🏞｜general-with-images

[dry crow]

Shark from Nod.ai supports Intel Arc, it also should be faster for that card, its still wip but you should try it

[onyx abyss]

Mixed ww cosplay w a bunch of different words and got 1 out of 100 that turned out good. Wdy guys think?

[smoky oak]

my LoRA data set is about to reach 150 images

[spice flower]

This is crazy

[smoky oak]

It is single-handedly going to take my mega LoRA to the next level, I hope

[novel ravine]

@smoky oak

[smoky oak]

looks pretty good, definitely needs more work IMO

doesn't look that Na'vi

[glossy herald]

looks gigachad na'vi

[smoky oak]

I have a feeling thats exactly what it is

[glossy herald]

yeah

[novel ravine]

because its half gigachad lora and half avtar lora

[glossy herald]

love to see gigachad like that x)

[novel ravine]

u mean just gigachad?

[smoky oak]

so now for the second time you have gone out of your way to try and undermine something I am working on

[glossy herald]

blue gigachad

I think it was a compliment though

[novel ravine]

let's exchange same prompts and see how different results we get

[glossy herald]

saying you got this, with that big dataset

[smoky oak]

A compliment how?

[glossy herald]

not sure, you were flexing on a big dataset, seemed appropriate x)

[smoky oak]

I am not sharing any of my info with you. None

[novel ravine]

just prompts

[smoky oak]

You already took an idea I wanted to work on well after I said I didn't want to say it, out of fear of somebody taking it... to which you took it

You have none of my trust, especially not after that

so yeah, I'll stick on my own, thanks

[novel ravine]

why can't we work together>?

[glossy herald]

I'm all for sharing, but I can get that people want to keep their ideas for themselves too.
Just keeping it civil and moving on to something else, or to PM maybe ?

[smoky oak]

I don't trust you enough to actually work together, rather than just steal my work/idea just like you did yesterday after you told me you wouldn't lmao

[novel ravine]

but why so much hatred

none of us own avatar

[smoky oak]

it has nothing to do with working together, or avatar. Its the fact that I specifically said I did not want to share what I was working on, out of fear of somebody taking the idea. You assured me nobody would, to which I shared the idea. And then you came back 30 minutes, having done that exact idea. Again, I have no trust for you

went out of your way to tell me you wouldn't, and then turned around and did

if that isn't shitty, then IDK what is

[novel ravine]

but how does that affect your own work?

unless you are selling loras

[smoky oak]

It doesn't. All it did was show me you are not to be trusted. Last thing I need is for you to say you wanna work together, then steal my idea and beat me to if even though I am well over 15 hours in

Its a project for myself, at my speed. I don't wanna have to deal with somebody else

[novel ravine]

I beat you huh, well . um.. thanks

[smoky oak]

congrats on the low blow man

[smoky oak]

you really got me by saying you wouldn

[novel ravine]

well

welcome to discord

you can't tell anyone to NOT do a training because it was their idea first

this is something younger audience do, in school

[smoky oak]

I didn't say not to, you said you wouldn't. AKA, you lied. So you aren't worth my trust

moving on

[novel ravine]

I didn't say anything about avatar

[smoky oak]

never said you did. This discussion is over

[novel ravine]

yeah

[smoky oak]

Please, do me a favor and do not interact with me or ping me again

[novel ravine]

ok

I can chat with other though, right?

[smoky oak]

Anyways, moving on

[novel ravine]

pandora legacy

dreamlike photoreal is an amazing model

[spice flower]

Hello guys, do you know if there is a script to automate the test of differentes parameters to generate a table like this ?

[smoky oak]

Yeah! I use it all the time

its in the bottom of A1111, in the scripts

X/Y/Z plot

[spice flower]

Nice thank you it will help me

[smoky oak]

you can find good info by googling how to use it. I use it for basic syuff

[smoky oak]

no problem. I know how much it has helped me lol

[novel ravine]

I can only tell you that this is the XYZ plot that is in the scripts tab at the bottom of txt2img section

but ashamed to say I have never used it

[glossy herald]

I love to use it too. For lots of things, comparing parameters is one but also prompt testing, or checkpoint comparison when doing training

Really useful once you get the hang of it

[smoky oak]

Yeah, I use it for all sorts of things now

[glossy herald]

If that is how you feel, you should just block the other user and not even see their messages again. Just wanted to say thanks, you voiced a disagreement and managed to keep it civil, it's a lot too rare a thing, that I need to say kudos

[bitter pike]

So we can use X Y Z plot to compare the outputs of different checkpoint but with the same prompt?

[glossy herald]

Yes !

[smoky oak]

I don't like to block people, and he hasn't done anything severe enough to justify that. I learned to not waste time on people who are clearly trying to get a rise out of you/be malicious. Thanks for the recognition <3

I just wanna have a fun time and do what I do

[smoky oak]

Yuppers

it can do allll sorts of things

[bitter pike]

Wow that's really awesome. This will save a lot of time for me when I'm checking which model gives the best image

[glossy herald]

There is a "checkpoint name" in the drop-down in the XYZ feature, you just put the names of the models you want to compare

And you still have 2 axis to play with

[smoky oak]

prompt S/R will look for a word in your prompt, then replace it with whatever you type

[bitter pike]

I see

[smoky oak]

so for example

Prompt S/R :.1,.2,.3,.4,.5,.6,.7,.8,.9,1

It will look for all .1 text, and replace it each time with the next number

or you could have it be a word, which I use

Prompt S/R: realistic, unrealistic, hyperealistic, ultra realistic, abstract realism

and so on

Like so

[bitter pike]

Wow this just became even more helpful

[smoky oak]

Same prompt, but the word changes, and so does the sampler. This allows you to see how different words affect different samplers with the same seed and prompt

[bitter pike]

Absolutely perfect

[smoky oak]

for example

[bitter pike]

[smoky oak]

you can see here, this model reacts a lot to hyper realistic

[bitter pike]

Using the same seed right?

Ah wait that's independent of the layout

[smoky oak]

yeah, everything is the same, unless you use one of the X/Y/Z axes to change something

I actually use this specifically to test my LoRA values

[bitter pike]

Can't wait to try it out once work finishes!

[smoky oak]

You can see how I used it here

1 LoRA, 2 subjects. Wanted to do a batch test to see how much they affect each other

[smoky oak]

trying out my first 2 concept dedicated LoRA

[bitter pike]

Quite impressive

So basically u can set any parameters on X Y Z

Be it checkpoint, seed, cfg scale

Btw how good is it with vram usage?

I currently use SD with 3080 so 10 gigs vram

Is it more burdensome on the gpu?

[dry crow]

It will still generate each image after the next one. So no real difference to normal batch count

[bitter pike]

Ah so it's sequential

Makes sense

That's perfect

[dry crow]

Yea also every output gets saved as 1 image before it makes the comparison collage

If you compare 3 models, you will get 3 images and the comparison

[bitter pike]

[chrome verge]

heyo, wonderful SD community ^^
i have a question and im sure it already has been answered before, but i dont know what to even search for...
basically it is about those blue-ish smudges like the one i marked in the image. SD based models seem to generate these when there is a sharp color gradient from dark top-left to bright bottom-right. i get those with many different models (original SD and tons of custom ones) and many different samplers.

based on how often i get those, i assume they are well known by now. do you have any explanation of how they come to be?

[dry crow]

Hey which model is that? You may need a vae file for it

Mostly these issues are from not using a vae file

[chrome verge]

ooh i thought vae was only for face restoration ^^

that particular model is CounterfeitV25_25, but i get those smudges with every model i use, roughly one smudge per 50 generated images or so

[dry crow]

Nope gfpgan and codeformer are for face restoration, also upscalers help

[chrome verge]

shows what i know xD

[dry crow]

Counterfeit needs a vae, let me search it quick

[dry crow]

Do you have any other anime model ?

So Counterfeit uses the vae of AnythingV3

[chrome verge]

im in that phase where i go on a download spree on civitai and try ALL THE THINGS 😄 so yeah i have many anime models lying around by now ^^
how do i know which models need that vae stuff? id like to learn, so i dont need to steal community time every time something is off 😅

[dry crow]

Okay a VAE file is used for color and Detail correction and goes on top of a model while generating. Most models have the vae already "baked in".
98% of the Anime models not.
So they need a .vae.pt file together with the model named like:
Example123.safetensor
Example123.vae.pt

Most anime models can use the same vae because there are mostly merges.

So your Counterfeit model is based on AnythingV3 and this needs a vae. So the vae will work for Counterfeit too if you rename it accordingly.
Here is the file:
https://huggingface.co/AdamOswald1/Anything-Preservation/blob/acc67d36406e41252aa936c43248c4ad988db33f/Anything-V3.0.vae.pt

You then need a restart and switch models that it applies

[chrome verge]

wow, thanks for the explanation. you are awesome ❤️

[dry crow]

No problem, you want to learn, then i explain it 😉

Most model creators also Provide a needed vae for their model or mention the one they used

[vague oasis]

im pretty sure my pet store used midjourney to make this cat, what yall think? lowkey cool

[storm badger]

Open pose + unipc varient bh2,codeformer+gfpgan+gpen,inpaint face.No extra post processing.Overall satisfied with the result.

[molten hedge]

finally trained dreambooth lol

[sonic vessel]

a rembrandt and van gogh cyberpunk collaboration

[glossy herald]

wow

wow wow even

that's incredible

everyone click that ⭐

[sonic vessel]

haha, thanks!

[glossy herald]

trully, great composition

[fleet goblet]

Thank you for the recommendation.

[fleet goblet]

I looked into Nod.ai's implementation and I can't find any documentation regarding intel arc implementation.

You certain it supports arc?

Supposedly a month ago someone in that discord mentioned they were working on it 🤷‍♂️

[fleet goblet]

Well I'll still use directML for now.

[dry crow]

You can ask on their discord, someone of the Team mentioned they have a arc770 and test with it

[fleet goblet]

I have oneAPI's base toolkit.

I've also tried the oneAPI implementation of Automatic1111's WebUI.

The problem with it unlike a normal WebUI interface is it requires extreme levels of careful prompting unlike a typical 2.1 inference where simple prompts can be used to achieve great results

The DirectML implementation also currently works, and that's what I'm using right now despite it being slower.

But it does affect it.

Especially considering I've compared the implementations, OneAPI and DirectML.

Same model.

SD 2.1 x768.

I know what model I had loaded.

I literally copypasted the safetensors version of the model between.

That's just how it be my guy

[pastel bluff]

#1084896022368624640

[fleet goblet]

Literally in the reddit post for the oneAPI implementation:

Based on my experience on A770 LE, the second implementation requires a bit of careful tunings to get good results. Aim for at least 75 positive prompts but no more than 90. For negative prompts, probably no more than 75 (?). Anything outside of these range may increase the odds of generating weird image / failure to save image at the end of inference but you are encouraged to explore the limits. As a workaround, you can repeat your prompts to get it into that range and it may somehow magically work.

https://www.reddit.com/r/IntelArc/comments/11an12q/stable_diffusion_web_ui_for_intel_arc/

If I had an answer as to why it does this I would say it.

I just don't know why.

I'm not ignoring you, but I am directly telling you that I've tested it for over 30 minutes on both implementations.

The OneAPI version is much less stable.

[cyan snow]

i made an everywhere at the end of time lora

[fleet goblet]

I don't agree with you.

[grizzled sage]

oh. we're there then. nevermind forget i said anything. You do you. Figure your own way through this maze.

[fleet goblet]

Not sure why you deleted your last comment.

Seems you took me the wrong way, clearly.

🤷‍♂️

Oh well.

I'll see my way out of this convo.

[grizzled sage]

asks for help, argues with advice lol

[fleet goblet]

I never argued with you. I literally told you I disagreed.

But instead you went the other way around and continue to enforce that what I've tested is wrong.

[grizzled sage]

bro just stop. i gave some good explanations but you disagreed lol. it's the same typical "i know better than the people i'm asking advice from" nonsense that is all over

[fleet goblet]

Never once did I say that.

Thanks for putting words into my mouth.

[grizzled sage]

no those words are all over. yours are just very similar

[fleet goblet]

Yeah? Where.

Insert Saul Goodman here

[grizzled sage]

yes i see. you're very good at stable diffusion. good job

you are the one who knocks

[fleet goblet]

Okay, now you're just being rude. It's obvious in the context now.

[grizzled sage]

😉

[fleet goblet]

Disagrees with you
Throws everything under the bus because I disagreed with what you initially said.

[grizzled sage]

i'll burn everything down if i'm a little annoyed. no lie

[fleet goblet]

No shit.

Seems a little bit more like it's the other way around.

🤷‍♂️

[grizzled sage]

good luck with your intel arc gpu!

[fleet goblet]

LMAO

Ah, paintings.

Beautiful.

[cyan snow]

is this good for my first lora? i attempted to make an Ivan Seal lora

this is the same prompt, but without the lora

[grizzled sage]

Is Ivan Seal a frost zombie?

[cyan snow]

no, he is the dude that made the abum cover for everywhere at the end of time

[cyan snow]

somewhat

[grizzled sage]

i'm not sure of the good tricks to train loras but i think you might need to tune some parameters or captioning maybe 😉 heh

OH its his art style

not a famous biker

[cyan snow]

bruv

what did you think

[grizzled sage]

i was like "that doesn't look like a guy at all"

for an art style it comes out good

[cyan snow]

with lora

without lora

with different VAE

[fleet goblet]

This is what I mean btw

In OneAPI for WSL2's implementation I get images like this half the time.

[cyan snow]

i made an everywhere at the end of time LoRA

[pseudo aspen]

@desert sandal It's this box

[cosmic lodge]

This doesn’t really fit anywhere but this the first actual physical result of my experiments with SD. A 12x12” framed print of ballerina Elsa for my daughter’s room. I’ve never been able to draw or produce any art at all, so it feels great to be able to do this.

[fervent crypt]

Got zoom_enhance working with multiple subjects, should be out later today 😉

[glossy herald]

you're the one that made that post one reddit ?

[fervent crypt]

yeah

[glossy herald]

it seemed very interesting, I didn't have the time to try it out yet

I didn't get one thing, you said it detects lots of things, thinner details to be upgraded, but in your examples you explained it did faces first and you needed to specify what to upgrade ? or did I missread ?

I need to test :p

great choice of a name btw

[fervent crypt]

There are two settings: the search query, and another for the thing you want to replace it with. Looks like this:

[native granite]

yay for handposting

[glossy herald]

that's using only the "depth" preprocessor

it was an example for @supple ravine

all hands on deck !

[native granite]

6 months ago this would have been the goat

yeah

[glossy herald]

yeah

[native granite]

Advancing ❤️ ❤️

[glossy herald]

it goes fast for sure

[native granite]

Now maybe I can finally do this

[glossy herald]

I had a blast, remaking some "crazy girlfriend "meme

[native granite]

some what?

[glossy herald]

let me find those

[supple ravine]

Things i've been able to generate.

[glossy herald]

elden cat

[supple ravine]

wow

[glossy herald]

(thanks Paint 3D for the controlnet picture there)

[native granite]

dark cat is great

Oh yeah, Paint 3D is a thing lol

who'd have known

finally an excuse to use that

[glossy herald]

#🏞｜general-with-images message

using it on "overly attached girlfriend" memes

[supple ravine]

https://tenor.com/view/persevering-face-johnny-depp-gif-4714987

[glossy herald]

that one was the best

like...

[cyan snow]

EW

[glossy herald]

thanks SD

also made that one for fun

loss

[hasty nova]

I wanna do some sd stuff I just don't know what to make

Maybe some desktop backgrounds

[glossy herald]

desktop background is great to do

[supple ravine]

make someone that seems uncertain about what to do

[hasty nova]

I always enjoy making them, I already have a few in a folder that my computer cycles through

But not enough

in #1002293361526460608 a while ago I suggested a place for sharing desktop background gens but I guess it never happened

just "desktop background" as the prompt

[cyan snow]

god cat

[hasty nova]

good image, but seems to have a few... extra limbs

very godlike

[cyan snow]

its probably an alien cat, idk man

[hasty nova]

Gods don't abide by the rules of mortals, obviously.

[cyan snow]

cool i guess

[glossy herald]

just another cat hidden behind

[hasty nova]

That makes sense

[glossy herald]

strangely fits

[cyan snow]

dude idk why it sometimes comes out with duplicates

like 5% of my iterations are conjoined twins

[hasty nova]

its just an issue with how sd works

[cyan snow]

it used to work better :/ i didn't change anything with my settings

[hasty nova]

negative prompting can help, but not always.

Are you using different models? Also higher resolutions often cause duplication, which is why hires fix exists.

[cyan snow]

what do you think i use? i use high res 1.5 upscale

[glossy herald]

well, to be fair, you are generating 1568x1024 on a model trained on 512x512

even with high res, this tends to show

[cyan snow]

merged models

one of the models is actually capable of wide images

[gilded widget]

😮 nice clip

[glossy herald]

CLIPY ! I didn't managed to make him as alive as I'd like

but yeah, controlnet didn't do better than that

trying to mix Edward Norton and Brad Pitt into fight club though...

half the face is supposed to be of each actor

@barren anvil

this is one example, there are multiple models in the controlnet extension that work very well with architecture

this is the depth model
and here is a guide on it : https://www.reddit.com/r/StableDiffusion/comments/119o71b/a1111_controlnet_extension_explained_like_youre_5/

[cyan snow]

how can you mix images? artbreeder? or there is an sd extension already?

[glossy herald]

nah, I just had a base controlnet image with both the faces stitch together badly, and used latent couple to have a different prompt in different part of the picture

[hasty nova]

not wide enough

these are some other backgrounds I have made

[glossy herald]

love the top right especially

[hasty nova]

Thanks!

[glossy herald]

a little under my ratio though :p

I'm working on getting some good ones on mine but it gets harder

[hasty nova]

I have both a ultra widescreen and a widescreen monitor and it looks fine on both

I just have it set to fill the screen

so it zooms in a little

[glossy herald]

(I see you writing LOX :p give it to us :p)

yeah I'm bored too x)

I'm going to play a little monster hunter afterwards. Still wondering what model to make out of the monsters

[barren anvil]

@glossy herald Noob questions here, up to now I've been trying to coerce various online interfaces to render image A in style B (SD generated based on a prompt, so it knows the style) and I'm not getting anywhere. But what you show is way more promising. How are you generating a depth map? Any good resources for starting with depth maps?

[glossy herald]

check the guide I sent, but yeah, i'll try something on your image.
No way I can do any bush there, there would need to be any scribble of those at least in the source image, but let me see if I get somewhere with that.
Also to answer you, the "controlnet" extension does that quite well. you just put an image and choose the "depth" preprocessor

but for architecture

mlsd seems even better

so, I'm not the most trained on that mode, I don't do architecture a lot, but here is what I can get from what you sent

I'll go less cinematic

lots better

a modern glass building, office, green grass, scenery, sunlight, highest quality, RAW photo, 8k

using the mlsd controlnet model

damn it's quite impressive even

it's a very basic prompt I used, and the rendering is also quite basic, so i'm sure with a little more tinkering you could get better

model is SD 1.5 by the way

a modern glass building, office, green grass, path,professional building, scenery, sunlight, highest quality, RAW photo, 8k
Negative prompt: reflections
Steps: 15, Sampler: UniPC, CFG scale: 7.5, Seed: 3860954534, Size: 640x512, Model hash: 4d4f85a738, Model: Base_1.x_sd_v1-5_vae, ControlNet-0 Enabled: True, ControlNet-0 Module: mlsd, ControlNet-0 Model: controlnetPreTrained_mlsdV10 [e3705cfa], ControlNet-0 Weight: 1, ControlNet-0 Guidance Start: 0, ControlNet-0 Guidance End: 1

Time taken: 10.67sTorch active/reserved: 4802/5600 MiB, Sys VRAM: 8087/24564 MiB (32.92%)

[barren anvil]

Man... Yeah this is great. It's a major use case IMO.

[glossy herald]

it should also work on interiors

would you have some sample pictures for me to play with ?

and for you to have more examples to show friends I guess :p

[barren anvil]

Yeah let's switch to dms?

[fiery lichen]

I am waiting for mlsd for 2.1 to show up eventually. There are a couple of 2.1 controlnet developers out there. no mlsd so far.

[glossy herald]

I'm not sure how people train controlnet yet. making a dataset for that seems like something out of an horror movie

[dry crow]

@glossy herald thanks for the Kind words ^^ nice that you have the mod finally back

[glossy herald]

security was key, but yeah, happy to get it back too. and thank for real, you know it and you rock still like a ball in a field, never stoping

[dry crow]

Yea everyone should be able to use SD 🙂 so im gonna support that the best way i can

[glossy herald]

that's how I started around here

when i got it running myself, I couldn't stop helping

it felt like Narnia or a magic wand or whatnot, and a right for everyone to get it up and running

[dry crow]

Yea to get the full sd experience ^^

And unlimited posibilities

[smoky oak]

day number however many its been working on the ultimate Na'vi LoRA

Having to deal with new LoRA issues I have never had before. Hmm

[sonic vessel]

Hey! If anyone is interested in checking out my new model I'm going to be hanging out in a voice channel for a bit

[mystic jetty]

anyone know this extension? for the buttons on top

[static tusk]

Settings > User interface > Quicksettings > add , sd_vae, sd_lora, sd_hypernetwork on the end

[smoky oak]

there we go

I was trying to remember how I did it lol

[mystic jetty]

@static tusk brilliant bro thank you, one more quick question for ya if you know, is there a good way i can automate different lora weights with xyz script?

[static tusk]

never tried sorry. not sure about that

[mystic jetty]

no worries ty

[static tusk]

there is add weight 1, maybe play with that

[smoky oak]

I know how to do that!

I specialize in LoRA XYZ plots

so, can you explain a little more what exactly you are trying to do?

[mystic jetty]

@smoky oak i just want to automate the weight of the lora on my prompt from .1 - 1

[smoky oak]

boom, super easy

I got you

[mystic jetty]

pagbounce

[smoky oak]

XYZ, prompt S/R

[mystic jetty]

okay

[smoky oak]

set the value of your LoRA to .1

then in the value, put this

and you're done

it looks for the .1, and it replaces it with the next every image

[mystic jetty]

oh sick, should i be careful if i have .1 anywhere else, like in the weight of a keyword, will it do that too

[smoky oak]

Yeah, it will change all .1's

so, I recommend possibly doing this

(lora:VALUE)

VALUE,.1,.2,.3,.4,.5,.6,.7,.8,.9,1

[mystic jetty]

ofc awesome

[smoky oak]

and I mean the word value

[mystic jetty]

yessir

[smoky oak]

you can make it whatever you want, just make sure the first value matches

[mystic jetty]

nice ty bro such a crucial thing to know

[smoky oak]

no worries. Very useful!

I used it to test a Hybrid LoRA I have

1 LoRA with 2 very different styles/characters in it

[mystic jetty]

looks amazing

[smoky oak]

Thanks. Though I did find they look way better on their own haha

as you can see haha

doing 2 subjects in one LoRA lowers both a bit

[gusty solstice]

Quantum Headquarters 🔥
Follow my IG https://www.instagram.com/p/CpvpqC3vpMK/?igshid=YmMyMTA2M2Y=

glitchzai

Quantum Headquarters 🪄🖼️..
.
.👈🏾👈🏾S W I P E ❗️🔥
.
.
.
.
.
.
.

Instagram

[smoky oak]

Anybody know of any GANs I can run an SD image through to get rid of the oversharpening effects?

[ripe cedar]

Idk but I remember animeganv3 online I used to use

https://huggingface.co/spaces/TachibanaYoshino/AnimeGANv3

[smoky oak]

oh, thats nothing like what I am looking for 😅

Thank you though

I am trying to keep realistic images as realistic haha

I am really trying to reverse oversharpening

it happens when you get really high resolution outputs out of SD

[fleet goblet]

Arc is working for me now on SD.

I'm good.

[smoky oak]

Whats the perf look like? I know Arc is a mess with support/drivers

[fleet goblet]

6-7IT/s on Euler A at 512x512.

With OneAPI*

[smoky oak]

which arc GPU?

[fleet goblet]

A770.

[smoky oak]

not too bad

[smoky oak]

thats close to what i get on my 3060ti

[fleet goblet]

Which is expected because the cards aren't far in terms of hardware specifications

Especially so considering intel arc's drivers are still new.

So not bad at all.

[smoky oak]

oh my bad

I get 10it/s

[fleet goblet]

Still stands.

Newer card. Newer drivers.

🤷‍♂️

None of the current intel implementations fully utilize the card either from my knowledge.

[smoky oak]

yeah, doesn't seem too bad

[fleet goblet]

10IT on what sampler?

[smoky oak]

DDIM. It goes from 9-13 it seems

trying Euler A

looks like 8-9 on Euler A

[fleet goblet]

If DDIM worked properly on OneAPI it'd be nice

But I get 7.8-8.2 with Arc on DDIM

[smoky oak]

I am using A1111, which has a considerable loss compared to other GUI's, IIRC

[fleet goblet]

despite having botched outputs

[smoky oak]

not bad

[fleet goblet]

So am I, Sytan.

[smoky oak]

Oh, I thought you were using OneAPI

oh my bad, I see

nevermind

I only use DDIM now

[fleet goblet]

I AM using OneAPI.

[smoky oak]

its better than Euler A in terms of speed, and matches in output (even better with LORA's)

[fleet goblet]

A OneAPI fork of Webui.

[smoky oak]

yeah, I realized what you meant

[fleet goblet]

Remind me to not attempt DDIM on the oneAPI version

My PC froze.

Lol

[smoky oak]

if you can get other versions, DDIM is the best from my tests

oooof

[fleet goblet]

It wasnt actually Ddim that froze me

[smoky oak]

ah

[fleet goblet]

But me attempting SDE Karras

Which doesnt even work

[smoky oak]

ahhh, I see

[fleet goblet]

Lol

[smoky oak]

I tested all of the samplers, and Euler, Euler A, and DDIM came out on top

Euler (non A) actually managed to be around 6% faster than Euler A on average

[fleet goblet]

What's weird forcme

is that I need to hit 75 tokens

Otherwise I get botched outputs lmao

[smoky oak]

oh, thats weird haha

[fleet goblet]

Simple though since you can repeat s prompt to reach said token count

[smoky oak]

DDIM gets decent results on 15 samples IMO

Damn, even 10 isn't that bad

huh...

Let me do a comparison, just a sec

[fleet goblet]

Wanna know what I get if I try DDIM?

[smoky oak]

oop-

[fleet goblet]

I can do Heun though

even though heun is known to be really damn slow

[smoky oak]

dang, DDIM is working hard right now haha

even 5 isn't that bad

10 is close enough to the others for batch work

10 batch

9.6s

so

[smoky oak]

1.2s/img

[fleet goblet]

Hmm.

[smoky oak]

I am using a realism focused model and tweaked prompting

[fleet goblet]

I'm just using 2.1.

Lol

[smoky oak]

I have been testing extreme upscaling

native (512x768)

1.5x high res fic (768x1152)

Additional 3x super upscale (2304x3556)

[smoky oak]

This is fucking cursed, and I love it lol

with his big ol

( . ) ( . )

[sonic vessel]

What is super upscale and how do I get it?

[smoky oak]

You can download it in the extensions tab

its called Ultimate SD upscale

I recommend googling how to use it

[dense tapir]

If you have a fast modern card kde karras and 20 steps for photo realism seems to be the best. Second would be euler_a 20-30 steps

this is, of course, for 2.x

[sonic vessel]

I have that, bit mine never come out that good

[smoky oak]

I recommend going to their page where they talk about how to use it

[sonic vessel]

Oh I’ve done that too

[smoky oak]

I mean, I just read and did what they say to lol

base res

upscaled

[sonic vessel]

Ah this is kinda my problem. See how the fur changed?

[smoky oak]

so let me get this straight, you want more res, with the same look?

you're not gonna find anything that does that

[gusty solstice]

2x Upscale + Color correction goes a long way 🪄🖼️

[smoky oak]

very nice :>

[gusty solstice]

Thanks 🙏🏾

[spice flower]

Your 3060Ti have more than 8GB ? I want to buy something in that price range but in RTX I dont find any except 3060

[smoky oak]

no, its 8GB, which is pretty much all you need for most stuff

I would get a 3060ti with 8GB over a 3060 with 12 any day. The 3060 may have more VRAM, but its also a way slower/weaker card

[glossy herald]

I'm trying to get some nice pics out of a logo

anyone want to play with me ?

using it as controlnet

not there yet

hard to keep the logo intact while pushing some real style into it

I can't find the right controlnet mode for it yet

normal map strangely makes it easier to get portals like that

cute one

trying mlsd

[smoky oak]

Oh man, I need my hands on Comfy UI

Nodes are my jam

[smoky oak]

Comfi UI makes me so fucking hyped for the future of SD, OMG

Now its a WAY more personal creative process

[queen lagoon]

/drean

[glossy herald]

you convinced me to try :p

seems fun

[crystal delta]

@queen lagoon

FAQ: I'm new here! How do I get started and is there a bot?

Welcome! There's currently no bot on the server to generate your images. Start by heading over to #1072220168534642768 to get yourself situated and help find the channels you are looking for! Please make sure you review our #✍🏼｜rules-and-tos and feel free to assign yourself some #👥｜roles as well! Answer any questions your may have at our #1072229020520947753. There are many ways of accessing Stable Diffusion, take a look at #1080946152318443610 to start your journey!

[glossy herald]

like, it starts in 3 seconds ? nice

ho damn it's strong

ho that's going to require some adaptation and setting up

but if it has most modules I need I'm going to switch for sure

[tame bridge]

Here's a demo of a technique for animating Stable Diffusion images now using controlnet and a custom application I made. Could be for live music visuals https://twitter.com/vibeke_udart/status/1635627814476029953?s=20

Vibeke Bertelsen (Udart) 🕹️💿 (@vibeke_udart)

Still experimenting with animating #StableDiffusion images through the use of 3D depth and light. Next step a tool for live music visuals using these techniques #AIart #AIIA #ControlNet

▶ Play video

Twitter · Mar 14, 2023, 1:03 PM

[long monolith]

How to try and crack this?
I saw it here - https://twitter.com/dannypostmaa/status/1635630263148355585
didnt even know this was possible

Danny Postma (@dannypostmaa)

Guess who can dress their synthetic models in anything now?

Hint: It's me 👋

Likes

145

Twitter · Mar 14, 2023, 1:13 PM

[glossy herald]

quite interested too, i'll follow that

[long monolith]

Yup it is...!

[glossy herald]

next week out ? we'll see

I've been teased more than once in my recent past

[long monolith]

Cldn't get you

[stoic pivot]

I often think about what folk write in their prompts and want to know what they want for result from those tags/words, the word I got stuck on now is "ultra realistic" and there is so many that swear by this prompt. So for any that use "ultra realistic", can you create an example with two images that show me the difference between normal realism and what you want a ultra realistic image to look like?

Edit: I looking what folk see as "ultra realistic" in an artistic way not is this is better or worse, just what folk see as the difference between realism and ultra realism.

[sterile kiln]

people who have test ComfyUI?

[grizzled sage]

I've had some logo remix fun in the past. It was very cool when Control net came out just a couple weeks after stadia burnt down, so i made these

i plan on having a lot more logo fun in the future

[sonic vessel]

https://seek.art/images/e694d82b409ba0029cf182fd72041e91

https://seek.art/images/4cfdab25c5026b9e7821bfc6aece5c6d

https://seek.art/images/3f557fdf52a7d4ea89bbdffe18e247fe

[sonic vessel]

https://seek.art/images/43788faab7e4598d6a41ea7ac5d200be

[gusty solstice]

CYBORG SOLDIERS 🤖🦾

https://www.instagram.com/p/Cpx6CSZvg6z/?igshid=YmMyMTA2M2Y=

glitchzai

War ready Auto Borg 🦾🔥
.
.
.
.
.
.
.
.

Instagram

[fiery lichen]

I see people in midjourney doing this sort of stuff

[wheat valley]

using Niji+StableDiff i manage to make genshin-like characters leeet's gooo

[wheat valley]

time to make some fake leaked characters 😈

[smoky oak]

I think that is actually against TOS

[smoky oak]

boddul of whine

[wheat valley]

i mean, it's not like they're going to take me seriously, chances are none of them will fall for the bait 🫠

[smoky oak]

true haha, just letting you know in case tehre are some problems is all

[wispy nest]

why is it so bright o_O

[fleet goblet]

Soap?

Cheese?

[wheat valley]

idk 🫠

[smoky oak]

Been messing with comfyUI

its super cool, but needs a lot of work/optimizations to match A1111

it doesn't seem to have nearly as good of an xformers implementation

and because of that, my resolution outputs are greatly limited, and generations are considerably slower

I wouldn't say its that much slower

but the resolutions are a huge limit

[fiery lichen]

theally made a pretty tall portrait scale demo image using comfy ui

[smoky oak]

to put it into perspective, I hit a VRAM issue in comfy when doing 8x 512x768 images

I regularly do 8x 540x1280 batches in A1111 with no issues

[fiery lichen]

ok 8x images.

[smoky oak]

8 images at the same time, yeah

[fiery lichen]

i guess theally made just 1 tall image (looks like maybe 3x normal 512)

so not pushing like going 8x

I hope there will be a colab for comfy coming out soon

[smoky oak]

I just hope comfy can get on the same level of A1111 in terms of VRAM efficiency soon

[fiery lichen]

someone's gotta try the avengers end game battle type of big scene with many custom prompted characters lol

[fervent rune]

New Locon trained on watercolor and sketch drawings

[fiery lichen]

using comfy, 1 prompt per character x 100 lol

[smoky oak]

So I would say ComfyUI is about 20% slower, and uses at least 2x the VRAM of A1111 (effective to output)

[smoky oak]

Speaking of watercolor, I made this an hour ago

original image was

10,240x4320

I downscaled to 5120x2160

[fiery lichen]

what is the selling point of lycoris? I read this and don't quite get it yet: https://github.com/KohakuBlueleaf/LyCORIS

[fervent rune]

Honestly you're asking the wrong person. Loha can't even be used in any of the UI's yet and Locon throws a key error in both Comfy and Automatic1111, but still works. I had better experience all around with LoRA but am starting to see sucess with Lycoris/Locon

[smoky oak]

I am just on normal LoRA's still as they give me what I need haha

[fiery lichen]

success in what? what is it doing better than a lora?

[fervent rune]

I like Locon as it trains more with less dim and alpha from my experience. So a Lora that I use 128/128 for, I use 64/64 for Locon and my understanding is Loha would be even lower to learn the same thing

[smoky oak]

LoRA's are still treating me well

and these results are from less than 1/20th of my full dataset

from what I know, Locon uses a shit ton of VRAM, right?

[fiery lichen]

Does 64 vs 128 mean shorter time, or more accuracy? (but that also means you get the same face every generation) or does it mean you get more variety that look like the same style?

[smoky oak]

LoHa too, right?

[fervent rune]

Yes I can't go over BS2-4 with a 3090 on Locon and no Gradient checkpointing

[smoky oak]

means you can fit more things into a single LoRA

[smoky oak]

Ah, I am limited to LoRA then

[fiery lichen]

like more characters into 1 lora like that set you made?

[smoky oak]

yes, basically

after a certain point, it can't fit new info, so understanding more using less is huge

[fervent rune]

You have to adjust Learning rate accordingly when changing dim and alpha as well. There is also a second set of Dim/alpha for locon called convolution Dim and Convolution alpha

[smoky oak]

I'll just stick to LoRA, tho my mega LoRA I am working on is looking like its gonna have to be 2 LoRA's sadly

[gusty solstice]

Cyborg Suited Female Soldiers 4x and 2x 🙏🏾🔥🎯 Follow my instagram https://instagram.com/glitchzai?igshid=YmMyMTA2M2Y=

[fiery lichen]

ok so i understand the interest in lycoris now. It was just that I keep seeing single characters on some civitai lycoris so I thought, why not use a lora or t.i.

[smoky oak]

From my understanding, there is no major reason to use anything over a LoRA in that case, as a LoRA is faster to train for the same VRAM

and LORA's are just stupid fast to begin with

[fiery lichen]

any good tutorial for style type lora? I read a basic one on reddit. not training for a character but an art style

[smoky oak]

These are from my LoRA I trained in less than 7 minutes

[smoky oak]

I don't train for style, but I am sure I could if I wanted to

[fervent rune]

I don't have tutorials written up but I only do styles if you want to add me and dm for help

[fiery lichen]

I just read that fine tuning method is good for style type lora

[fervent rune]

Finetuning is not good imo because you have to extract and it defeats the purpose of Lora, but the results are still good

[fiery lichen]

ok thanks. I am limited in some ways. I cant use bf16, only fp16 on colab

[smoky oak]

one thing to know is you should only caption/tag things you don't want the LORA to hold onto

It doesn't work how AItrpreneur says

what you are tagging are inconsistencies

if there is something consistent you want it to learn, do not tag it

[fervent rune]

Yeah the visual similarities it should catch onto from the images itself

[smoky oak]

^

[fiery lichen]

I can't DM you Devin.

[smoky oak]

now if you are training on a dark skin woman with different hair styles, tag the hair styles so it can reference later, but it will learn "oh, this is what her face looks like all the time, and those hair styles are some of the time"

[fervent rune]

Sent you a friend request, crypto servers ruined me leaving DM's open lol

[smoky oak]

you also wouldn't tag her having dark skin, otherwise it will make her ligth skin when you don't include dark skin

We really should make a collective place to share LoRA training info

info that is actually good, rather than most of the shit you read online

I think @green plover and @dense tapir would be great additions

[fiery lichen]

yeah, maybe we can suggest this to server. there is prompting help. so there should be model-train help ; t.i train help ; lora train help

[smoky oak]

I don't know about active help, more than just a location with good information

[fiery lichen]

yes they train a lot of stuff too

[smoky oak]

I work with them closely

[fiery lichen]

we can pin those important info and update as necessary

[smoky oak]

I am sharing my documentation on multiple subject LoRA's

[fiery lichen]

having an open channel for training help will be cool

[smoky oak]

like the one you saw

I noticed that it is indeed better to keep 2 characters separately trained in the LoRA when using the same rank size, otherwise they both degrade

[green plover]

waiting in anticipation to see if blip2 is going to crash

[smoky oak]

@green ploverOh, I tried comfy UI. Its cool, but A1111 is wayyy more efficient in terms of gen speed and gross res

big bummer

[green plover]

Give it time

[smoky oak]

on average, it seemed to be about 20-30% slower, and used wayyyy more VRAM

[smoky oak]

Yeah, I am hopeful for its development

[dense tapir]

I spent all day looking for a card to train on that was decently priced. Doesn't exist, and moore's law is dead on youtube just showed the 4070 is DOA as nvidia is 700+ for it.

[green plover]

And we have captions!!! 10 per picture

[smoky oak]

nice :>

[fiery lichen]

4070 is just the new 3060

[smoky oak]

Yeah IDK man. Nvidia is shitty for their VRAM cuckery

[dense tapir]

btw, comfyui actually beat auto but 25-30% for me on my 1060 gens

[green plover]

actually 11. Now I need to do the research paper thing like we did in grade school and make those into 1 caption per picture

[smoky oak]

interesting, it was considerably slower for me

[dense tapir]

I was shocked at how fast

[smoky oak]

I was getting about 7it/s on my 3060ti on comfy, and about 10 on A1111

the speed isn't much of an issue to me, but the horrible VRAM use is

errors in comfy when doing 8x 512x768 images

whereas I can do 8x 540x1280 in A1111 with 0 problems

I wonder if I can do native 1080x2560 in comfy

[dense tapir]

I don;t look at its, but I can say 56s became 35-40 using the same everything

[smoky oak]

yeah, the benefit may be much bigger for non tensor cards

[dense tapir]

the problem with the cards is I need 16gb min and the price for a decent one from last or current is insane

I bet so

no TC for me on 10x0 cards

[smoky oak]

for me on 3xxx series, it is considerably slower, but the kicker is the VRAM constraints

which means I can't play with the stuff I wanted to do sadly

[dense tapir]

well, he is anti 100% gradio bullcrap so I am sticking with him to see what he puts into it.

[smoky oak]

oh wow

comfy actually managed to pull through on a single native 2560x1080 image

how weird

but 8x 512x768 crashed-

[dense tapir]

btw, he is a good guy and actually programs unlike auto who merely copies and pastes so if his is slower for you ticket it and he will get right back, and/or fix it.

[smoky oak]

I think the issue is just that its not using the same Xformers implementation

[dense tapir]

I am leaning to buy an ARC790 as it trains now, and gens. About the level of a 3060 but with 24gb of ram

[smoky oak]

cause its slower and less VRAM efficient

[smoky oak]

I wouldn't do it yet. You're gonna have a lottt of problems

[pulsar trout]

GUYS I HAVE AP ROJECT WHERE I NEED TO SOAK A COIN IN VINEGAR AND SALT FOR 2 DAYS AND ITS DUE TOM CAN I GENERATE LE IMAGE

[smoky oak]

I forgot who it was in this server, but they were talking about how bad generating on their arc is, as almost all samplers crash

[dense tapir]

well, yeah, but all I know is I am hanging on by a thread and not even wanting to touch SD anymore.

[smoky oak]

Seriously

[dense tapir]

I am not seeing that from the recent threads

[smoky oak]

I personally wouldn't go from having one set of problems, to another set of problems, but having spent a lot of money

[pulsar trout]

yes seriously

[smoky oak]

ARC GPU's are just a very incomplete product, sadly

[smoky oak]

good luck

[dense tapir]

It will get there but the way I figure it is either them or time to wander off into the desert and see how things are with SD, and the gfx card pricing, in 2024

[pulsar trout]

ty

[dense tapir]

I can't handle colab any longer

[smoky oak]

I guess man... I just personally cannot recommend such a lack luster/problem filled product

why not just save money and go with an actual 3060? No problems

[dense tapir]

I know. I am trying to see more about it for SD. DX9 it is shite. DX11 iffy DX12 it is nice

12gb

I barely have enough on colab with 14.7 and sometimes get OOM

[smoky oak]

thats just down to how you had it set up

[dense tapir]

Naw, I do all sorts of testing and esoteric shit

16gb is my min

[smoky oak]

IDK what to tell you, you don't need that much VRAM to train LoRA's lmfao

you can do it on 6GB

you needing 14+ is just down to how you set it up

[dense tapir]

I do dreambooth, lora, loha, locon, ti, HN

[smoky oak]

alright man, it sounds like you have your mind set. For your sake, I hope it works out, but based off of the real world, its is a very risky buy

[dense tapir]

no, my mind is not set but is set against anything less than 16gb, yes.

[smoky oak]

also, where does one even buy an arc A790?

[dense tapir]

Nvidia cucked 3k series for vram and 4k is stupid priced

[smoky oak]

agreed

[dense tapir]

I can buy them on newegg and amazon

not that I want one

[smoky oak]

however, I don't think buying a drastically inferior product in teh right solution here. IDK what else to offer in terms of advice

[dense tapir]

I know on gamers nexus Intel hired the Nvidia dev guy and they were already 2 gens ahead and 3 in the test lab so I suspect soon next gens

[smoky oak]

I cannot find anythinga bout an Arc A790, only a750 and a770

[dense tapir]

Oh, 790 doesn't exist I meant 770. What I meant was a 790 would be nice

[smoky oak]

wait, but 770 doesn't have 24GB VRAM

[dense tapir]

yes, it does

[smoky oak]

no, it doesn't

it has 16

[dense tapir]

I am tired. yes, 790 would have 24 gb and why I mentioned it

that is why it is missing and would be nice

770 is the min I would consider due to 16gb

[smoky oak]

I am not trying to be rude, just trying to make sure you have all the information right is all

[smoky oak]

I see. Yeah, that GPU gens about as fast as a 3060 it seems. No idea how training speed is on it

[dense tapir]

yes, gens nice too

training I am unsure

[smoky oak]

and it wouldn't be able to benefit from xformers, so you would have 16GB VRAM, but have a much lower max res to generate at than an 8GB NVIDIA card

But I think you care more about training than high res generations

[dense tapir]

as I said 16gb 770 is my min card but no real verifiable info

[smoky oak]

yeah, i get what you mean now

there is somebody here who has a 770 and uses it for SD

[dense tapir]

I am wondering where the 790 is with 24gb? bit odd

[smoky oak]

They said that their 770 is around 3060 speed, verrified off of the it/s, but that they have to use OneAPI fork of A1111, and it crashes constantly (they also can't use DDIM, and they have to have a minimum of 75 tokens in the prompt to generate)

[dense tapir]

you know something I am wondering?

[smoky oak]

where the fuck AMD is in all of this?

[dense tapir]

besides that

[smoky oak]

no clue

[dense tapir]

I am wondering if the people with the most issues are AMD cpu users?

[smoky oak]

no idea

[dense tapir]

Intel has some of this in their core cpus now and I wonder if they stupidly used that so those who don't get bad performance or crashes

[smoky oak]

oh, its dan

@fleet goblet

are you here? Could you maybe answer some questions about your expeirence with your 770?

[dense tapir]

I

[fiery lichen]

ever tried renting a GPU on runpod instead?

[smoky oak]

thats a770m, which uses a different arch/drivers, so be careful

[dense tapir]

not for me I want local which is one reason I have become burnt out on colab

[smoky oak]

a770m is Xe based

[dense tapir]

m being mobile?

[fiery lichen]

runpod is not like colab

[smoky oak]

not Arc Alchemist based

[dense tapir]

oh

Naw, not for me renting on runpod. I know what it is and steered some people to it but not for me

[smoky oak]

A770m is the name intel uses for their beefy AF iGPU's on their mobile CPU's

see, its Xe HPG based

not Arc Alchemist microarchitecture based

[dense tapir]

I am tempted to get a P80 but it is same gen as my 1060

*P40

[smoky oak]

they use completely different drivers

[dense tapir]

Yeah

that's dumb :/

[smoky oak]

thats what lead to so many people buying 770's, cause the 770m is pretty dope, but then the 770 have massive driver issues

[dense tapir]

like 3090ti and 3090 their own set of drivers.

[smoky oak]

even worse than that

[dense tapir]

where does one find a 770m?

[smoky oak]

its legit two different architectures, different in every way.

[smoky oak]

in laptops only

intel laptops only specifically

or well, you can get them in intel iNUC's

but they are spensive

[dense tapir]

Oh, wow

https://www.youtube.com/watch?v=u8IxJQUgDDo

That thing is a beast alright

[smoky oak]

yeah, the A770m is pretty dope

[smoky oak]

thats a whole PC

thats an intel NUC, micro PC

[dense tapir]

Pretty damn tight

[smoky oak]

its one of these, but next gen

[smoky oak]

a NUC 13 I believe

[dense tapir]

That isn't a bad price

though that is a nuc 11

[smoky oak]

oh god yes it is lol

that has no GPU

[dense tapir]

ffs

[smoky oak]

and an 11th gen i7, which are very bad

(extremely hot, low perf compared to new options)

[dense tapir]

I thought it would be like a mobile gpu soldered in since it is a chip. screw that then

[smoky oak]

the a770m is the name of the iGPU you find on special 13th gen NUC devices

its the same as intel HD760 or such, just BEEFED

basically, its a super beefy APU

CPU/GPU in one chip

and there is one more thing to keep in mind@dense tapir

A770m shares RAM with the CPU

it has no VRAM, instead using system RAM just like all other iGPU's

[dense tapir]

that is shit, but howcome so fast?

[smoky oak]

a mobile 3070 is slower than a desktop 3060, thats why

and the a770 is their top of the line best offering, where as you can get a 3080ti from nividia

[dense tapir]

ahhh. Well, no hope then as I will only pay a set price which Nvidia is twice more than I will or I go to pascal server based cards so slow

bad fp16 performace which is emulated like mine

they do well with fp32 though

I was also looking at the 7900XTX

of course AMD is sorely lacking as they have nothing to compete with cuda

@smoky oak This is the only card from last gen I would want but that 12gb sucks ass. TI version is when the cuckery began.

3 fans vs 2 or some with only 1 :/

[smoky oak]

I have a 3060ti

the issue is sure, the 3060 has 12GB VRAM, but its really bad

they went for quantity over quality

[dense tapir]

rather have bad than 8gb

[smoky oak]

so its 12GB, but that GPU is often times less than half the speed of a 3060ti. As slow as 1/3rd the speed in things like rendering

[smoky oak]

in a case like this, I can understand that

[dense tapir]

for gaming ti for this AI/ML more ram is king even if slower

I saw some chart and in SD 3060 vs 3060ti training was about 1it/s slower

I am also HEAVILY looking at the A4000 RTX

I think that is my only option but that is 500 bucks. Now if the 4070ti were priced right it would be about 500-600

[green plover]

Welp! At least I didn't pay too much for car insurance

or to put that another way, can I have a button like the Staples "easy" button, except when I hit it, it says "That was stupid"

[glossy herald]

hello team

[glossy herald]

let's go back to that comfyUI tutorial 👀

for reference, this is the end of the first chapter

sure

I'm scared of where this is going

[iron jacinth]

how to do it? nidji is not free right?

[glossy herald]

working on an album for a book illustration, "broken"

that's all :p

[wheat valley]

They give you 25 free gens, but yeah, i have the subscription, first i make a character in Niji and then I use a Lora on StableDiff

[fierce crest]

NO FAIR

SHARE YOUR METHODS

HUEHUEHEUHEUEHEUEHEUE

im sorry i came off creepy

but anyways

[green plover]

Me in a bad mood? Never!

[oblique nimbus]

what artist are still working in the prompts these days?

[green plover]

Proper keywording works best, but I have a list

[oblique nimbus]

mind sharing?

[green plover]

https://drive.google.com/drive/folders/11S9AOcyaMeXHrvH99VaXbxEu3vd7Iud-?usp=sharing

[oblique nimbus]

do they all work?

[green plover]

Those are the ones in stable diffusion's vocabulary

[wispy nest]

Appreciate the lists. Just a headsup, you've got "Isaac Levitan" listed twice in the Public Domain Artists list 🙂

[green plover]

Thanks. Probably listed twice in the database

[odd coral]

hi

[wispy nest]

(in automatic1111) Is there a way to use Prompt S/R without generating the grid? I do use the grid occasionally, but usually just end up browsing the results individually in a different image viewer

[odd coral]

#1047610792226340935 hi

[crystal delta]

@odd coral

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[keen hollow]

Same Prompot in different models with same seed
Dreamshaper_331

Deliberate_V11

[calm canopy]

Deliberate

[keen hollow]

Chilloutmix

[sonic vessel]

what's the prompt?

[oblique nimbus]

are you an expert on root beer?

[sonic vessel]

got it from metadata

[calm canopy]

yes =]

[keen hollow]

its just a copy from chilloutmix prompt

[sonic vessel]

oh you've got loras in here

[oblique nimbus]

then do you know where I can find some in Florence Italy?

[sonic vessel]

you should try it why my model =]

[keen hollow]

thats what openjourney_V4 did

[calm canopy]

ive heard it's actually pretty uncommon in EU, so you may have to import it unfortunately

[oblique nimbus]

too poor for that

guess I'll have to learn to make my own

[keen hollow]

dont know your model

[sonic vessel]

https://civitai.com/models/1315/seekart-mega

[dry crow]

@calm canopy original

@calm canopy same but upsclaed with highres fix

[oblique nimbus]

I haven't had a rootbeer in a little over 17 years

[keen hollow]

what is rootbeer

[dry crow]

@calm canopy same again but now with sd upscale script

[calm canopy]

Whoah those eyes look great

Thanks for the comparison

[dry crow]

yea no problem. thats the power of upscaling 😄

[keen hollow]

its running

[keen hollow]

@sonic vessel i mean its not a women as the prompt says, but i think its still really cool

same seed as all.

[keen hollow]

@sonic vessel okay with random seed it crates a women

[sonic vessel]

ah, its alright. not awesome

[keen hollow]

i mean chilloutmix is something completely insane

https://civitai.com/models/17599/putin-the-loser lmao

no way

[sonic vessel]

I'm playing with your prompt.

[oblique nimbus]

some of my cars are coming out looking like Hotweels and that's not what I want

[keen hollow]

its not mine and i dont want credit for it

https://civitai.com/gallery/174703?modelId=6424&modelVersionId=11745&infinite=false&returnUrl=%2Fmodels%2F6424%2Fchilloutmix

its this prompt

[sonic vessel]

with the better models you can usually get away with fewer filler words "8k, best quality, etc"

https://seek.art/images/ff9b9b95f7a34a72fcadb2ccd84e2666

That's my take on that prompt.

https://images.seek.art/fddef1c5150a85449cab70c2a9189f2e.png?auto=webp&optimize=high

another one I like

[oblique nimbus]

Ide like to have that orange one in 3d

[oak plume]

RAW SDV2 no VAE no Controlnet poses, merge-models etc, ask anyone this stuff, this is difficult

[fleet goblet]

Walter white isn't human.

[smoky oak]

Heisen-burnt

[wise forge]

Hola

[smoky oak]

hounty

[wispy nest]

Interesting style that came out in one image in a batch

trying to replicate

[sonic vessel]

https://images.seek.art/77d61834f058e9c99b09d90da75b1b51.png?auto=webp&optimize=high

https://images.seek.art/ee7363ce099f8ef0e6f8e8b62a4ce85b.png?auto=webp&optimize=high

[oak bay]

what section of sd is this?

[worldly loom]

@boreal falcon

[boreal falcon]

damn xD

[ripe cedar]

What type of skills are best supplementary to AI? I think 3d software like Blender perhaps.

[wispy nest]

im working on a comic, WIP.

[cunning pond]

#1072229020520947753

[torn valley]

/chat

[crystal delta]

@torn valley

FAQ: I'm new here! How do I get started?

Welcome! Start by heading over to #1072220168534642768 to get yourself situated and help find the channels you are looking for! Please make sure you review our #✍🏼｜rules-and-tos and feel free to assign yourself some #👥｜roles as well! Answer any questions your may have at our #1072229020520947753. There are many ways of accessing Stable Diffusion, take a look at #1080946152318443610 to start your journey!

[plucky spade]

@digital tendon

https://tenor.com/view/sniff-gif-26460635

You

Are u sick

[glossy herald]

that's a strange way to post for the first time on the server

[smoky oak]

@glossy heraldOranch can't

Real orange cat moment

He's a shit head, but I love him