when retcon happens it angers me.

[violet tulip]

idk wut u mean

last thing i learned from a gay friend of mine was the bandana code

[dense tapir]

retcon is when they retroactively change the continuity of a character, or plot, etc...

[violet tulip]

and some of them are not cool

at all

[dense tapir]

They made a lot of the Star Wars characters into the LGBTQIA group now.

[violet tulip]

have you seen the gay bandana code?

👀

look it up

[dense tapir]

Oh, yeah flagging

flagging became sagging etc...

[violet tulip]

that's how some of the gay chomo's find each other

[dense tapir]

I grew up when flagging was a thing so I know.

The Hanky Code

[violet tulip]

ye, its more of an 80's queer thing to do

mhm

[dense tapir]

70's and 80's which led to sagging a decade later in the prisons

[violet tulip]

which some how, became cool?

weird huh

[dense tapir]

yeah, blew my mind

[violet tulip]

all the thugs wanna let everyone know "hey, im open for business"

[dense tapir]

I still see it here but rarely now.

yes

Anyway, ai art is allowed for real peeps else I would have tossed all of this out the window.

[violet tulip]

i'm straight, but there is one cool thing about gay bars, well in the town i was in, they would sell cigs for 1/2 price, and had unlimited drinking cup night

[dense tapir]

I just can't use an image of a real person to make money with legally unless they gave me permission. Politicians I can.

[violet tulip]

i would go and buy the max 2 packs at 1/2 price get a cup and pound a 6pack n leave

[dense tapir]

As my late wife used to say they can throw one hell of a party even with straight people around. No sex or anything just fun.

[violet tulip]

but on that night it was common to see some of my straight friends also taking advantage of that night

[dense tapir]

smh at half priced smokes, lol. How?

[violet tulip]

idk, i think one of the bar tenders was getting smokes from mexico

it was a bar in San Diego

[dense tapir]

WOW, my friend was right a carton of smokes in the UK is about 150 pounds. FFS, that is almost 200 USD.

[violet tulip]

fuuuuuuuuuuuck

i pay about $30 a carton

[smoky oak]

My v2 freshwater shrimp LoRA was a huge success

[dense tapir]

smh, I used to pay 6 with 2 off coupon. I quit long time ago though.

[smoky oak]

untrained protogen

trained Lora v1

and my new v2

[dense tapir]

I am really loving LoRA

[violet tulip]

nice

[smoky oak]

Same here, fast to train, easy to do, fantastic results

[violet tulip]

i can't use them

[smoky oak]

this is just 30 images

[dense tapir]

fast as all get out and uses so little vram

[violet tulip]

i use invokeai

😦

[smoky oak]

yeah

[dense tapir]

btw, over 20 LoRA degrades

[smoky oak]

I have been training these LoRA's sub 10 minutes

[smoky oak]

oh does it?

[dense tapir]

Yes

[violet tulip]

ya'll use auto1111 huh

[smoky oak]

yes

[violet tulip]

yeah its nice

but i cant use it

i have to use invokeai

😦

[dense tapir]

Fo styles I use 6-10 and for subjects no more than 20. I hang out in some LoRA groups and yea, my 30 was too much. Trains faster and far better.

[smoky oak]

is it a sort of thing where I should get 20 good images and give it more steps?

[violet tulip]

wish i knew wtf u guys are saying

[dense tapir]

Well, depends

I train for about 2k steps (1600-2k).

[smoky oak]

I trained this LoRA on different species, colors, and poses of freshwater shrimp

30 images, 100 steps an image, 3k

[violet tulip]

i want to train on something but invoke doesn't allow training

[dense tapir]

I am surprised that worked but when I am talking I mean one concept

[violet tulip]

i could just do it cmdline but ugh

too lazy

[dense tapir]

I haven't tried multi concept and not sure how to or what to change or even why

[smoky oak]

ah, so giving it more variations across things should still be beneficial?

I put in male, female, several colors, different directions, and different grades

I mean clearly, this image worked haha

[dense tapir]

On a single concept it is very detrimental. Multi concept is actually a whole different thing.

Multi concept depends on what you are training with in how it is done

[smoky oak]

one of the images I trained on

[violet tulip]

u really like shrimp huh?

[dense tapir]

Bubba gump

[smoky oak]

I breed freshwater shrimp. It was just something easy to train on cause other AI's are trash at them haha

[violet tulip]

ah

[dense tapir]

What is your training tool?

[smoky oak]

I am using the one AItrepenuer uploaded

Kohya

[dense tapir]

Ahhh, that one. Listen, his vid was good but his technique is really way off

[smoky oak]

I mean, I got damn good results with all the models I have made, so I am happy

[dense tapir]

It was obvious he was new to it and needed to make a vid

Oh, hard to get LoRA wrong

[smoky oak]

My first one I trained on Taylor Lautner cause protogen AI made him look terrible, and i knew it would be easy to get the pics of him

[dense tapir]

I just think in your shrimp instance a multi concept would be the best way.

[violet tulip]

lora's are like embeddings, right?

[dense tapir]

sorta but no

to the end user yes

[violet tulip]

in like you add them to the model u choose

[dense tapir]

yeah

[violet tulip]

so in my case loras are like emdeds

[smoky oak]

yeah, they are low range focused data sets basically

they add focused information/subjects/styles into bigger models

[dense tapir]

Tell you the Hierarchy is like this DB, LoRA, HN, TI

[violet tulip]

idk wut that mneans

[dense tapir]

oops

[violet tulip]

DB database

[dense tapir]

DreamBooth

[violet tulip]

idk wut that is

[smoky oak]

Protogen Taylor lautner

[dense tapir]

HN HyperNetwork. TextualInversion

[smoky oak]

Protogen with my LoRA

[violet tulip]

textual inversion are embeddings

i know that much

[smoky oak]

that was before I used the proper value, it way too high so there are artifacts

that was like .9. I later found that LoRA works much better at like .3

[dense tapir]

I will break this down. Prompt an actor. Does SD know the actor or was it utter shite? If it looks like a body double then LoRA if it is utter shite DreamBooth it.

[violet tulip]

but i use invokeai

so i dont get the options u all get

i can add emdeds, models, and do inpainting thats it

[dense tapir]

Yeah, and I complained to them but no dice they are too afraid of their niche market of the M1. Never going to get extensions with them.

[violet tulip]

i think he's talking to u, i have no idea wut that means

[dense tapir]

I am not even sure if their dream of nodes will ever see the light of day with their fear they have.

[wispy nest]

my second public mix is coming out in like 30 mins, it's SD 2.1 based

[dense tapir]

I was talking to you about InvokeAI

their attitude sucks

[violet tulip]

invokeai doesn't even have 2.0 working 😢

[wispy nest]

2.1 needs a lot of love atm which is why I'm releasing this and free prompts to help people get started with it

[violet tulip]

yea, once i got auto111 working on my gpu, idk how, but it never did it again

[wispy nest]

I have a 1.5 mix up aswell incase youre interested

[dense tapir]

They still don't? Naw, I thought they had it finally at least it worked for me a couple of months ago.

[violet tulip]

still only goes into cpu mode

[dense tapir]

What does it do with 2.1?

[wispy nest]

its more bias toward anime styles and things but quite good for general use

[violet tulip]

i have every correct ROCm software installed

[dense tapir]

Oh, AMD. I feel ya

[violet tulip]

i have gfpgan installed manually too

[wispy nest]

no I mean I have a public mix for 1.5 and one for 2.1 which is coming out shortly :p

[violet tulip]

and k-diffusion

and still

only inokeai knows how to tap into AMD

[dry crow]

You should look at Shark too

[wispy nest]

hey dude you good?

[violet tulip]

idk what shark is, i can't swim 😢

[dense tapir]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs

[dry crow]

Shark from Nod.ai is a webui made for amd (nvidia and intel) works on windows and linux

[violet tulip]

i know i read it

[dense tapir]

if you can't get that to run you have more issues than just your AMD

[violet tulip]

they lie

[dense tapir]

I know people who run it

[violet tulip]

nope

[dry crow]

Hey yea!, what about you?

[wispy nest]

not bad bro just been working like crazy !

[violet tulip]

i have a 6700XT and know how to use anaconda

[dense tapir]

Oh, you are on windows then? Yeah, you are fucked.

[violet tulip]

me?

[dense tapir]

yes

[violet tulip]

Arch Linux

[dense tapir]

well, wtf

[wispy nest]

I thought that AMD perform fine on Linux this whole time?

[dry crow]

Nice you push 2.1 forward! Keep an eye on the weekend there will be a nice model release from cacoa or so

[violet tulip]

it does

invokeai works amazing

but it has zero extra features like auto1111 has

[wispy nest]

awesome, cant wait , more people making TI and Lora for 2.1 sooner we can patch up some of the stripped dataset

it's so fast and makes great highres

[wispy nest]

also really good with correct negative prompting

[violet tulip]

looks nice

maybe there will be an update

and it'll work

[dense tapir]

@violet tuliphttps://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5896#discussioncomment-4559845

read that link

[violet tulip]

sigh, i'll try

[dense tapir]

has a vid on it too

[violet tulip]

dont care about videos

i click the link

[dense tapir]

SOB, they removed it?

[violet tulip]

thnx

doesn't work

[wispy nest]

waifu diffusion 1.4 latest epoc is helping 2.1 quite a lot aswell, not made my mix for that purpose but its in there and damn the quality of some prompts is neat

[dense tapir]

https://github.com/nod-ai/SHARK/

there that works

made for windows though

10/11

[violet tulip]

😒

great suggestion for an Arch user lol

ur heart was in the right place

[dense tapir]

Welcome to minority status

[violet tulip]

ty

[dry crow]

I would join their discord, there are the most amd, Linux, users ive seen, they can help you with rocm, oonx, Linux and shark

[dense tapir]

which card do you have?

[violet tulip]

I have all ROCM software installed

6700XT

[dense tapir]

well, here ya go

https://bbs.archlinux.org/viewtopic.php?id=279528

[violet tulip]

👀

ugh not the arch bbs

[dense tapir]

My way of making it work on an AMD RX 6700XT was as followed:

[violet tulip]

i'v read all of them

[dense tapir]

then something else on your system.

[violet tulip]

I appreciate you

how, lol?

when invokeai taps into my gpu perfectly

[dense tapir]

I am just saying I know Linux users on that same card who use it. Dunno which distro they use though. I know on my Nvidia and Linux Mint I said screw it after 3 weeks thanks to xformers.

[violet tulip]

that use auto's or invoke's

[dense tapir]

auto

[violet tulip]

¯_(ツ)_/¯

[dense tapir]

I actually know of no one that touches invoke

[violet tulip]

thats the cool and shitty thing about linux

now u know 2

[dense tapir]

I lost my root drive of D and all my bookmarks too. Man I was mad and never got it to work. Got everything else but not xformers.

[violet tulip]

wtf is drive D?

👀

[dense tapir]

they were ZERO help

my windows drive I dual boot

[violet tulip]

i havn't used windows since win3.1

[dense tapir]

sorry for ya

[violet tulip]

why

im sorry for u lol

i use linux

[dense tapir]

cause I ceased to like Linux around 2009

[violet tulip]

i'm free

[dense tapir]

I rather be in chains and able to use what I want than to be free and can't use shit, or go through all the hell I did with xformers again.

[violet tulip]

xformers took me like 1min to install

1 pip command

[dense tapir]

on my 1060 never worked

always something

[violet tulip]

true

[wispy nest]

https://tenor.com/view/gigachad-chad-gif-20773266

this is how you look when you use whatever tool is most efficient regardless of the operating system

[violet tulip]

2009 was the last year of the good ubuntu desktop

[dense tapir]

thing about SD shit is everything is so intertwined and this can only work with this and when you get that it breaks something else

[violet tulip]

once someone actually makes a self contained app for SD it'll always be hit n miss

[dense tapir]

I dunno but I gave after 3 weeks of 10h a day on it. Too tired and beat up to continue

[dry crow]

@violet tulip @dense tapir to clarify

[dense tapir]

All I wanted was the little extra vram Linux gives me over windows

[violet tulip]

yes , i realized it after reading it

all the shit i need is in the repo

but why its not in the readme, is kinda fucked up

and lazy

brb, gotta wiz

[dense tapir]

One thing I noticed about the github Python devs these days is that they can't write docs for shit.

[violet tulip]

i have github pro

and there's no "github python devs"

there are just ppl's repos on github

lol

but i know what i think u mean

ppl just suck at documenting there software

which idk y they dont just pay the $10 a month for copilot, it'll do all that for you

[wispy nest]

https://civitai.com/models/7615/czeros-sd-21-space-fi-mix new release

[violet tulip]

i love copilot

i have in on my visual studio code for linux

[dense tapir]

I hate Python with a passion and how it is being misused. Python was made for research purposes so something to get ideas out there fast not to be the finished program. Akin to 3d printing being made for rapid prototyping not the end product on shelves.

[violet tulip]

and i'll credit windows for making a awesome text editor

me too

dude, me too

like if your updated python version breaks the version behind it, how is any better than an Apple product

🔥 burn

lol

[dense tapir]

Well, who'se brilliant ass idea was it to make the guis browser based?

[violet tulip]

that joke piss's off so many python devs

well, the JS guys did that

[dense tapir]

gradio is shite

[violet tulip]

i just installed that but havn't played with it yet

i been playing with making chatGPT api calls

[dense tapir]

my browser eats gigs of ram eventually whereas a real gui sits in k's of ram all nice and tidy

[violet tulip]

browsers do suck ass these days

[dense tapir]

yes

[violet tulip]

they've turned into apps it seems like

[dry crow]

True but not for Firefox 😄

[violet tulip]

bloated

[dense tapir]

I have 16gb of ram and sit with a 12-20gb page file when in automatic1111 so even with another 16 I will still be using a page file sometimes.

[violet tulip]

yeah, but firefox admited to changing the results you get when quering the internet

[dense tapir]

I dropped woke non free speech Mozilla like a hot potato a few years ago now. Was with them from their inception to then.

before them I was Netscape

[violet tulip]

and until someone makes a new browser and not one based from chomium we're all kinda fucked on which one we use

[dense tapir]

😦 yes

[dry crow]

They cant and dont do this, just use a different search engine

[violet tulip]

since they are all made from chromium

they all are made from the chromium base

[dense tapir]

I used Brave until they became woke and evil to their end users to the point they reminded me of the Mozilla devs when they changed. I then came to Vivaldi but ugh. Nothing else left as I tried them all now.

[violet tulip]

they can and admited it

[dry crow]

Source xD

[violet tulip]

it's on one of distro tubes videos on yt

[dry crow]

Then use librewolf if you dont trust them

[violet tulip]

he made a video on why he's leaving firefox... but then like 5 moths later will make a video on why firefox is better

idk, ppl r fucking stupid

[dry crow]

But chromium based Browsers are all bad except ungoogled chromium

[violet tulip]

or im getting confused with duckduckgo

infact ye, i think i am

[dense tapir]

duckduckgo is bad

[violet tulip]

nevermind