🍥｜anime

[elfin bolt]

Some of them might be outdated

[pastel bluff]

not showing up very weird

[elfin bolt]

its likely outdated

7 months ago last commit

check for alternative

[supple raptor]

Agreed. That extension is ancient.

[elfin bolt]

try this

https://github.com/vladmandic/sd-extension-aesthetic-scorer

from vlad

[supple raptor]

I think we've had like 3-4 gradio version bumps since December.

[elfin bolt]

alternatively you can revert to the version where those extensions were supported

[supple raptor]

Cute!

[barren anvil]

Which model?

[slate night]

Oooh that style is really cool!

[final vault]

chotto matte why is she crying?

[supple raptor]

~~The prompt was just a bunch of wildcards plus the style portion ~~ Because nobody gave her headpats https://cdn.discordapp.com/emojis/953452520682975294.webp?size=64&name=ReisenCri&quality=lossless

[final vault]

cruel nubby
gib her headpats

[final vault]

lmfao

[supple raptor]

These are pretty dope. Are you using the bot or local SDXL?

[slate narwhal]

The bot

[supple raptor]

Nice! I'm really looking forward to seeing how this model does once it's finetuned on anime. The test model for WDXL based on 0.9 is looking very promising and 1.0 seems considerably better at anime than 0.9 did.

[slate narwhal]

I'm very limited right now, I'm in another city, in a hotel/guest house, and I only have a phone (not mine) to generate stuff so I'm using the bot for everything

[slate narwhal]

Yeah, it's getting better and better these last days before the release. They will release a prompt book with instructions on parameters, and I hope they keep the promise of releasing some official tools for fine tuning. Yeah I'm also looking forward to the anime fine tunes, the style possibilities are virtually endless

Lots of stuff we would need to do merge and mess with VAE to get good eyes using 1.5/2.1 are already great in the base SDXL model it will help a lot for future tuning

And the hability to mix different materials and art mediums in base is already great, future fine tunes, Loras etc will get really crazy from there

I've read a little about the experiments with refining, if the community improve this refining stuff, maybe it could be possible to port 1.5/2.1 models or fuse 1.5/2.1 models with SDXL, and it would create even more possibilities

[supple raptor]

From what Haru has said, it also seems to take to finetuning much faster than 1.5 or 2.1 did. His model is at epoch 6 right now on a much smaller dataset than WD1.5 used and it seems like it's similar in quality to base NAI already from the previews I've seen. He plans to train 1.0 on the full dataset once it's released.

[magic wave]

though, from what i heard wdxl is training with booru tags, I know someone who is training a anime model using both natural prompting and booru prompting, so I'm looking fowards to that

[supple raptor]

I have a theory that if someone can figure out how to train the small TE with booru tags and the large TE with natural language it might work better than other methods. I don't feel like the small CLIP can understand natural language well enough and I think that the large CLIP is underutilized with just booru tags.

[slate narwhal]

This is really great news, didn't know about WDXL. Things will change a lot

[worthy delta]

the illusion of choice

[wispy canopy]

yes, i will need to be kept in the loop ❤️

sdxl 0.9 so good as base model

[loud falcon]

very nice

[worthy delta]

the noising strngth was at 0.71

forgot to tweak it

[magic wave]

What you mean? :o

[wispy canopy]

you mentioned someone who's not training on booru?

that's good news for me 😄 now i just need to know who to follow 😛

[loud falcon]

i'm out of ideas

taking a selfie before the end

[wispy canopy]

got wildcards installed? good for random ideas

[loud falcon]

aren't wildcards built-in?

[wispy canopy]

it's in dynamic prompts afaik

vlad has it built in a1111 doesn't

[supple raptor]

there's a basic script to add it to Auto

or this UmiAI fork comes pre-loaded with a bunch of very useful wildcard lists and also adds the functionality https://github.com/Tsukreya/Umi-AI-debloat.git

[worthy delta]

gotta ask what is dynamic prompts and wild cards?

[wispy canopy]

script/extension that can change your prompt on the fly

[worthy delta]

ooooo how does it work?

[wispy canopy]

simplest example "{blue|green} hair" would select either one of them

[loud falcon]

automatic1111 has the [x|y|z] built-in right?

[worthy delta]

oooo thats pretty cool

[wispy canopy]

but that's just a small example, it can be made pretty extensive

[supple raptor]

wildcards are a list of words, partial prompts, or even full prompts that can be randomly and dynamically added to your prompt. For example, this image was made with the following prompt
1girl, __derrian/expression__, __derrian/traits/hair/hair_gen__, __user/haircolor1__ __derrian/posture__,\n__Colors__ __derrian/clothing/clothing_gen__, __derrian/clothing/clothing_gen_helper__ BREAK __derrian/backgrounds/backgrounds_gen__ BREAK superb, (sketch:1.1), [monochrome::0.8], (graphite \\(medium\\):1.1), [anime lineart:0.2], (spot color:1.1) <lora:animeoutlineV3:0.5>

each thing surrounded by double _ is a wildcard. the actual prompt used for generation with the wildcards filled was this
1girl, ashamed, medium hair, (multicolored hair:1.2), (light brown hair, light blue hair, light blue hair, dark green), (pastel rainbow hair:1.27628), sitting, red orange bikesuit, necklace, high heel boots, choker, black leggings, lace-up boots BREAK cave,enchanting, underground, crystals, hidden tunnels, flowing streams, bioluminescent mushrooms, shimmering minerals, fantasy atmosphere, adventure vibe, ethereal ambiance, darkness BREAK superb, (sketch:1.1), [monochrome::0.8], (graphite \(medium\):1.1), [anime lineart:0.2], (spot color:1.1) <lora:animeoutlineV3:0.5>

[wispy canopy]

and with wildcards you can do stuff like __artists__ which would select from a premade list called artists

[worthy delta]

DAYUM

thats alot of stuff it can do

[supple raptor]

All of those wildcards that I used are ones that are included in that extension I linked above.

[loud falcon]

people playing chess 4D with their prompts

[worthy delta]

im over here playing hopskotch

[supple raptor]

wildcards are just lazy prompting XD

[worthy delta]

BAKGROUNDS SUPREMACY

[wispy canopy]

llm prompting lazy prompting too 😄

[loud falcon]

you need to see regional prompter

[supple raptor]

Don't have to think of what you want, just have to think of the things that should be described in the prompt.

[loud falcon]

they taught me yesterday and my mind blew

[supple raptor]

also true

[worthy delta]

i see

its a shame no models are properly trained on dudes from what ive seen

only anime girls

[loud falcon]

welcome to the future

[supple raptor]

Some models are better at dudes than others. Here's a super low effort prompt on Kakigori. No cherry picking, these are just the first 4 images it generated.

[worthy delta]

Not abd

Bad*

Very nice ngl!

Wonder if I can generate a room then inpaint areas I want someone to be

[supple raptor]

Yep, I'd suggest using controlnet inpainting instead of normal inpainting though. Here's the extension for controlnet, install it the same way as the others. There are models that you have to download for this one so you'll also need to open the link in a browser and find the link to the controlnet 1.1 models in the readme section. https://github.com/Mikubill/sd-webui-controlnet.git

[worthy delta]

I'll give it a shot when I get the chance

Thank you!

[worthy delta]

That looks pretty good

[cyan burrow]

You should probably delete that before a mod finds it

[cyan burrow]

I use InvokeAI or Gyre AI's webui for inpainting, really well made

[merry tendon]

god i love the ne thoughts approach

[supple raptor]

Unless they're using controlnet for their inpainting, it won't receive the same level of context from the input image as controlnet inpainting does. For actually adding things, I haven't found an option that can compare to controlnet inpainting. For regular inpainting, I'd much rather use the Krita extension than anything else.

[harsh fractal]

exactly！

[cyan burrow]

If I put a box in an area and include a small portion of the existing part of the image (or should I say 25%) then it will continue the image really well 90% of the time

[supple raptor]

You should try controlnet, it's much better. It gets context from the full input image rather than just the bit that you include in the box.

For outpainting specifically, I have a guide on how to do it with controlnet inpainting in the pins of this channel.

[cyan burrow]

well you could include it all in the box if you wanted

shrink it down, include it all, resize it img2img latent stuff and maybe control net to reference it more

though controlnet might be better once set up

I don't have the energy to set anything up though

[supple raptor]

shrinking it down is going to cause you to lose a ton of quality and still wouldn't work the same as controlnet.

[worthy delta]

I barely understand inpainting

[cyan burrow]

Something I'd love is to be able to just select a blank area on say a canvas resized image, and mark it down with a selection box or something. Type in what I want there and have it be generated, either using tiered layers or only tiles

Ive considered buying the Gyre AI plugin for photoshop

it's come a long long way

setting up renders in a literal 3d space with selection tool and placing it in the 3d space where you want it

can create a whole big scene with everywhere where you want it

[bitter latch]

figured out why lora's where not working. seems they somehow got turned off in extensions. had to reselect them and reload. oi vey

[supple raptor]

Adobe will have to pry CS6 out of my cold dead hands and even then I'll choose other software over switching to their subscription BS so Gyre doesn't exist to me. Besides, I thought you had already bought that back when they were flyingdog. Did they change names and stop supporting the old version already?

[worthy delta]

Hey u don't need cs6

I got full Adobe think latest version

And u don't need a subscription

[supple raptor]

cs6 is the last version before they went subscription based afaik and I already own the master suite

[worthy delta]

I took a course for it abt a year ago

And never used it again but with ai imaging

Lord im boutta make masterpieces

But what is the master suite?

[supple raptor]

master suite is what they called the package that came with all of the adobe software from the generation.

[worthy delta]

Ahh gotcha gotcha

Want me to show ya how u can get the latest version of adobe

I really gotta put it on my ssd

[supple raptor]

Thanks, I use it for my business though so I can't do any sailing of the seven seas for my adobe stuff.

[worthy delta]

Ahhh gotcha gotcha

The seven seas are perfect example of use for Adobe stuff LOL

[supple raptor]

I agree. I'm not a fan of doing it normally, but for Adobe I would make an exception any day for personal use.

These days, I use Krita way more often anyways though.

[cyan burrow]

You can get their latest creative studio releases....elsewhere

no they didnt stop supporting flyingdog, that's the company's name

which somehow got confused with the plugin at first

[worthy delta]

I agree, I barely do it anymore aswell only reason I'm on the seas cause the school forgot to change my password so I couldn't work for home

[cyan burrow]

so now they've clarified it as Gyre

they stopped the Krita plugin because Krita publicly stated it would do everything in its power to not let AI onto its software

their justification was "If they are so fucking opposed, we don't even wanna do it"

and I agree

also the Krita one was free

so nobody lost out

[worthy delta]

What is krita

Seems complicated

[supple raptor]

Krita is FOSS so it really doesn't matter what the main dev team wants. Anyone can develop whatever they want for it.

[worthy delta]

Isn't bad idk what a Foss is

[supple raptor]

Free and Open Source illustrating software that's actually good. One of the few FOSS softwares I've used that can actually compete with the paid equivalents. https://krita.org/en/

[supple raptor]

FOSS is the acronym for Free and Open Source Software.

[worthy delta]

Aaahhh gotcha gotcha

I'll give it a look over!

Thank ya

[worthy delta]

DAMN

How do u keep character consistency?

[cyan burrow]

where is the source code for it?

and afaik there is only one Krita atm

[supple raptor]

https://invent.kde.org/graphics/krita

[cyan burrow]

thanks

I wonder if anyone could make Krita into a full AI drawing software. Where they apply an AI behind any tool that can become better from it and run the server from within it

[supple raptor]

Running the server within it is pointless. No need to try to reinvent the wheel when your specialty is upholstry when you can just have it attach to APIs from existing AI backends.

[worthy delta]

It's like a fantasy angel vibe

Very nicely done

[loud falcon]

i want to fix the dress looking so short, but nothing so far

[worthy delta]

What's the prompt of the dress u have?

Is it just dress

[loud falcon]

clothes and accessories,
crystalline cloak,(detailed winter off-shoulder dress),detached sleeves,fur trim,jewelry
fur-trimmed dress, fur-trimmed sleeves,

[worthy delta]

Try adding long dress prompt in there it might work?

[loud falcon]

can try

[slate night]

Yeah long dress and flowing dress usually do the trick for me

[meager sun]

ahoy
been trying to recreate the same drawing style, yet could not get anything actually close to it
the goal is to make it look like a sketch/something hand-drawn idk
but could not get with nothing better than putting tokens like: sketch, sharpie sketch (goes really wrong, SD just adds a sharpie and a notebook to the pic), lineart and sumin else, but wont eren remember
any ideas, or without a specific Lora (which I haven't even found yet) it doesn't seem possible?

[worthy delta]

Which is sd n which is the style?

[meager sun]

both r pics from pinterest, not ai-generated

also sorry for my dyslexia 🤕
had a stroke trying to read it

i could give you a couple of full-size examples if u'd like to try to play with it by yourself

[supple raptor]

@meager sun you could pick apart the prompt from these ones that I was making last night, the model and the parts after and including the last BREAK in the prompt are what you want to copy the style.

Remove the spot color if you don't want the image to sometimes be partially colored.

PNG info is in all of those images.

[meager sun]

ok, I'll give it a try
thank u 💕

[zinc flax]

Not as rough of a sketch though

[supple raptor]

https://cdn.discordapp.com/emojis/1024970429422653540.webp?size=64&name=headpat&quality=lossless

[meager sun]

daaamn
but that's for sure way closer than absolute most of my attempts

[zinc flax]

graphite \(medium\), 1girl, monochrome, style of William Kentridge, grin, arms behind back, leaning forward, t-shirt, short sleeves, collarbone, medium shot, sketch, rough sketch

Negative Prompt: (EasyNegative:0.3)

[supple raptor]

Here's what I was using for mine, added at the end of the prompt: BREAK superb, (sketch:1.1), [monochrome::0.8], (graphite \(medium\):1.1), [anime lineart:0.2], (spot color:1.1) <lora:animeoutlineV3:0.5>

[zinc flax]

Interesting

I've never tried using the [ ] thingy though

No clue what that does lol

[meager sun]

https://www.reddit.com/r/StableDiffusionInfo/comments/ylp6ep/some_detailed_notes_on_automatic1111_prompts_as/

haven't seen there anything about BREAK tho

[supple raptor]

There's actually 2 different uses of that in there:
[monochrome::0.8] - this is removing the word "monochrome" from the prompt 80% of the way through generation.
[anime lineart:0.2] this is adding "anime lineart" to the prompt 20% of the way through the generation.

The full syntax is [from:to:when] you can also change words partway through by doing something like [monochrome:vibrant colors:0.5] which will use monochrome in the prompt for the first half of generation and switch it to vibrant colors for the last half.

[meager sun]

wow

[supple raptor]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#break-keyword

[meager sun]

https://tenor.com/view/swag-spongebob-gif-20705524

[supple raptor]

The reason I'm using BREAK before starting that section of the prompt is that it allows the style portion to have the same relative strength in the prompt regardless of how long I make the character description in the beginning. Without it, the style would get varying levels of effect depending on where it falls in the final chunk of 75 tokens.

[meager sun]

oh
got it

thank u a lot buddy

[zinc flax]

Ooo very nice. Gonna play around with it some time later

Don't have webui open, and that last gen I did was using pixai lol

[supple raptor]

It can give some really neat effects. You can also use it with embeddings which can be pretty interesting.

[gusty palm]

should break be used for prompts over 75 tokens

[worthy delta]

@supple raptor I keep seeing stuff like (worst quality 1.2) but I got no clue what it does

[gusty palm]

negative prompt to discourage bad quality

[supple raptor]

When finetuning a model, people generally intentionally include some bad quality images and tag them as such. This is important because it allows the AI to understand the concept of a bad quality image so we can use the negative prompt to tell it that we don't want something that looks like that. Most often images are processed through some type of an aesthetic scorer which assigns images to groups such as worst quality low quality, normal quality, best quality, and in some cases masterpiece as the absolute highest scoring images. Usually the worst quality and low quality images are what we want to avoid, and we want to encourage the best quality and masterpiece so we put something like (Worst Quality, Low Quality:1.2) in the negative prompt and masterpiece, best quality in the regular prompt.

Finetuning is the term used for continuing training on a model in order to "finetune" the quality and/or style of the outputs as well as it's responsiveness to prompts.

[gloomy lodge]

these are beautiful what prompt and model you use?

[supple raptor]

Aurora with a bunch of wildcards for the characters and backgrounds and BREAK superb, (sketch:1.1), [monochrome::0.8], (graphite \\(medium\\):1.1), [anime lineart:0.2], (spot color:1.1) <lora:animeoutlineV3:0.5> at the end of the prompt for the style.

[gloomy lodge]

tyty

[worthy delta]

what does the numbers mean?

like (Quality:1.2)

also i cant thank u enough for teaching me all this stuff

[brittle rampart]

how to bypass the limit of 75 tokens? if it is not possible to reduce the number of tokens

[supple raptor]

it's a way to add more or less emphasis to a word or part of the prompt. Here's the full explanation.

[worthy delta]

i see that makes sense!

[supple raptor]

If you're using WebUI, that's just a soft limit. It will increase the limit by 75 each time you exceed the current limit. The reason it's displayed is that tokens are strongest at the beginning of the block of 75 tokens and weakest at the end. Also, if a word ends up being split between two blocks of 75 tokens, it will not be understood. You can use BREAK in all caps at any point in your prompt to fill the rest of the current 75 tokens with blank tokens and force starting a new block.

[worthy delta]

my img2img wont generate

and idk why

[supple raptor]

Do you get an error in the CMD window?

[worthy delta]

i dont believe so

it isnt a ram error

u fixed that completely

txt 2 img works

[supple raptor]

The most common problem I can think of is selecting a script at the bottom left side, but not filling out whatever it needs. Make sure the "Script" drop down says None.

[brittle rampart]

so i should just use this keyword to split the string by about 75 tokens?

[worthy delta]

it does say none

[slate night]

COURBO MiSH

TTHIE OYISST

[bitter latch]

shrugs

[loud falcon]

it worked

[supple raptor]

If you're getting close to 75 tokens and the next word would be split then yes, this also applies to words that are separate but need to be related to each other such as red hair. If you were at 74 tokens then red would be token 75 and hair would be token 76 meaning they would would be in separate blocks of tokens and the AI would not have any way to understand that the word red goes with the word hair. You can also use it to sort of reset how important the next part of your prompt is since tokens closest to the beginning of the 75 token block are strongest.

[worthy delta]

ayyyeee

[bitter latch]

Correa

[loud falcon]

a bit

[worthy delta]

NOT BAD

try doing

((long dress))

and in genative

negative

do

((short dress))

also THIS

img2img still aint working

might have to

reload the page

[supple raptor]

I'd try restarting SD. Without an error, there isn't much trouble shooting I can do.

[worthy delta]

how do i just restart sd?

like close it

and re open?

[supple raptor]

yep, close the CMD window and re-open it. It would be best to also refresh the webpage afterwards to make sure that the parameters are reset to default in case that's the problem.

[worthy delta]

tbh been tryna figure out how to set the settings to the ones i want in the .json

havent figured it out

[supple raptor]

what an odd combination the wildcards landed on this time XD

[worthy delta]

LAWD DAMN

[loud falcon]

which tokens you use for more mature looking?

[brittle rampart]

word should be separated by commas?
For example :
(craving lusty look:1.3) BREAK (An abandoned factory with rusted machinery and broken windows:1.2)
or
(craving lusty look:1.3), BREAK, (An abandoned factory with rusted machinery and broken windows:1.2)

[worthy delta]

put childlike in negative and put mature in positive

or variations of adult and grown up

[supple raptor]

Я использую приложение-переводчик, так что извините за ошибки. Я бы использовал первый. Не нужно запятых вокруг ключевого слова BREAK

[bitter latch]

huh?

[worthy delta]

img to img still broke sadly

[supple raptor]

I just translated my message to russian since they asked the question in russian at first.

[bitter latch]

oh

[worthy delta]

google translate?

[supple raptor]

yeah

[brittle rampart]

thank! but I still get a warning that my string has been truncated.
Warning: too many input tokens; some (12) have been truncated:

[supple raptor]

Are you using the Cutoff extension by any chance?

[brittle rampart]

not.

i must use this is extention?

[supple raptor]

No, I thought that might be the problem because that extension only allows 75 tokens in the prompt.

@brittle rampart You are using Automatic1111's WebUI right?

[brittle rampart]

yes

[supple raptor]

Which version number do you see at the bottom of the page. Right here:

[gusty palm]

the autocomplete thingy stopped working

[worthy delta]

think the issue is my browser

[worthy delta]

mmmmm idk what the problem is for my img2img

[bitter latch]

readjusts the weights as they are to high

[supple raptor]

restarting SD should fix that

[brittle rampart]

not any veriion info

[gusty palm]

oh nvm its working in the big prompt box but not the little one

[bitter latch]

well that made a mess

[worthy delta]

are those onions in the fruit bowl

[bitter latch]

shrugs

[supple raptor]

That's ok, the commit hash that's there gives me the same information in a different way. Your version is outdated (it's from Febuary 5th) so you might want to update at some point, but it's not old enough that it wouldn't support more than 75 tokens. Are you using any extensions currently?

[brittle rampart]

only this is

[worthy delta]

imma restart the page

[worthy delta]

yup restarting fixed it

[supple raptor]

Ok, those should be fine. I'm not sure what else it could be. Try looking for this setting in the settings tab and make sure it's set to something higher than 0

If that doesn't fix it then I would probably just reinstall SD if I were you. That way you'll also be on the latest version.

[worthy delta]

omg its loading my last image i loaded

annnd that is depressing wth

[supple raptor]

Don't worry, she's fine. She's just being dramatic after she asked me to carry her the rest of the way and I refused.

[brittle rampart]

what section is this setting in?

[worthy delta]

thats a mood

[supple raptor]

Should be in the "Stable Diffusion" section unless he moved it there from somewhere else.

[worthy delta]

yall think ai art is the future of game concepts?

[bitter latch]

used aloy lora

[runic vault]

ok

[worthy delta]

only wanted a room

got a visual novel yandere scene

[bitter latch]

eek...

[supple raptor]

I think AI will continue to be integrated into other tools to make them better. Unless we have some major breakthroughs it isn't really good enough to be used as the primary tool for serious creative projects.

[worthy delta]

ngl i think theres is gonna be a future where u pay ppl to make ai art and clean it up

[bitter latch]

Fallout

[gusty palm]

ive been wanting to do something like this with 3d modeling

cleaning up 3d models or prep for 3d print

[brittle rampart]

I can't find it in any section. maybe it was added in the latest version and I will have to update or reinstall

[supple raptor]

Yeah, I would probably just reinstall since we use a newer version of pytorch now anyways. You can move your embeddings, outputs, and models folders out of the stable-diffusion-webui to keep them then just delete the webui folder, clone it again, and move those folders back into it.

[bitter latch]

"just a little bite"

[gusty palm]

when we were talking about saving vram and stuff yesterday with vae tiling etc does that under the impression of using a batch size of 1

i forgot ive been running size 2/3

switched to 1 and was able to run a bit higher resolution

[gusty palm]

actually still exceeding 100% but its not cuda OOM its just dipping into chared memory

which i dont mind i think its just a bit slower im not exactly sure

[worthy delta]

is it okay if i move my ai stuff to my ssd

will it still work?

[cyan burrow]

You will need to go into settings and change folder settings etc if they're drive specific. Same with extensions that have folder destinations. Other than that, if it's auto1111 anyway. It should work

[gusty palm]

i wish i had an ssd id imagine putting your models there would result in faster loads

[cyan burrow]

I use a SATA SSD for my ai imaging, no room on my nmv

gonna buy another one though so maybe make room on that

[worthy delta]

is there a dif tween sata n nvme

cause i got a nvme one on the cheap

[cyan burrow]

yeah, one is 100x faster

nvme that is

[cyan burrow]

mine runs at 3800 mbps and that's still barely mid range

[wind ermine]

Dieselpunk wermaght stormtrupen

[cyan burrow]

high end nvme runs above 10gbps

great, they have a street racing party nearby

[worthy delta]

not bad

also i got a problem

when i click send to img2img

it wont

but it does send to inpainy

[cyan burrow]

I never had that issue so I don't know

[gusty palm]

is there like a simple way to outpaint ,

or generated expanded borders around the image

[gloomy scroll]

[worthy delta]

@supple raptor save me nubby it keeps generating the same scene, also it wont send to img2img

[gusty palm]

thanks ill check

oop

[prime hearth]

good evening and beunvenue

[gloomy scroll]

Did you remember to check that seed is -1 and does terminal have any errors when you click send to i2i

[worthy delta]

oooooh

OOOOOH

it aint -1

[gloomy scroll]

There's many leafs here so I'll just drop this here

[worthy delta]

thank u for the help

[gloomy scroll]

-1 means that it picks random seed every time

[worthy delta]

ahh good to know

thank you

what was u guys first image

[gloomy scroll]

something that would get me banned from here

[worthy delta]

me too

LOL

istg its a universal thing

[slate night]

Mine wasn't ban-worthy thank goodness but just outright horribly ugly lmao

[slate night]

Lemme see if I can even find it

[slate night]

I didn't know what vae was lmao or how to get good faces or anything

[prime hearth]

what is a LoRA

[worthy delta]

okay so basically its just a dlc for the model

it adds on to the model and makes it better at a thing

u can use them on any model

[gloomy scroll]

Loras are files that have been trained with Style, Character, Concept & etc.

[prime hearth]

what are these models?

[bitter latch]

I used the Aloy model from that game, can't remember the name

[worthy delta]

like cetus mix

use civitai to find them

warning alot of nsfw content on there

[prime hearth]

actually, i should ask what all of this is related to

[bitter latch]

welp i'm off again.. back to work to listen to people bitch about fedex

the delivery service we use

[gloomy scroll]

[bitter latch]

they are all BLEEP

[gloomy scroll]

Remember customer is always right

[bitter latch]

nope

[worthy delta]

we allowed to swear in here

cause i been holding back LOL

[prime hearth]

what is a model?

[worthy delta]

basically the heart

[prime hearth]

of?

[worthy delta]

its what u use for a certain type of style

some trained on anime

some trained on realistic

[gloomy scroll]

#1072220168534642768

[worthy delta]

is my understanding of it

[prime hearth]

oh. this is for the ai to make art

[worthy delta]

this is a god send

[gusty palm]

any way to specify a colour pallete on generation

ive seen the pixel art options but i dont want that

[gloomy scroll]

Well you can add color names to your prompt

[prime hearth]

you can download the whole ai and tweak its accuracy using models and Loras

[bitter latch]

https://giphy.com/gifs/cat-funny-crazy-bfcGcG0ceYZva

[gloomy scroll]

Either you use some online service or if you want to do local setup you need some GUI and Model everything else is extra

[prime hearth]

ic

[gloomy scroll]

Automatic1111 webui and Comfyui are most common local GUIs

[prime hearth]

interesting

[gloomy scroll]

Comfyui is quite bit more complex for beginners

[prime hearth]

i dont think ill even begin cus i have no space left

[gusty palm]

that does sort of work but im wondering if i can input pride flags somehow like the way the pixel art stuff works but without pixelizing

although i have some apps that do this in post so its not super huge deal

[prime hearth]

time to invite all my friends here cus free ai

dont suppose the bot can be invited

[gloomy scroll]

You can try taking pride flag and throwing it into img2img

[prime hearth]

2nd one is better

[worthy delta]

i cant really see it

[prime hearth]

it has darker details

[prime hearth]

oops sorry. i meant the first one

[gloomy scroll]

If you want better face quality in these images you could look into Adetailer extension for webui

[worthy delta]

gotcha gotcha

[prime hearth]

300gb of ai generating software

how

[prime hearth]

are you a pro?

[gloomy scroll]

hobbyist who has been doing this for ~9 months

[prime hearth]

pro detected

[prime hearth]

all 300 gb of that can be replaced by 9 months of actual drawing skills

fancy ngl

ig ill stay here and download all your wallpapers

[gloomy scroll]

I guess I'll throw my entry here

[prime hearth]

are you able to make a redhead, red eyes, holy knight in heavy armour with a huge lance

[prime hearth]

may i ask you to dm me everything you have if you dont mind

[slate night]

Lmao if you want free art just go to google

[prime hearth]

wait. 9 months of work means its way too mch

[worthy delta]

ong

or do it yo self

[gloomy scroll]

Yeah not gonna happen

[prime hearth]

google also gives me stuff i dont like

[slate night]

Ok

[worthy delta]

then download the ai for yo self

[prime hearth]

data 😭 wifi space effort

[worthy delta]

its easy..

[slate night]

Yeah it's super fun and easy to pick up just knowing the basics, and there's tons of people in here who can give advice and stuff. I mean just scroll up through today's posts and you'll see what I mean

[prime hearth]

look at that

[prime hearth]

and that

[worthy delta]

ong like nubby taught me so much within an hour or two from asking

[worthy delta]

thats when u get deep into it

[gloomy scroll]

you need like one 2GB model and loras are just extra

[worthy delta]

i keep stuff minimal

[prime hearth]

its easier to ask a pro cus they already have all the nice things

[gloomy scroll]

Also <200GB of models is considered small amount

[slate night]

Yeah I'm making gens on a 6 year old laptop with 128mb vram, you can do it lol

[prime hearth]

thats more than my remaining space

[gloomy scroll]

But you need like ~5GB for Webui setup

[worthy delta]

thats with a couple of models

[prime hearth]

my space is being taken up by... thats a lot more resonable

[worthy delta]

yeah thats with bigish models

i watched a 3 min turotial of it

[prime hearth]

nice nice

do your ai hands have 5 separated fingers on each hand

[prime hearth]

and thumbs

[worthy delta]

negative hands is blessed

[prime hearth]

im not downloading 300gb tho

[worthy delta]

u dont need to..

[gloomy scroll]

I don't even know what prompt/neg I used for that image

[gloomy scroll]

You don't need to

[prime hearth]

so i can get 5 separated fingers on each hand and thumbs with 20gb of ai softawre?

[gloomy scroll]

Yeah

[prime hearth]

My investment levels have increased

[gusty palm]

those ar pefrfect

[gloomy scroll]

👋 is cheat word tbh

[gusty palm]

what about feets

[gusty palm]

mine always look like they been throu a blender

[worthy delta]

thoose feet are damn good wtf

[gloomy scroll]

I haven't even inpainted this one

[prime hearth]

bro. now i want this ai so bad

but first. may i request a picture of Yuzu Rorormori from anyone. A good one

[slate night]

sorry, I heard we were posting feet pics

[worthy delta]

its easy to download it

[worthy delta]

amazing

we need male trained models

HE LOOK LIKE A WOMAN

[slate night]

rugged and handsome are your friend

[gloomy scroll]

[worthy delta]

LAWD

WHAT MODEL

GIB RN

[slate night]

My own mix

[prime hearth]

lol

[worthy delta]

it public?

[slate night]

No, not sure if I'll release it or not tbh

BUT it does have MechaMix in there if you want to try that model out for making android/cyborg husbandos

[gusty palm]

is colour loss an effect of too many prompt tags or exceeding vram

[slate night]

https://civitai.com/models/69438

[slate night]

Do you have a VAE?

[gusty palm]

ya i used the one nubby linked

im trying a gen without exceeding vram to test

[gloomy scroll]

Our feet pics have called the master

[supple raptor]

This is what 10 months of hoarding looks like. Not all image gen stuff though. I have my text gen AI stuff in there too.

[slate night]

LOL

[worthy delta]

LOL

is nubby just the feet pick enthusiaist

[prime hearth]

wow

[worthy delta]

THE ELDEN RIGHT MORE LIKE ELDEN BED

[prime hearth]

wait. do you have yuzu roromori feet pics?

[prime hearth]

lol

[worthy delta]

bro is desperate

[gloomy scroll]

Is chibi feet pics fine?

[worthy delta]

i say not

they look child like

[prime hearth]

Yuzu Rorormori is the only female character in anime i can respect

[worthy delta]

bro

go download it for yo self

[prime hearth]

google have none

google trash

[worthy delta]

bro...

look on youtube

if google is trash we have forsaken the internet highway

[worthy delta]

NOOOOOO

LEWD

BANNED

[gloomy scroll]

[slate night]

RIP vil

[gloomy scroll]

One more before hammer hits

[prime hearth]

yt have even less yuzu roromori

[worthy delta]

I MEANT

THE AI GENERATOR

[supple raptor]

I can offer you more marisa kirisame feet pics than any one person should probably have.

[worthy delta]

nubby

u gotta teah me how to use wild card

[prime hearth]

thats good

[prime hearth]

ah. time to learn how to use the AI

[prime hearth]

20gb is a scam

[gloomy scroll]

1. Install Dynamic prompts extension
2. go to webui\extensions\sd-dynamic-prompts\wildcards
3. Create text file and add whatever you want to that file
4. add __textfile__ to you prompt
5. Profit

[prime hearth]

[gloomy scroll]

Not rly

[worthy delta]

so it rando generates words out of it

[worthy delta]

i just set the denoising too high

[young fjord]

that easy?

[gloomy scroll]

Yeah

[gloomy scroll]

It takes one line from that text file every patch

[prime hearth]

oh.

[worthy delta]

ahh gotcha gotcha

thanks

[prime hearth]

ok i belive you now

[gloomy scroll]

Ofc you can do more complex wildcard prompts but I haven't bothered to learn that

[worthy delta]

im still tryna figure out good prompts for men

rugged and handsome are blessed

[supple raptor]

here's some reimu toes too

[prime hearth]

there was a bot owner that wanted chibi art at some point and was willing to pay for it

[stable crane]

quality

[worthy delta]

god u are a foot gremlin

LOL

[supple raptor]

I made this emote https://cdn.discordapp.com/emojis/1025474924825022525.webp?size=64&name=aifeetpics&quality=lossless

[worthy delta]

yup honestly i aint surprised

[supple raptor]

that was on Waifu Diffusion 1.3, back before we had good models XD

[gloomy scroll]

@supple raptor was so in pain so he fixed these toes

[young fjord]

I want hong meiling feet pic's

[prime hearth]

now i wait to go to school to use uncapped fast wifi to download all the resources i need

[gloomy scroll]

You need to install Webui which downloads some packages that take few GBs and then after that just download some 2GB model

[worthy delta]

is all u need to do is just download webui from github n that it right

thats basically all u need cept a model?

[gloomy scroll]

Yeah

[prime hearth]

ty ty

[gloomy scroll]

Oh and VAE is also good to have

[prime hearth]

now i can fake drawing skills

[gloomy scroll]

@supple raptor You got the VAE link

[supple raptor]

https://huggingface.co/nubby/kl-f8-anime2-blessed/blob/main/WD1-4-kl-f8-anime2-bless09.ckpt

[young fjord]

a woman I can simp for

[worthy delta]

agreed

[worthy delta]

how u keep the character consistent

[prime hearth]

ty ty

[prime hearth]

cute

[young fjord]

consistent?

[worthy delta]

yeah the same generally

[worthy delta]

ADORABLE

[young fjord]

just by generating over and over again with prompts and different seeds

here's the prompt

[gloomy scroll]

Make good prompt or lora

[young fjord]

oh I can't post it

[worthy delta]

💀

we both know why LMAO

[young fjord]

we?

dude idk

[worthy delta]

oh rip

[young fjord]

ok there

it worked

[worthy delta]

ayye not bad!

[gloomy scroll]

I was expecting something like 2000 discord limit

[worthy delta]

i was expecting that too

which is the best outta these?

[supple raptor]

obviously my model https://cdn.discordapp.com/emojis/1069076345260343318.webp?size=64&name=anyasmug&quality=lossless

[young fjord]

aurora or cetusMix

[gloomy scroll]

[slate night]

I prefer cetusmix over aurora tbh

[stable crane]

you gotta get lottamix2

[gloomy scroll]

This bugged?

[stable crane]

that's what it looks like to me

[worthy delta]

same

[young fjord]

your abt the pfp?

[stable crane]

I assumed he was copy+paste URL to make mega emoji

[supple raptor]

nope, i just don't have a nitro badge.

[gloomy scroll]

[young fjord]

you know that you can just download the image right?

[supple raptor]

https://cdn.discordapp.com/emojis/1093686767808425994.gif?size=64&name=Fran_Nod_Yes&quality=lossless

this is easier though

[stable crane]

https://cdn.discordapp.com/emojis/1069076345260343318.webp?size=64&name=anyasmug&quality=lossless

[gloomy scroll]

I may be wrong but I feel like Mango got inspiration from one of these

[young fjord]

yes because I stole borrowed your prompt

[supple raptor]

China toes, as requested

https://cdn.discordapp.com/emojis/770784612116856833.webp?size=64&name=meowsalute&quality=lossless

[young fjord]

gasp

[stable crane]

that emoji is mad cute

[young fjord]

very nice toes

thank you

[supple raptor]

ffxiv servers have some great emojis

[stable crane]

oh you play?

I used to play a ton

[stable crane]

I wish they made progression more like FFXI though

[supple raptor]

https://tenor.com/view/honest-word-its-honest-work-it-aint-much-it-aint-much-but-its-honest-work-gif-13763573

[worthy delta]

https://tenor.com/view/jojos-bizarre-adventure-joseph-joestar-gif-9842584

Omg I could make it so hermit purple is actually a good stand

[supple raptor]

Yeah, I need to renew my sub, I let it lapse about 2 weeks ago.

[stable crane]

lately I just play the new expac, and wait til next expac comes out to catch up

[gusty palm]

i think it was the 'fast encoder color fix' checkbox
i just ran a gen using huge resolutions above 100% of vram and got normal colours

didnt think i actually needed that checked derp

[supple raptor]

I still have plenty of jobs to level up so I renew for a couple months every once in a while to play a bit.

[stable crane]

hm https://civitai.com/models/50290/remre0maid-onichibi-best this one doesn't work at all for me

this aint Rem

I wonder if my webui is too old

or something

[gloomy scroll]

Here's another blue haired girl

[supple raptor]

even my wildcards know that barefoot is da wae

[gloomy scroll]

Toes
Toes
Toes
Toes
Toes```

[young fjord]

feet

[stable crane]

toes, feet, dainty, this little piggy went to market

[supple raptor]

I didn't even make my wildcard lists. They mostly came from Derrian and ao汁. With Ao involved, I'm surprised I get anything other than reimu armpit and polar bears

[stable crane]

hm maybe my webui just stopped loading loras

yep that's the case

can you make custom plugins in comfyui to generate all kinds of prompts for you?

[supple raptor]

Yeah, you can make custom nodes. There's also some sort of wildcard support, I don't know exactly how it works though.

[stable crane]

I figure I might as well try it out

[stable crane]

one thing I'd like to do is a "gallery" progression

[gloomy scroll]

https://tenor.com/view/raccoon-steal-stealing-yoink-seeya-gif-10550534

[stable crane]

cute pic, cute pic, increase lewdness, increase, ..., a whole progression of poses

some LORAs still load, some don't, I guess I'm on the verge of being on obsolete tech now

[gloomy scroll]

https://tenor.com/view/nsfw-levels-rising-levels-rising-nsfw-rising-gif-22822536

[supple raptor]

https://tenor.com/view/chuunibyou-anime-bully-bonk-kumin-gif-20952854

[stable crane]

mfw I click "Generate"

[gloomy scroll]

I was supposed to watch couple episodes of anime and go sleep but I was only able to watch 1ep and got stuck here for 2 hours but oh well it's sleepy time

[young fjord]

https://tenor.com/view/kamiko-kana-kamiko-kana-vtuber-kamen-gif-25239783

[gloomy scroll]

[gritty sundial]

OOOOOOO MA GOOOD, what is your model?

[supple raptor]

Who's model?

[gritty sundial]

the stabble diffusion model

[stable crane]

lol

[gritty sundial]

this model

[supple raptor]

Oh, that's Aurora with some LoRAs. It has PNG info.

[worthy delta]

Png info?

[young fjord]

bro is someone giving birth rn?

[worthy delta]

Dayum

[supple raptor]

WebUI saves your prompts and generation parameters to the image by default. If you drag and drop your images from the folder they're saved in, into discord then that gets sent so that others can put the image into their PNG info tab to view it. Sharing prompts made easy.

[stable crane]

did webui update to avoid the discord stripping?

[supple raptor]

discord backed out of stripping metadata after like a week.

[stable crane]

oh wow

[supple raptor]

They've done that twice now and it makes no sense, either strip the metadata or don't XD

[stable crane]

ikr

it can't have been just for AI posters

the smart decision is to strip EXIF to protect your camera data

[worthy delta]

That makes alot of sense so just go to png info n boom?

[supple raptor]

Yeah, I have to imagine camera data is the reason for doing it. AI is the only reason to not want it stripped that I can think of.

[stable crane]

they could target GPS fields and such though

strange they went back on it

[supple raptor]

Yep, just have to open the image in browser or download it. If you open in browser you can just drag it from one browser window into your PNG info tab without downloading.

[supple raptor]

Yeah, the only problem with that is that it opens them up to potential liability if someone's camera encodes it strangely and their system misses it.

[stable crane]

yeah

surprised nobody made a "Send to Discord" button for webui

though not sure how much you can do with their API

[worthy delta]

I see I see, thank you!

[stable crane]

I think I'd like a "Send to Twitter" and "Send to Pixiv" or something

maybe I can build this into comfyui

[supple raptor]

I think Discord would consider that a self bot which would make it banable. You could build an actual bot to do it though. Some discord chat bots can use webui api to post images.

[stable crane]

that's annoying but perhaps understandable

[supple raptor]

Yeah, they don't want any interactions from a user account to be automated outside of the limited tools they provide.

[stable crane]

I like posting here because we can have conversations in a realtime sense

it's annoying to post to other places comparatively

[gusty palm]

thats so dang cute

[gusty palm]

what kidna tags are u using for the robotic stuff

[supple raptor]

Yeah, when discord becomes a forum where you're waiting for a reply, I lose interest in using discord.

[gusty palm]

ive always liked chats better than forums

[gusty palm]

remember icq

[worthy delta]

anime man

[stable crane]

vaguely

[gusty palm]

the internet was amazing in the 90s for a 12 year old

[stable crane]

I remember using some IRC client and spamming channels with macros

[gusty palm]

lol

[worthy delta]

OMG

[gusty palm]

i remeber figuring out how to play .wav over voice chat in counter strike

[worthy delta]

this new generation looking FIRE

[stable crane]

the new generation:
slurp slurp slurp
yum yum ice cream
slurp slurp

[gusty palm]

there used to be this chat client that , the room would change depending on what browser url u were at

[stable crane]

oh, you mean image generation

not human generation

[stable crane]

I love this idea, to chat to everyone else currently visiting that website

you know some math nerd would be hanging out on the math wikipedia pages for chat

get help with your homework

[supple raptor]

That's neat, but also sounds like a privacy nightmare.

also, how would it pick which chat room when I usually have like 30+ tabs open XD

[gusty palm]

it was called gooey

i dont know why it didnt catch on more

[stable crane]

all of them XD

[supple raptor]

emo cirno hates headpats

[worthy delta]

IT LOOKS LIKE A MASTERPIECE

[gusty palm]

is there a way to check or know which proper nouns work like touhou project or names

or do mostly everything from the autocomplete extension guaranteed work

[supple raptor]

in general, characters with over ~3k tags on danbooru will be semi reliable with some additional description of them, and over 10k tags will usually work pretty well even on their own. The number next to the tag in the autocomplete extension is the number of images that tag has on danbooru.

It'll vary from model to model though.

Pretty much the only constant is that nearly every model has no problem making Hatsune Miku and Hakurei Reimu.

[gusty palm]

;p

i need some good female zelda and other gender bent stuff

[supple raptor]

Link and Princess Zelda both work pretty well. I'm sure both have LoRAs on Civitai as well.

[gusty palm]

annnd some shy guy girls

and goombas

and somehow hello kitty

[supple raptor]

Here's my approximation of Saria from OoT

[gusty palm]

i wonder if i could create a character in koikatsu party and use screenshots as img2img

[supple raptor]

Yep, that's a pretty common workflow. Most people use Controlnet instead of img2img for that now days though. https://github.com/Mikubill/sd-webui-controlnet.git

[gusty palm]

i was worried that might be over my head or pc resources

it sounded useful from what i heard something about being able to create poses

[supple raptor]

It'll slow down your generations a bit when you use it, but with the Low VRAM option in the extension, it should be doable for you.

[gusty palm]

i forgot to use that cmd line option

[supple raptor]

I wouldn't use the cmd line option. Use --medvram instead. The controlnet has it's own low vram checkbox though that's worth using if you're vram limited.

[gusty palm]

generation took 14 minutes almost though for 4 batches size 2

oh ic

ill try medvram

--opt-split-attention --opt-sub-quad-attention

whould these be helpful

[supple raptor]

which gpu do you have?

[gusty palm]

gtx 1060 6gb

and 16gb sysram. usually about 10gb of it is free

[supple raptor]

xformers might be the best optimization for you. Don't install it with the commandline arg though.

Do you know how to manually activate your venv?

[gusty palm]

oof i did di the xformer cmd line

its not on rn but i tried it

[supple raptor]

That's ok, we can uninstall that version and install the newer one. It's just one extra command.

[gusty palm]

ok, i know a bit about the python stuff if thats what u mean

but not actually that specificaly

[supple raptor]

Ok cool, it's easy to do. Open a CMD window in your .\stable-diffusion-webui folder

easiest way is to just click the address bar while in that folder in your file explorer, type cmd and hit enter.

[gusty palm]

ok im in

didnt know that trick

D:\automatic111\webui

[supple raptor]

Once you're there you're going to type
venv\scripts\activate
which will activate your python virtual environment that's used for WebUI.

[gusty palm]

here is right?

D:\automatic111\webui>venv\scripts\activate
The system cannot find the path specified.

D:\automatic111\webui>

should i use conda instead

[supple raptor]

You'll know you did it right because the new line in CMD will say (venv) before the path like this

[gusty palm]

hmm

[supple raptor]

No, you shouldn't need to use conda. WebUI uses python's built in venv tools.

The folder that you should be in is the same one that has your webui-user.bat file that you use to launch WebUI with.

[gusty palm]

ok thats correct then

[supple raptor]

what happened when you ran venv\scripts\activate?

[gusty palm]

theres no venv dir in that folder

in the above dir theres a system folder with python in it

[supple raptor]

Ok, you must have installed from a different fork or one of the 1click installers.

Do you have a link to the repo or wherever you installed from?

[gusty palm]

i used the update.bat and run.bat from the official github repo

should i run the environment.bat i dont think i ever ran that

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

[supple raptor]

Yep, that should be it

[gusty palm]

this is how i installed

[supple raptor]

that environment.bat is what you need. Go ahead and run that.

[gusty palm]

that flashed a cmd prompt window but nothing changed

[supple raptor]

ok, open a cmd window from the folder it's in and run environment.bat

it's possible it could say something else between the parenthesis instead of "venv". As long as it has (word) CurrentFilepath it's right.

[gusty palm]

change skipvenv to 0 maybe?

( i didnt change that thats default)

[supple raptor]

oh :/ I really hope that installer doesn't install everything to your system environment or that's going to be really annoying for anything else that uses python.

[gusty palm]

would a powershell window or something worth a try

ur prob right about system

environment

[supple raptor]

Ok, I just downloaded it and it's using it's own copy of python instead of system python. So at least it isn't going to mess with everything else.

[gusty palm]

oh alright

i thought i installed python when i set up sygil

i have miniconda or anaconda and gitbash

[supple raptor]

Now I just have to figure out how to install to the environment that it's using since it isn't really a virtual environment.

[gusty palm]

im looking around and trying things

[supple raptor]

When you run WebUI, which version of torch does it show that you're running?

[gusty palm]

my python version if difrerent than yours

could i set venv dir manually

[supple raptor]

Yeah, differnt python version is fine 3.10.10 and 3.10.6 are the same other than a few bug fixes. Let's just check your xformers version before doing anything else. You're on the right torch version which was the main thing I wanted to check first because the webui.py is showing a really old version of torch in the zip I just downloaded.

Go ahead and launch your webUI with --xformers in the commandline args and check which version of xformers it's using in the same spot as torch is shown.

[gusty palm]

ok

should that skip_venv=1 remain default

[supple raptor]

Yeah leave that at default for now

I'm assuming you're going to have xformers 0.0.17 which is an older version of it.

[gusty palm]

ok relaunchn

[worthy delta]

Can I make images on my phone?

[supple raptor]

You can run SD on your computer and access it from your phone.

[worthy delta]

Teach me this eldritch magic

Cause I sometimes be lazy not to get out of bed

[supple raptor]

find your webui-user.bat file open it in a text editor and add --listen --enable-insecure-extension-access after the = in the line that says "set COMMANDLINE_ARGS="

[worthy delta]

I'll try it out later!

Boutra watch me some tulsa king

[supple raptor]

I would recommend also configuring your router so that your PC has a static local IP address because you'll need to use the local IP address to access it from your phone and it's a pain if it switches every time the IP lease needs renewed.

[supple raptor]

Oh nice! He finally made it install the right version. You can forget everything I was going to have you do. Just run your WebUI with --medvram --xformers in your commandline args.

[gusty palm]

lol xG

ok gonna do that

i used to use sygil from my phone over wifi

[supple raptor]

Yeah, I do that most nights with WebUI

[gusty palm]

just needed pc on same network and needed wan ip

does share=true need to be set or just that argument

[worthy delta]

God I love ai

[gusty palm]

and thank u so much for going through all that just now

was you trying to figure out how to update xformers or im not exactly sure what our goal was there

[supple raptor]

--share allows external access from WAN. It will give you an external gradio link that lasts for up to like 3 days

--listen allows LAN access and will work as long as you know the LAN IP of the host and have SD running on the host.

[gusty palm]

oi

[supple raptor]

Yeah, if you git pull the repo instead of using the .zip download it (to my knowledge anyways) still installs xformers v0.0.17 which has problems with torch2.0 so I was just assuming that it was going to be the same way with the zip, but apparently not or he finally fixed the pinned xformers version in one of the last couple of updates.

[gusty palm]

ooa

[supple raptor]

if you use --listen then you mostly likely also want to use --enable-insecure-extension-access which is just to allow installing and updating extensions when the listen flag is used.

[gusty palm]

good shit

[worthy delta]

Nobby is the goat istg

[supple raptor]

And if you have anyone sharing your LAN that you don't want to be able to potentially access your WebUI, you can also add --gradio-auth username:password (replacing "username" and "password" with an actual username and password combination that you want to use) which will make WebUI require you to login before you can use it. You also want to use that with the --share flag because that one creates a public link that anyone with enough determination could find and access your WebUI remotely.

[gusty palm]

thats cool

[worthy delta]

I've been wondering how to get ideas to make images

Is there some sort of prompt stuff where I get ideas?

[bitter latch]

there are more than a few

[worthy delta]

I heard abt a place where u can pay for prompts but ain't no way my cheap ass paying for that stuff

[gusty palm]

u can ask chat gpt to help u

shes touchy with nsfw stuff tho

[worthy delta]

I made a server for myself

And post alot of the stuff I make on there that I like

Shame servers like this don't allow nsfw stuff cause it kinda limits stuff but understandable tbh

[gusty palm]

anyone wanna share insta or threads

i usually get away with slightly nsfw on insta

i moved my pc into another room last week i need to call the cable company to move my mode, til then im stuck on wifi 😦

[worthy delta]

Get ethernet cable?

It's what I use

[gusty palm]

id need like a really big one

the cable line is already installed it just didnt connect when i tried

i think the isp needs to come do a thingy

[worthy delta]

All ethernet should connect

[gusty palm]

itd be on a different coaxial , but i dont see why it shouldnt work

i might just try again

[worthy delta]

If it fits in the pc ethernet port

It should work

[gusty palm]

i mean im moving my modem to connect to a different coax line in my house

i dont know why but ive had to call the isp every time i do that

[gusty palm]

I lost my Tumblr and Twitter for bullying Elon and posting nsfw

https://instagram.com/xelonxmuskxhaterx?igshid=MzNlNGNkZWQ4Mg==

[worthy delta]

good work!

[bitter latch]

watching a guy wondering around Tokyo, somewehre

[gusty palm]

more details lora pretty cool

[worthy delta]

looking good!

[bitter latch]

thats not a car lot.. wtf

[gusty palm]

that hirez fix option really does make a big difference

[bitter latch]

extra arm

[gusty palm]

it makes my renders t ake way longer tho i eneed more testing

[bitter latch]

I find it works on some but not on others.

hit or miss

[pure oyster]

I find a weight of 0.5-0.6 to be the best for that lora.

[gusty palm]

lora:more_details:0.6
is that correct for putting the weight below 1

[pure oyster]

correct

[gusty palm]

it looks weird

ok

[worthy delta]

i normally have it at 1

[gusty palm]

its making people look muscly and boobs bigger

i think

hard to identify with

[pure oyster]

Women look good in suits

[worthy delta]

anyone got model reccomendations?

like general purpose

people n enviroments?

[pure oyster]

80% of my stuff is made on https://civitai.com/models/49037, It is disgustingly good at both people and landscapes.

[gusty palm]

i love those

[pure oyster]

This is a sample https://files.catbox.moe/6acwr3.jpg

[worthy delta]

lord have mercy that looks amazing

[worthy delta]

nubby approves

@supple raptor

[supple raptor]

https://cdn.discordapp.com/emojis/1025459379287376023.webp?size=64&name=feetpics&quality=lossless

[worthy delta]

knew it

[stable crane]

https://civitai.com/models/94401/lottamix

[worthy delta]

nubby what is 1 model u would reccomend and dont say ur own even tho its goated

[stable crane]

https://civitai.com/models/64843

[supple raptor]

flatter anime style or more 2.5/3d?

[worthy delta]

i say one that can do good enviroments and people

in general

[worthy delta]

looking for enviroments

get women