🏞｜general-with-images

[dense tapir]

I found it does a better job as anything I did in PS it hated

[dark flare]

KEWLS, THXXxXxXx

[dense tapir]

you are welcome

play with sliders if needed but I found I rarely have to

canny by itself I don't like but when combined with depth is great.

[oak osprey]

if you wanted to manually do it, the Controlnet 1.1 Nightly repo on github on its README on the front page describes how it all works with code examples that utilise OpenCV for generating the canny

[dense tapir]

I can't wait to finally upgrade so I can take advantage of all these things because 5-8s/it sucks.

[dark flare]

kay wait ... 2 models at the same time?

I'm, um, pretty sure I'm doing it wrong

[dense tapir]

yes, each tab

Can have as many as you, or your card can handle, want

[dark flare]

hehe yep, I did it wrong, I had "CNet is more important" from a previous test

[dark flare]

oooooooooo

All I gots is this

[dark flare]

At least I figured out the error 😆

[dense tapir]

change in settings->controlnet

[dark flare]

Great success!

[dense tapir]

I would do 4 even if not using currently

[dark flare]

How much VRAM do you have?

[dense tapir]

6gb

[dark flare]

Cool

[dark flare]

Do you use the same YAML file for all your models?

[dark flare]

This guy reminds me more of Keaton's Batman for some reason

[dark flare]

And that's the richest man in the world now, if I'm not mistaken

[dark flare]

...or maybe not

[kind quartz]

@lilac pecan some such quality?

[kind quartz]

if this quality is o.k. all info is in metadata.
point is to have not photoshoped like photos.

[sterile kiln]

https://www.youtube.com/watch?v=n82Z5-CDbKQ

[dense tapir]

I wonder how many people will leave Automatic1111 for Nvidia's webui?

[dense tapir]

A fox among the foxes

[oak osprey]

this is art nouveau style

[oak osprey]

https://images-ext-1.discordapp.net/external/bhCDkNTLEFNfdZ-zNAsNeBw8IVVfj3M3z7cX0qSU6s0/https/tripleback.net/public/discord/1685378838.116824478ffc808dae124ae4e2bc78319eb9ca.png?width=1507&height=1004

[steel cipher]

Yeah totally it's m'y favorite style

[dark flare]

Seen this?

https://twitter.com/CoffeeVectors/status/1661543152967118848

These guys HAVE TO get damn good at AI or they lose $$$$$$

CoffeeVectors (@CoffeeVectors)

Timelapse of using #photoshop’s new generative fill feature to connect two images and build a scene around them using blank prompts. Was inspired by @MatthieuGB’s post doing something similar! Notice how I’m not adding any descriptions, but letting gen fill present options for…

Likes

1146

Retweets

210

▶ Play video

Twitter · May 25, 2023, 1:22 AM

My first depth map. Grabbed a rando from my stockpile of stock images left over from my designer days...

The result ... .. . .. . . . wow actually pretty cool?!

OK wait, the head goes ... where again? 😅 not a very fair test I suppose

Slightly more "fair" depth map...

I guess if I ever need to make corporate stock images, I'll be set for life! (provided I can travel back in time 2-3 years)

The man's wider body is throwing off the proportions a little 😉

[marble wasp]

I'm still new to all of this and have a lot to learn, but I'm interested. I'm taking an EDX course right now on AI so I can get a better understanding of it, but, as far as a creative angle, anyone have an idea what kind of jobs would be open to those who do AI art?

i am loving making photorealistic portraiture, but I guess I'm having one of those impostor moments where I don't see where this is going.

and, outside of this discord, not much in the way of support for this kind of technology, which is a shame because it's fucking amazing.

[dark flare]

Apparently there's big bucks in becoming a prompt engineer. The art itself isn't worth anything, because anybody can do it, anybody can steal a prompt ... but truly understanding why and how things work and don't work ... apparently it takes a degree in computer science though?

[marble wasp]

hmmm

[dark flare]

If you could install the stuff on your machine and travel back in time 2-3 years, you'd be MEGA RICH. But today it ain't worth nothin'. Sorry 😄

[oak osprey]

i have a simple BS in CS (lol) and not a masters or anything, but it's just stumbling through it and picking up the pieces as you find them, that is how you will learn

[dark flare]

New depth map! Tourists

[worldly whale]

wow

i didnt know depth maps were a thing

[dark flare]

Even the hands look OK

[dark flare]

I got it from
https://huggingface.co/thibaud/controlnet-sd21/tree/main

control_v11p_sd21_depth.safetensors

All thanks to @dense tapir 🙏

[worldly whale]

🙏

[dark flare]

Creepy guy...

Smoldering viking, my sister would like these guys 😆

[marble wasp]

are all or most of you in computer sciences?

[dark flare]

No, I'm a general nerd, I know a little bit about everything 😉

Not enough to be "good" at anything, but enough to squeak by if I need to

[worldly whale]

did like 3 years in college then dropped out xd

[dark flare]

ooh nice! I only did 1 year of college ... totally different topic though

[boreal falcon]

Prompt engineers are a whole different thing than what people here like to be called as prompt engineers

For art itself, its easy if you speak about AI art which is true

[boreal falcon]

Jobs related to AI art? Doubt so, but since you do a course on EDx on AI you could take one of the professional AI courses with certificiate or even degree

I might do so

Right now im more on Coursera but that doesnt matter

Andrew Ng is one of those that are notoriously known for his AI knowledge and courses

Deeplearning.ai

[marble wasp]

yea, considering it. Just have no clue wtf if anything I'm aiming for, or if I should just enjoy the art as art instead of pressuring myself to make something of it. It is all fascinating and yall keep posting about things I have never heard of so I know I'm no where close to the ceiling of AI and AI art.

[boreal falcon]

with art you mean AI art i guess?

[marble wasp]

both, either

just need to chill about it

and enjoy the learning

[boreal falcon]

for non AI related art you should learn the industry standard tools ofc

Maya, 3ds Max, for VFX artists Houdini is becoming mandatory etc

Thats what im going to do

but my plan wasnt to have a full time job in this sector

But rather secondary and passive income ideally but i dont learn and do this simply for money, its a hobby

And: Dont rush it

Dive in into AI, programming, art see what fits you 😄

[oak osprey]

not how i expect a coffee robot to be but i guess it is what it is

[versed hedge]

bit random to casually chill on the beach while sitting on that thing but looks cool

[sacred marsh]

banjaran ravina logo

[wispy ether]

his reaction was insane

he said what the fuck

BRO I CANT

[neon crystal]

Bollywood flower world goddess polka slime

[wispy nest]

something is messed up with my a1111, its been unusually rendering distorted images when i prompted it for beautiful portraits

haven't changed anything with models i was using either, it's whacked up fo some reason and idk what

if the GPU was multi tasking would that affect the output ?

[smoky oak]

Oh hey, did you use my Funko Pop LoRA? Lol

[wispy nest]

nice inventive mix.

[elder locust]

I got these dots on my image (this is a small part of it)

there's not too many, but this is the worst spot

it's kinda like having a bit of dust on top of it

should I use inpaint to get rid of it, img to img, or just manually edit them out?

[oak osprey]

i think i have A1111 running on my 4090 now

[elder locust]

(a good number away from this spot are just little dots in the shadows)

[marble wasp]

idk, they look as if they have a depth of field, or bokeh, applied. Maybe in the negative prompt section add something like particles, floating particles, dust, or like floating artifacts. not sure

[elder locust]

I actually did that initially, but they came back after I went through waves of iterative refinement and upscaling on a image

[marble wasp]

hmmm

[oak osprey]

https://images-ext-1.discordapp.net/external/Y9dvoJit3lKst2ElDCeURerHCPAU3M47kUAKzIzBO8o/https/tripleback.net/public/discord/1685399210.4271786a4ea5c89fe201870c608474309986da.png?width=1481&height=987

[marble wasp]

i suppose you could try "drawing" them out before you run new renders

[elder locust]

that's what I was thinking of

either editing it out or using inpaint, which you think would be best?

I mean I've been working on training a model and generating this image for 5 days so learning quite a bit about iterative refinement

[marble wasp]

suppose either can work dependent on what you're more comfortable with. but i know nothing of training models, only touching up photographs, so bear that in mind

[elder locust]

alright, I'll just edit it out

with art software

most places just need a black dot to fill the void

yeah I can easily zoom in and eye dropper the surroundings

[dense tapir]

No, but I have seen it in action.

[dense tapir]

Depth, Canny, Reference...done

[sterile temple]

need to speak to the mayor about getting storm drains

[oak osprey]

@smoky oak whatever reason you can't run my model in A1111, isn't because of my model

it's your config 🙂

same for you @dense tapir

[smoky oak]

Peeng

[oak osprey]

probably some LoRA or TI or TE or some other thing i don't know anything about is loaded when it shouldn't be

and in #🤝｜tech-support one of the community guides told me you can swap the VAE for 2.1 so make sure you're using MSE 840k on it

does this thing make the image jump around for anyone else?

when it's previewing it shows the prev image and makes the new one all blurry and stuff before it shifts over top again

made with A1111 and my v2

[dense tapir]

@smoky oakhttps://www.youtube.com/watch?v=IlWT_TdOK6s Oh, my. Jensen is actually defensive of their GPU pricing and is enamored of Apple. As I said a few weeks back Jensen is going to go the Apple model of doing business which sucks for all of us.

[wispy nest]

Was hoping it came out of its pp

[oak osprey]

😐

yes

[wispy nest]

there we go, I kinda like the style :D

[wispy nest]

as long as one doesn't think too much about the limbs, then it's not a problem ;P

[oak osprey]

what model is that? the output is very dark

or, rather, high contrast

[gentle orchid]

I connected AI outputs to Unet directly via Reference controlnet. This is so messed up

[oak osprey]

in Comfy? lmao

[gentle orchid]

WebUI, reference controlnet model, so no prompts have been involved

[wispy nest]

it was the one Sytan showed me, but I'm not sure if it's based on the mode, this is the same prompts and stuff but on another model :P

[gentle orchid]

After attenuation of neural artifacts reached some critical value, all generations turned into deserts/bushes/gore etc

[static tusk]

remember when I said it needed VAE to work 😆

Im downloading it now

[oak osprey]

i haven't gotten to test this new one much tbh, but i know it works best when you have the right sampler

DDPMScheduler in Diffusers but i'm not sure what the name is in Auto

DPM Adaptive looks alright but a lil weird

DPM++ 2S a looks close.. weird face on the kiddo tho

[boreal falcon]

they already are

[oak osprey]

oh

the v1 model file still doesn't work on my a1111 but i've recreated it and now it does

one made NaN errors an the other doesn't

[wispy nest]

how were you able to get a photo of me in the morning when I'm readying for work? 😱

[sterile temple]

https://tenor.com/view/stalking-watching-you-gif-25961549

I'm always watching

I could use another cup of coffee right about now

[wispy nest]

https://tenor.com/view/angry-cat-instant-gif-24474761
But yeah, now I need to brew some more coffee! for no reason at all :P

[smoky oak]

whats up butt nuggets?

did I miss anything?

probably 3 major extension releases, a whole new AI startup, and 6 more failed NVIDIA launches lol

[sterile temple]

enters butt nuggets into prompt

enters nsfw into negative prompt...

[dark flare]

mmmm, butt nuggeeets...
/homer

[static tusk]

PNY and Cooler Master have teamed up for a new RTX 4090 design, called VERTO ARGB Dual Fan (which is actually a very short name for a PNY product). This graphics card features two massive 12cm fans developed by Cooler Master.
4.5 slot cooler 😲

[smoky oak]

I did actually see that

I also saw a new solid state cooled PC I think

[static tusk]

and an Asus 4070 with no power cables

[smoky oak]

yeah, that too

[oak osprey]

i figured out how to make my model work in a1111

[smoky oak]

oh, sick!

[oak osprey]

you simply load it

pure_fire says he has to force the 840k VAE on it? for some reason. which is odd because that's not 2.1's default VAE

i thought it was. so did one of the community guides. but he found out, it's not. and he showed me the results of switching VAEs

voila

[static tusk]

yeah if you leave vae on automatic you get NaNs errors

[oak osprey]

mine is on automatic

[smoky oak]

Alright, so I can't use it, cool, not gonna waste my time with it lol

[oak osprey]

no errors

well in your stack trace it shows you loaded Loras too

[smoky oak]

but I didn't

I don't have any LoRA's active

[oak osprey]

hmm well they did say that pytorch version you're on and the xformers, have issues with this

and if you can't upgrade, oh well - but maybe you can run a seperate install for pt2

[smoky oak]

alright, then I won't bother with it

thank you for the info, I will keep that in mind for the future

[oak osprey]

oh

also i did get the NaN error on the v1 model file, when i tried loading the one from civitai

i had to rebuild that file so it loads

the v2 model i made today worked correctly the first time i tried it, so that was hella confusing

also didn't need a yaml file, it seems to understand it's teh 2.1-v model

but i also did discover how to make that yaml and so that's also on civit now

[smoky oak]

I'll give that a try, if it works then cool, if not then I am not gonna bother with your model anymore

[oak osprey]

yeah i'm just happy to know it works for anyone on the latest version

[static tusk]

it works. the model needs more coherency though

I posted a few sample tests #1045349359044280360 message

[oak osprey]

was that on the v1 or 2

[static tusk]

v2 from civitai

[oak osprey]

yeah it's more heavily trained than v1, try that if you have the time

[smoky oak]

@oak ospreyCan I have a link to the page?

found it

[sterile temple]

name of model?

[static tusk]

https://civitai.com/models/73341?modelVersionId=78058

[smoky oak]

@oak ospreyModel works for me

but it does not look good at all

[oak osprey]

don't use DDIM

it's using terminal SNR, so

you need to figure out which scheduler A1111 has that works with terminal SNR that you like the results of

like here's an example of what i can get out of it

[smoky oak]

didn;t use DDIM

[oak osprey]

the thing is i don't think A1111 has a correct implementation of DPMScheduler

i don't know why everything looks terrible through every sampler it has in that tool

[smoky oak]

The images I am getting look like signature 2.1 images TBH

[oak osprey]

probably because they're using k-diffusion which is incorrectly implemented too, there's some open bugs on their tracker

idk all i can tell you is i was thoroughly unimpressed with stable-diffusion-webui

my image pipeline with a single python script works much better

[static tusk]

if that were the case noone would have good image results

[smoky oak]

right lol

I feel like he is always making excuses

[oak osprey]

it doesn't have to be absolutely incorrect in order to be incorrect. things can be off by a little bit

[oak osprey]

don't try and start flinging shit when i'm the one looking at the source code and actually making it work

you did nothing to make it work, just whined and said you'll delete the model and you're done with it etc, as if it matters

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/2940

this is a partial list of all the missing or broken stuff in sd-webui

check how many broken k-diffusion things there are/have been. and there's likely more, because sd-webui users just seem to care about furries and NSFW and think "2.1 is just bad" because the tool is broken

and i don't think SDXL is going to work very well in there unless they fix all of that because it uses terminal SNR too

[static tusk]

I havent used 1.5 for months. only 2.1 and get good results

[smoky oak]

I wasn't the one that was doing things that needed fixing lmao

[smoky oak]

Alsom, your looking at the source code hasn't been the most reliable, Mr. Commas do nothing lol

[oak osprey]

i also pointed you to the bug that they fixed 11 days ago on that

but you haven't apologised and still do this petty shit

[smoky oak]

my A1111 install is over 2 months old, and yet it still matters

[oak osprey]

well it's more like 13 days ago now

i'm not going to go git-bisect when they broke it

you seem to think your singular experience is the only way it can go

[smoky oak]

2 months old =/ within the last 13 days, therefore I do not get the update, and therefore it still made a difference back then lol

[oak osprey]

did you ever think they'd broken it a month ago? i haven't been looking at its source code for longer than 3 weeks

it's just disappointing, how annoying you can be

[smoky oak]

no, but did you? You just up and said it makes no difference, without checking over versions, so I mean

[oak osprey]

why are you like this?

[static tusk]

these are from aetherLux sample tests I did #1045349359044280360 message

[smoky oak]

@oak osprey I will wait for you to flip into one of your other various moods/personalities before I continue this conversation

[oak osprey]

you can talk to yourself or someone else about it now, i'm done with that lol i got it working for others, that's all i care about

[smoky oak]

I'll just wait until your mood flips again before we continue talking. See you then

[smoky oak]

AetherLux does seem to be a pretty solid 2.1 model as far as all of the other ones I have tested

probably the best

[static tusk]

its on par or better than realism engine and Illuminati for sure

realism engine has more depth in subjects though

they probably had more images

[smoky oak]

it still has 2.1 issues for sure, but they are the best I have seen thusfar

[oak osprey]

no, lol, because you will just keep a tally list of all the things you project onto me and then lash out when you disagree with a technical assertion i'm making

like i give supporting documentation and then i dunno what to call it but you just decide, "oh, that's just you being stupid, you're always stupid"

"here's everything stupid you do all the time to prove how stupid i think you are"

[smoky oak]

I literally have never called you stupid, or insinuated that you are stupid. Please, grow up dude

[oak osprey]

you don't do it directly, duh. that's against the rules of this server. you tell me through your words implicitly when you say that i'm not able to read source code. you know what that sounds like to me? jealousy, bro

i know between the two of us who can or can't read or write source code

[smoky oak]

Oh come on dude. My stuff works, I do what I wanna do, I am not jealous of you. You are being a little ridiculous, especially when i have also beem helping with things in DM's

I do appreciate the things you have helped me with, but this kind of childish behavior is legit sad. If you really feel like me saying that you having problems that nobody else has ever had (because you say that you don't wanna use the tools that everybody else uses, but then get annoyed when your products don't work with the tools we make) is me insinuating that you are stupid, when you are clearly not, I don't know what to tell you.

I swear you have bi polar tendencies, jumping from being nice and being an ass, and I am not the only one who has seen it. I tend to just stay away from you when you get in these moods, but now you are targeting me even more, so no thanks. Keep my name out of your mouth, respectfully, until you can learn how to have a conversation without making shit up about people. Thanks

I do appreciate what you have done for me, and the community, but I will not put up with this anymore

[calm canopy]

speaking of DMs

😏

[smoky oak]

lmao

[wispy nest]

people are result driven, well some are. But there's different kinds if results. Here I'd say there's "artistic" and code. How something work can be someone's good result while style might be someone elses :)

[smoky oak]

I also agree

resorting to petty namecalling and making stuff up about people is not how I roll. I am just trying to make some hot guys and LoRA's lol

[wispy nest]

And my knowledge of python script is really bad so I mostly want good images than code :P

but everyone likes different things and I'm just happy that many types can find it fun when it comes to ai stuff :D

[smoky oak]

yeah of course, I am glad you see ti like that :p

I am just here for fun, and the betetrment of my knowledge

[oak osprey]

the best part about knowing how to code is that if there's some limitation in an open source tool, it's not there for long. or you just write your own from the start, and it fulfils all of your needs, because you made it for you.

[wispy nest]

yeah, coding something and then using it is something special :D

[smoky oak]

I am still a beginner at python, but I do have some goals for things to make for myself

[oak osprey]

i could sit there and use a webui and just generate images all day but it's not as fulfilling as being able to learn and develop understandings of what's happening underneath

[smoky oak]

really just automation scripts, but I do wanna get TensorRT working ASAP so I can speed up my workflow a lot

[oak osprey]

was looking at khoya's code the other day and it's actually really nice to learn from

[wispy nest]

I'm more of an artist when it comes to webpages, I even got this! ;P

[smoky oak]

lmaooo, I love that haha

[oak osprey]

that's awwwesome

[smoky oak]

I do 3D printing a lot as well lol

[oak osprey]

is that 3d printed

[smoky oak]

Oh man, just reminded me that my free trial of Plasticity is probably over

[sterile temple]

I spend all day writing python at work, and I'm lazy. So I appreciate all the effort the community puts into make it easy to play around with SD and try all sorts of settings and what not without having to write code

[wispy nest]

yeah, but mine is just that image taken from a site a long time ago, I'm not rich enough to have a 3d printer :(

[smoky oak]

oh come onnn, 3D printers are cheap now :p

You can get stellar prints off a $200 printer now

[wispy nest]

I got too many bad decisions to pay off first ;P

[smoky oak]

hell, my printer was $160 when I bought it, and I upgraded it myself, and I can print better than some $2,000 printers on it cause I am using custom firmware

[smoky oak]

honestly, valid lmao

that reminds me, I need to get to 3D printig

[wispy nest]

and I also like to play computer games when I don't have a - sign before the number on my bank account :D

[wispy nest]

and when I got time I also try and find new cool things with the art ai, like this is from the model you showed me
thanks again by the way :D

[smoky oak]

Of course! Glad you like it :>

[oak osprey]

i started out just making a discord bot to entertain friends with

i want to add a guess the prompt game to it, like pictionary

[sterile temple]

like #1103141787960344576 ?

[oak osprey]

oh, that exists? 😅

that seems to be just counting the frequency of overlap with the prompt words but i'd check for their locality too using NLTK or something

or just check the indexes and see if their neighbours' indexes are the correct terms 😄

[sterile temple]

fuzzywuzzy

I use fuzzywuzzy at work for account name matching

[sterile temple]

Is it just me, or does setting the resize from seed at a higher resolution than the image resolution result in more coherent images with sharper details?

[oak osprey]

it hallucinates them

some of the GANs do that too

https://images-ext-1.discordapp.net/external/trA3ZyySrkV50w8IJevWqjO2pk1UzJI9bY__0oAFgBY/https/tripleback.net/public/discord/1685420895.5423048af5fde9147cd4193e5d636c5c264f94a.png?width=1267&height=844

lol i love how SD biases to cars if you just prompt testing

teamwork

[marble wasp]

Back again to where I started with AI, Scarlett Johansson

[oak osprey]

3.0 vs 7.5 cfg

radioactive wasteland

[wispy nest]

CFG is one of my weaknesses, I have no idea at all how it work because all I understand is that it just make everything stronger in a bad way when going over 5-6 :P

[oak osprey]

for v-prediction models it does better with higher values

i use around 9.2 for mine usually

the 2.1-v model uses a totally different loss prediction than 2.1-base or 1.5, and so it needs higher cfg and a few other tricks to make the output actually look good

[static tusk]

for most 2.1 models you want a lower cfg than you typically use in 1.5

but it depends on the model

dont go lower than 4

[oak osprey]

i'm doing inference in 32bit as well btw

[static tusk]

that changes things again

[wispy nest]

I go between 4.00 to 7.55 but it's mostly a "I have no idea what I'm doing" than skill :P

[oak osprey]

well 2.1 has numeric instability in half precision. it's not good to do that in half mode

can be pretty drastic

[static tusk]

just look at the output images. If they are cooked and oversaturated your cfg is too high

[oak osprey]

for v-prediction it looks cooked if it's too low

and also loses prompt coherence

eg. it is less likely to do what you ask it

[wispy nest]

yeah, that's why I stay within those numbers, anything higher "overcooks" them in my opinion, and lower messes it up as the ai think the prompt at that strength(?) too weak?

[oak osprey]

low guidance value

[static tusk]

yep

[oak osprey]

impressive mountain reflection tho

[marble wasp]

as fun as this is, I think I need to elevate my work and finally try out a1111 like I have been putting off. Idk what the CFG stuff is and I still am working with the Easy Diffusion UI. yet, It's hard to make an ugly Scarjo image 😉

I mean...

[oak osprey]

https://images-ext-1.discordapp.net/external/HYqFJT31iT36JrGtcV7d0Iz8GNOGycAD6LkMdzRsslg/https/tripleback.net/public/discord/1685424423.863274ff7a0ca3be9e660963d8509eab0fc685.png?width=1267&height=844

https://images-ext-1.discordapp.net/external/oWO2_rxQneMauaXOZA_Xsl_PP-gyRs4YyIpH92fGBNo/https/tripleback.net/public/discord/1685424505.491260591c329b549f9455c2821e32517ca2411.png?width=1267&height=844

[pulsar night]

#🏞｜general-with-images software development, Promotional image, cinematic lighting, UHD

[oak osprey]

clocks are hard lmfao

[sterile temple]

can it do a digital clock?

[oak osprey]

ok so salvador dali gets a working clock?

this is irony at its finest

i also quite like that img

https://images-ext-2.discordapp.net/external/qQOgkBiIvXgc6H2F4xqcgDmgXD20kTPbGtqqMpLCvyo/https/tripleback.net/public/discord/1685425219.206722528e053a6be27701f0ba8f0b586117cb5.png?width=1267&height=844

i should try giving it a time to display

https://images-ext-1.discordapp.net/external/5Stwb_dRZNd8aWIRjheKuKgem7aaceWwozCAnTn0erY/https/tripleback.net/public/discord/1685425292.96518165d0e7e665caf988094143336943349e1.png?width=1267&height=844

nay

#🤝｜tech-support

[marble wasp]

i really need to advance and start using controlnet.

[gaunt heron]

This is what clocks and screens in dreams look like

[sterile temple]

I tried the tensor rt thing, but it was 40% slower for me

[dense tapir]

which card? Tensor cores are the fastest of any GPU part. They are so fast, in fact, that even on the fastest GPU/Accelerated cards wait 50% of the time on the mem inside the GPU/Accelerator to hand them more information because it finished the job so quickly.

[sterile temple]

3060 6GB

[dense tapir]

That would be why

My 1060 doesn't even have any TCs 😦

they didn't come until Turing and I am Pascal based

at least your mobile has 120 TCs but at only 6gb you are mainly having issues with getting the data to the chip

If you had 8, or 10gb of vram, real vram, you would see a HUGE difference with tensor cores for this

know what is funny? Your 3060 mobile is basically a slightly revved up 1060. Sort of like a regular 3060 12gb had sex with my 1060 and had a baby.

[sterile temple]

hmm, oh well. I've still managed to generate well over 100k images in the last 6 months. So I think I'll manage with my laptop for now 😄

at least my laptop has the 115W version and not the 80W version

[dense tapir]

I am honest I never heard of your chip but I despise laptops since the early to mid 90s. Too much is soldered on with little, to no, ability to upgrade I just swore them off. I like freedom to replace parts not buy and stuck.

[sterile temple]

I was into building my own PC years ago, but now I like being able to sit in my lazy boy chair with my laptop 😄

[dense tapir]

No thanks. The day I can no longer build my own is the last day I use the PC and go out and do gardening, fish, etc...

[oak osprey]

mine has 150w but hits 115w when cpu boosts

love terminal snr

[timber frigate]

Ah yes, API-only closed source is the best part, my favourite

It's a feature that you can't run it locally, trust

Not like every locally runnable tech has a super user friendly way of using it made by the community within 2 days...

[oak osprey]

the part you circled is demonstrating their target audience is total idiots

maybe its nothing special and releasing it would just look dumb

the image demos are mostly bedrooms

easy stuff. it fails to keep the bedroom layout the same and changes the number of windows etc

[boreal falcon]

whats the point of Reimagine?

i saw some stuff but i dont get it quite yet

[oak osprey]

its just img2img

[boreal falcon]

so nothing special?

what sets it apart

[oak osprey]

some technical details

implementation details in other words

the heroes embark

to discover a sentient rock that manipulates time

[boreal falcon]

i mean okay they want it to make it easy for people instead of having to work around with some technical stuff

[oak osprey]

yeah thats the future and where the revenue is

you guys act like they can just do everything for free

lol

[boreal falcon]

i dont expect anything to be free

it comes down to people what they want ofc

[oak osprey]

not really specifically talking about you

its just how the community is

[boreal falcon]

oh

well many do yeah

and are almost triggered when they hear someone pays money for stuff

at least some

[oak osprey]

you will never convince me otherwise because ive seen it relentlessly and the shit stability goes thru im surprised they even plan on new open sourcr releases at all

it is less liability for them to not release weights

only a matter of time before they decide to do api only and no one here is going to train foundation models. theyre always whining about how hard and costly that is

deepfloyds license is so restrictive as an example of what we get to look forward to

[boreal falcon]

you mean they are going to charge money sooner or later?

OpenAI went from non-profit to for-profit after all too

[oak osprey]

they can charge and be non profit

[boreal falcon]

which they arent, i mean non-profit

but that is not an issue for me. Non profit or for profit

i can live with both lol

[oak osprey]

same

we already got enough from sai

they can close source and be paid only and i still would consider them generous

can't undo what they've done

[boreal falcon]

wont please a good part of the community i guess tho

reminds me somehow of Blender community with some situations

[oak osprey]

look how novelai was treated

a1111 stole their code and faced no criminal charges

community backs him

they just want it all free no matter the cost

[boreal falcon]

i mean no wonder, its also how they treat artists amongst all

i said people are using the cutthroat approach and mentality for a reason

which is also why i could spit on them every time they mention immorality of corporations for example

[hard wagon]

@wispy spindle Thanks for the pow win! Just wanted to share that the image was made with my new model Aether Lux, alongside the prompt.

ultra detailed marble and jade sculpture of a female necromancer, skeleton face, volumetric fog, Hyperrealism, breathtaking, ultra realistic, ultra detailed, cyber background, cinematic lighting, highly detailed, breathtaking, photography, stunning environment, wide-angle

[boreal falcon]

the only difference between the apparent or actualy cutthroating corporation and those users is that those users are on the bottom and the others at the top

[hard wagon]

Details on settings can be found here:

https://civitai.com/models/76276/aether-lux

[wispy nest]

I have an issue while installing sd on my laptop, how do i solve this?

[oak osprey]

[boreal falcon]

@oak osprey what is you workflow if i may ask? 😄

[kind quartz]

@wispy nest what is your gpu on laptop?

[wispy nest]

Could i dm?

[oak osprey]

my discord bot code on Github, https://github.com/bghira/discord-tron-master and discord-tron-client, it is a master-slave worker network that can utilise many remote workers to distribute jobs in a queue

trying to get the best default output settings for as wide a use case as possible

this is one of the ways you can still do SD on Colab though via free tier.

[kind quartz]

@wispy nest ask in #🤝｜tech-support i have not dm allowed

[wispy nest]

Oky

[boreal falcon]

oh, you work via Discord on your stuff? O_O

[oak osprey]

yep

there's nothing to really work on, the output is all great because the pipeline is hardcoded to a specific set of parameters, and the ones that can be adjusted are limited to ranges and values that have been tested to work well

most of the time i just allow GPT to generate the images

interesting prompt you got there @hard wagon

i should try that negative embed on this model

[hard wagon]

I found it on Lexica 🙂

[oak osprey]

your model is so dark! it's incredible

https://images-ext-2.discordapp.net/external/jNI_HD6JD0cBYlMtJRltRyAGvRNcGXuK6hV8FOcNYpY/https/tripleback.net/public/discord/1685426783.2596364351c656c217794cc72dcd6c3c0421211.png?width=1481&height=987

i am curious what it'll do with this prompt

Explore a rainbow-colored museum filled with floating jellybeans++

gotta love the deformed output of 1920x1080

this art installation would sell for like $760,000

[boreal falcon]

@oak ospreyinteresting, i see this one for the first time 😄

the one time i used generative AI on Discord was with Midjourney

[oak osprey]

i used Blue Willow which led me to making my own bot because their model selection is weird and reproducing a single result is too hard

i've never used Midjourney

[boreal falcon]

oh, i got some invites to Blue Willow, never touched it

i started with Midjourney back in july 2022 and ended the subscription in december, used Stable Diffusion/Dreamstudio on the side and then left it all to focus on my artistic and editing skillset and doing manual stuff and waited for Adobe to release their own one

meanwhile there are also those like Leonardi.ai and others but i never touched them

Leonardo.ai*

[oak osprey]

blue willow is free and you can invite it to a server any time

[oak osprey]

i guess the green tint from your output is the "jade" keyword, but mine seems to focus a lot more on it

[boreal falcon]

dunno, i heard its free but there is a system behind it

forgot how it worked

[oak osprey]

it's just free and there's no limits

you can submit 5 generation requests at a time per person

it returns about 20 images in idk, 10 seconds?

whatever they've done, they've done it really well and it scales

[boreal falcon]

oh okay

[oak osprey]

it's running on more than 20 million Discord servers

kind of like, i don't even know how to comprehend the scale of that

who is funding that

[boreal falcon]

i have found and pretty much built up a workflow for myself but its interesting to see others their workflow etc.

[hard wagon]

Yes. It’s jade. Try other other gemstones. See what happens 🙂

[oak osprey]

i should add offset noise alongside the terminal SNR for v3 of this

i really love how i can get images dark when i add appropriate keywords for it, just with terminal SNR alone. but there's some shocking contrast i see on the offset noise models, though i'm concerned it's not possible to 'turn that off' on those models?

[hard wagon]

No, you can’t turn off offset noise. You train it in. You can make neg embeddings though to intensify brightness all over, but that’s not really changing anything for real.

[hard wagon]

These are made with my model by a friend, @analog goblet using his neg embedding to focus on dark imagery.

I could’ve trained in more offset noise than what’s there already. If I really wanted to enhance that.

[oak osprey]

neg embeds = textual inversion?

i've not looked into either a whole lot

[hard wagon]

Yes 🙂 but surely they can be made as LoRA’s too, but that I have no experience of.

[oak osprey]

interesting, looks quite simple to implement

that doesn't happen here

[wispy nest]

Where to do it bro

[oak osprey]

#1100170312106127410

[wispy nest]

Ok thanks bro

[dark flare]

😆

[oak osprey]

anyone ever noticed the impact arafed and araffed have on the images made?

@neon crystal new ability unlocked

[wispy nest]

I didn't even know it is a word

[oak osprey]

it's not

it's just in the tokenizer vocabulary and quite prevalently so, and i'm not sure why

https://github.com/damian0815/compel/issues/34

more prompt parsing oddness

[versed star]

Hey guys, anyone knows if Stability AI is planning to enter the stock market?

[oak osprey]

why do all these bots keep coming in to ask that

[tall granite]

Hey guys, I just begin to use stable diffusion and realized It uses RAM instead of VRAM. Isn't that strange?

[woven dawn]

The models load into ram and vram, but run in realtime off thr vram.

[oak osprey]

ah crap comfy put a security hole into the tool today

[oak osprey]

guess you need to buy the rest now.

[dry crow]

7900XTX?

[dense tapir]

I think it is as it is an XFX brand. The picture shows how bad it is where even 170mm is not enough due to the power connector so 180mm is really the minimum safely. A lot of people removed their side panel just to use Nvidia cards.

[oak osprey]

that's the 7900XT Black edition.

https://youtu.be/hrw4EswaVQE

[dry crow]

its huge

[dense tapir]

They all are

[oak osprey]

pretty memorable card design. the white rings inside the fan shroud give it away

[dense tapir]

a lot of people are going waterblock just to remove 25% of the length but nothing can be done about the depth needed.

Not a lot of cases allow vertical mounting out there but then that brings in other issues unless properly made for airflow

[oak osprey]

that's not a big deal unless it's passively cooled.

😛

that's what fans are for

if you have some industrial equipment, like the celeron router i've got, needs to be mounted with the fins vertical so that the cooling works properly. but that's because it has no fan. add a fan, that problem disappears.

bigger issue is mounting it in an odd orientation causing sag but vertical mount is better for that issue too.

ComfyUI's code is a mess

https://github.com/huggingface/diffusers/commit/6cbddf558adc2992c77e640fdaadf0584fbfa233

diffusers now has native tiling upscaler 🙂

[wispy ether]

me when me when the

[oak osprey]

look at that gender reveal party

[dense tapir]

@smoky oak Oh, my.

[marble wasp]

"Error: index 1000 is out of bounds for dimension 0 with size 1000" tried to add another 25 steps. Know what this means?

" Gone are the old ways, Gone are the old days."

[tropic trout]

Hi all, is there a help channel where I can post an image/question? Or would I do that here?

[oak osprey]

depends on the Q

[tropic trout]

It's about img2img alternative test results

I posted it in tech support but maybe not the right channel

[dense tapir]

@smoky oak https://github.com/Uminosachi/sd-webui-inpaint-anything

[smoky oak]

So I keep seeing this, but I have yet to like... look into it 😅

[dense tapir]

I brought it to your attention since this is the biggest key feature of firefly.

[oak osprey]

isn't that, that gal

hannah owo

[tranquil token]

It is, tried keeping the Lora low, just to switch up the face of the model im using, wasnt going for full on her

not hannah

maybe this will tell you

[oak osprey]

nope

[tranquil token]

its Belle delphine

[oak osprey]

ah i'd forgotten her name, only saw some deep-dive on her weird content once

the heck

[tranquil token]

[oak osprey]

the before pics are frightening

[tranquil token]

doubt thats really her, but we never know i guess

[oak osprey]

train that lora

[tranquil token]

just wanted a good looking lora to switch up the faces a bit

[oak osprey]

do billy harrington

he's a cutie pie

[tranquil token]

ive never trained a lora.. dont know how

[oak osprey]

oh snap, do a Forsen LoRA

that'd be hilarious

you need like 5-6 images or so and then it takes maybe 5-15 min to train

[tranquil token]

is it really that easy?

[oak osprey]

YMMV

your mileage may vary but yeah there's complete idiots that pull it off, i know because i helped one with it 😄

[tranquil token]

maybe ill look into it one day

[tropic trout]

Anyone know why my img2img alternative test results are looking like this? Using that and Euler is the closest I can get to retaining the original likeness of the photo but stylizing it. Any other sampling method or not using img2img alternative changes it too much. But I can't seem to get it to not be all deep fried like this. It looks exactly how I want it otherwise, matches my original photo but stylized

[smoky oak]

I will have to look into it. I know that as the days go by, Firefly gets more and more outclassed lol

Now we can do the stylized text as well, and the vectorization, and SD already kicks the shit out of firefly with all forms of standard generation, and now this

This is why I think that having such an incomplete tool like firefly, where its definitely behind the curve, is a bad idea

Mainly cause they are showing all of their cards up there sleeves before they are particularly good at any of them, which means SD will have everything they are flexing before Firefly is even out of beta

[dense tapir]

The only thing I like about Firefly is that it allows the masses to be more at ease with AI as it becomes more common place. This is akin to how 3d rendering was treated in the 1990s to now it is so common place no one says anything negative about it but in the 1990s wow.

[tranquil token]

You not accepting DMs

[smoky oak]

yeah, true, but I am not sure having one of the most corrupt and malicious creative companies out there doing this will put a good taste in peoples mouths

I know it doesn't for me

[tropic trout]

@tranquil token Sorry how do I change that?

@tranquil token I changed it

try now

[dense tapir]

Most people don't care, like our EU person. NGL people pay for that so AI coming to them is a very good thing in the long run for us, over here.

[smoky oak]

yeah, true

I have just never been one to go after less for more

[dense tapir]

Not even about that it is just to acclimate people to AI so they leave us the fuck alone and stop talking shit about it, us (we are so toxic, remember?), and how we aren't really artists since we didn't actually create it. Telling ya it reminds me 100% of the days 3d rendering hit. Same shit different asshole syndrome.

[smoky oak]

Yeah, I can get that, doesn't mean I have to be happy that people are getting acclimated to the lesser of what this tech can actually do

[dense tapir]

Ahh, yes but then they get used to it to then see what SD can do via a plugin and all for free.

for them it is a complete win as they have PS, Firefly, and SD

[smoky oak]

Yeah, that is true I suppose

[dense tapir]

once the extension matures they will use SD more and more and firefly less

[smoky oak]

and its not like SD can just be bought out like older technologies were in their infancy

[dense tapir]

well, I wouldn't go that far

everything has a price, sadly

Did you see that 4050 img I pinged ya with?

Oh, and Titan will be about 3-3.5 grand

[smoky oak]

So they are actually making a titan this gen?

[dense tapir]

yes

[smoky oak]

so stupid

[dense tapir]

it is what killed 4090ti

[smoky oak]

Well ok, it actually makes more sense than a 4090 TBH

4090 is... so awkward

Incredible performance, good VRAM, useless in professional programs

[dense tapir]

a 4090ti would have used the last 16% of the die

if the 4090 is useless then the 3090 before it would have been too. The only real difference between the 4090 and the titan will be the 48gb of vram and not certain it will be 48 as it might be 40

[smoky oak]

The 3090 was useless for most actually professional work unfortunately

No Quadro/Titan drivers

It was so bad that teh 2011 Titan was actualyl faster in a lot of Titan programs

[dense tapir]

Yes, this is like with Solidworks a geforce was useless. Oh, it worked, but a lot of shit was greyed out and couldn't be used unless you had, at least, a Quadro crad.

*card

[static tusk]

honestly I think thats for the best. Have businesses using gaming cards for professional work would drive prices up like crazy

[dense tapir]

Oh, they do just not the datacenters

workstations I have seen a lot of geforce when a quadro is what they should have had

I still wish Nvidia had made an accelerator AI card for consumers. Hell, they just may but I think they are more concerned with Hopper and pro sales right now. Then when they are all sold and the companies don't buy again for 10-15 years Jensen will go crawling back to the gamers.

[oak osprey]

i don't get all the doom and gloom, there was some video released last week talking about how "IT'S ALL OVER FOR GAMERS"

[dense tapir]

Unis and datacenters are still using Maxwells so I suspect some of the hopper sales are to them.

[static tusk]

Nvidia would be planning and designing cards years ahead of release. Right now they would all be working on 50 series designs and testing

[dense tapir]

yes

[static tusk]

yep its not all doom

[dense tapir]

well, I have little faith in Nvidia considering their track record since 2020.

I figure this will finally come settling down around the 6k series though none of us may have the cash for one if Jensen had his way. We shall see.

[oak osprey]

what's wild is all of the people asking for lower priced, high VRAM cards, would be immediately met with shortages as scalpers come along to use them for AI

[dense tapir]

btw, shipping costs is affecting cases/chasis now with 20-40% price increases. 😦

[static tusk]

Chia farmers are using 3060Ti cards now

[oak osprey]

i thought that was a hard drive thi--- no, i don't want to know

@static tusk looks like HF is updating diffusers soon to have a new SNR training schedule on top of terminal SNR https://github.com/huggingface/diffusers/pull/3616/files#diff-8cc7d2ea599969de32066eed057eccbdd71f1f60382d9e78c0b6644c7124fb53R52-R53

pulled from https://github.com/TiankaiHang/Min-SNR-Diffusion-Training/blob/521b624bd70c67cee4bdf49225915f5945a872e3/guided_diffusion/gaussian_diffusion.py#L847-L849

Abstract. Denoising diffusion models have been a mainstream approach for image generation, however, training these models often suffers from slow convergence. In this paper, we discovered that the slow convergence is partly due to conflicting optimization directions between timesteps. To address this issue, we treat the diffusion training as a multi-task learning problem, and introduce a simple yet effective approach referred to as Min-SNR-$\gamma$. This method adapts loss weights of timesteps based on clamped signal-to-noise ratios, which effectively balances the conflicts among timesteps. Our results demonstrate a significant improvement in converging speed, 3.4x faster than previous weighting strategies. It is also more effective, achieving a new record FID score of 2.06 on the ImageNet 256x256 benchmark using smaller architectures than that employed in previous state-of-the-art.

[static tusk]

I have no idea if we can update diffusers for A1111. I do like the toolset in it

[oak osprey]

whatever Diffusers are in A1111 are just there for convenience, they've essentially lifted code out of it and put it into A1111 in odd ways. it's a collection of code pulled out of the repositories published by researchers, especially for their samplers

[static tusk]

I do a lot more than just generate images and save them, so having tools available is make or break

[oak osprey]

was looking at ComfyUI trying to figure out how they got scaled CFG working and it's like they wrapped around k_diffusion instead of diffusers so it's a lot harder to figure out what's happening. the terms in use between k-diffusion and Diffusers are not always the same

[static tusk]

yep different methods

[oak osprey]

there's also a lot of custom wrapping code in there

it's not really commented or organised in a way that helps understand it very well

[calm canopy]

you could also just ask comfy himself lol hes active in the other popular discords

olivio has a channel just for him that he responds to almost daily

[wispy nest]

done with my horror style, just look and imagine the rent of the place!

[oak osprey]

that new controlnet stuff is uhm, nightmare fuel 😄

[oak osprey]

i kinda did better but

it's hard to use

[tired peak]

is this good?

[wispy nest]

yeah, I also like the van gogh model :D

[tired peak]

there is a van gogh model? im using dreamlike photoreal 2.0

[wispy nest]

yeah, I remember using it to get a similar style some time ago, still good images you got :)

[sterile temple]

You should try Camille Pissarro

[tired peak]

pretty nice

i got an issue w many of the artsy results being fuzzy or grainy or low resolution looking

w dreamlike photoreal 2.0

[sterile temple]

have you tried many models?

[smoky oak]

I actually quite like the texture on this gen, but it might be your negative that could be causing issues

also, curious to know if you are using high res fix, which could be causing the issues

[tired peak]

usually the dreamlike photoreal 2.0

[tired peak]

not using high res fix at all. it's possible some neg prompts are hurting it. hm

[smoky oak]

Very interesting... If you would be willing to send me your prompts in DM's, I could download the same model and try it out

[kind quartz]

@smoky oak pnginfo 😄

[smoky oak]

not sure if they have that or not

I make sure to send my images without PNGinfo

[kind quartz]

it has it. i hope you dont want to earn money from it:)

[smoky oak]

I am not gonna use somebody elses prompts to make money lmao

I make money off my own creations :p

[kind quartz]

same shame 😛 Try earn money some harder way

[sterile temple]

I usually leave the png info when posting here. Stuff I'm playing around with. But I don't intend to make money off this, it's just a hobby for me.

[kind quartz]

yes thats the spirit

[smoky oak]

Not sure what thats supposed to mean

I make money off my work, it is what it is lol

[tired peak]

you can just put the same image into the png info box

[kind quartz]

o.k. if your job, then o.k.

[smoky oak]

@tired peakLooks like I get the same texture as well, which suggests its the prompt

I think its cause you are using words like Van Gogh, and Oil Painting, which make a more, well... painterly look lol

[tired peak]

ok sent u a msg

[kind quartz]

also paintings opening way CANT be visualised on photo.

[tired peak]

hm maybe

[dark flare]

CONTROLNET!!!

Before

After

[tired peak]

rly good

[dark flare]

OH YEAH 😆

[tired peak]

all this stuff is too difficult to install

i cant even find the SD 2.1 yaml file that everyone keeps talking about

so im using 1.5

[dark flare]

for CNet?

[tired peak]

just for anything

the yaml thing to put in the models folder together with SD 2.1

[dark flare]

hehe 😄

I can help you with the ControlNet yaml but not the SD yaml 😉

[smoky oak]

SD 2.1 Yaml's should be included with the download of the models themselves

Althought, i would say you are not missing out by using 2.1. Almost no good models out there unfortunately

[tired peak]

hm

so u guys just use sd 1.5 and models and controlnet?

[smoky oak]

I use just 1.5, yes

I also train my own addons, as I see fit

[tired peak]

ok where is the controlnet Yaml

[smoky oak]

that is actually what I make money off of

[dark flare]

I'm not using 1.5 presently

[tired peak]

i want to start making money too

[dark flare]

Making all my images with
https://civitai.com/models/27739/artius-v21

[tired peak]

i only sell tokens so far magic the gathering. but it's annoying and time consuming

[dark flare]

Making money with AI images is probably not realistic, because anybody can do it

[smoky oak]

I'll take a look, yet to find an actually good 2.1 model

[smoky oak]

It is very realistic, just gotta know what you are doing

[dark flare]

If you can take your machine and go back in time, 2-3 years ... yeah you could be mega rich 😄

Now everybody just knows "oh, that was AI"

In 6 months we might not be able to tell ... but again ... anybody will be able to do it 🤷‍♂️

[smoky oak]

interesting, this Artius model looks pretty good

[tired peak]

i've been doing things like this

[smoky oak]

looks really good for a 2.1 model

[dark flare]

Can you CAPTIVATE people? Can you INSPIRE them? That's about storytelling ... which ChatGPT is getting better at 😆

[smoky oak]

Aha, my friend and I were working on some MTG cards as well lol

[dark flare]

Those look good!

It's less about the images though, and more about the story behind them, or the purpose, the function they serve in the game

[smoky oak]

But anyways, there are many ways to make money off of AI still, just gotta know what you are doing

[dark flare]

You still need those "good ideas" and then "a good image to go along with it" .... is less important? I think

[tired peak]

it's part of a series of 7 different knights

ive made over 100 tokens so far. sold over 200

[dark flare]

I think it's the greatest freakin' model ever 😄 only thing it can't seem to do ... "darkness" images. Always so much lighting. Beautiful but garrr

[smoky oak]

I wouldn't say its that good, not even close, but it does look very good for a 2.1 model

I am testing it right now, seeing if it still has all the same problems associated with 2.1

[tired peak]

elemental token

[sterile temple]

back to headphone images

[smoky oak]

crusty ol werewolf haha

[dark flare]

Try my latest set of negative keywords

3d, 3d render, 3dcg, amputation, animated, bad anatomy, bad art, blurry, body out of frame, border, brothers, cartoon, clones, copyright, crossed eyes, cross-eye, deformed, desaturated, disfigured, doll, double, duplicate, extra arms, extra legs, extra limbs, frame, hat, illustration, logo, long neck, lowres, make up, mutated, mutated hands and fingers, mutation, out of frame, over saturated, photoshop, poorly drawn, poorly drawn face, poorly drawn feet, poorly drawn hands, repeated person, same face, signature, sisters, split image, text, tiling, twins, ugly, video game, watermark, wrong anatomy

[dark flare]

Nice, that doesn't reek of AI 😄

[dark flare]

hehehe yes sir

[smoky oak]

Thank you, I already have my own collection of netaives for my own use though

Maybe I will give it a shot if mine don't work

[tired peak]

i tried to use the lore for this poison token. elesh norn's fair basilica and the monuments, and the black ichor to symbolize corruption, leaking from the eyes . this one took hours of editing

[dark flare]

Please share! 😛

[smoky oak]

I personally don't like all of the problems that come with 2.1, and thats why I don't use it

[dark flare]

What are those?

At first I was annoyed because I couldn't make ANYTHING with it ... but I was educated that it's really about negative keywords now

[tired peak]

i used inpainting to fix his hands

[dark flare]

So I just stole a bunch of other peoples 😆 and I've been quite happy

Ah nice!

[smoky oak]

deformities, built in compression, bad proportions, several spatial and temporal artifacts, severe issues with out of focus or focal planes

Its a problem with base 2.1, been waiting for people to train it out

[sterile temple]

how's that for dark

[tired peak]

2.1 seems like a waste

i don't know how they went so backwards with the technology

[smoky oak]

it honestly is, and its a shame

Somepeople are finally learning how to train it, but its not worth it IMO

[tired peak]

it's like microsoft releasing a new console that is 2x less powerful

[dark flare]

Well, I was talking specifically about the "artius" model 😄 but yes, nicely dark

[tired peak]

in my models folder i see a ''control net'' folder and it's empty inside.

and i have a stable-diffusion folder with some models in there but nothing controlnet

[kind quartz]

you have to download them manualy.

[smoky oak]

@dark flareHonestly, this model is not bad for a 2.1 model

Its still not as good as a lot of 1.5 models, but this one isn't complete garbage, far from it infact! haha

[dark flare]

hehe 😄

[smoky oak]

glad to see its a big step in the right direction for 2.1

[dark flare]

I got into SD with 1.4; had some life changes when 1.5 came out; when I got back, it was all about 2.1 ... so I've skipped a lot

[smoky oak]

2.1 is pretty much dead now, unfortunately

[smoky oak]

The werewolves I tend to gen are... Hotter 😅

nice to see some creepier werewolves haha

[sterile temple]

hotter?

[smoky oak]

😅

[tired peak]

i will restart my pc. maybe that will get controlnet working

[smoky oak]

no, no lmao

just a sec

Hotter 😅

some Cerberus hell hound fun times lol

[sterile temple]

oh okay, gotcha 😉

[smoky oak]

This model is made to make more attractive anthro chacters haha

[sterile temple]

what's wrong with my roid dog? haha

[smoky oak]

@dark flareYou know what, this is hands down the best 2.1 model I have ever used. It still has some of the common issues with 2.1, but its dramatically better than most honestly

[sterile temple]

I believe the person who made it also won the last POW

don't think my model knows cerberus

[smoky oak]

I honestly don't think i will find a 2.1 model that will be able to match 1.5 realism for a while

Left is a 1.5 model, other two are 2.1 models

I should also note that the 1.5 gen there is far out of native res, so it is struggling harder than it normally would

[sterile temple]

I found 2.1 models too gaunt, like anorexic looking faces

[smoky oak]

Yeah, same here

[dark flare]

You can always add "obese" or something 😄

[smoky oak]

we shouldn't have to, thats the problem

[dark flare]

Models are certainly biased towards beautiful people, you have to say "with a big nose" or something to make them more real

The machine does "perfect"

Perfect faces, symmetrical, balanced, golden ratios, etc

But then it also makes horrible mutants 😄

[kind quartz]

@tired peak there is model for fantasy card or d&d

[sterile temple]

I just want an average body as the default, and clothes that aren't falling apart

[dark flare]

Unlikely .... .... ... since people are voting for "hot" people continuously

[wispy nest]

You got unlucky lol.

I'm playing around with 11labs, picking random stuff to let AI spell it lol

[tired peak]

is that an AI voice?

[wispy nest]

yea

[dark flare]

See? Even the voices are hot 😉

[smoky oak]

Sounds like one

[dark flare]

Can it make an ugly voice?

[smoky oak]

yeah, looks like it is

[smoky oak]

They can do trump, pretty sure that counts lol

[dark flare]

hehehe gurrrg

[wispy nest]

yea you can train any voice

[dark flare]

hehe nice, that actually felt like Trump at a couple points, the pacing

[wispy nest]

so far emotions , pauses looks random

[dark flare]

Was that made with the "free" package?

[wispy nest]

I mean it can read , , . e.t.c , but sometimes it just makes weird stuff

[wispy nest]

yea

[dark flare]

I keep meaning to play with it but then I forget 😛

[wispy nest]

today is the day to play, tomorrow it should reset lol

[dark flare]

Reset to what 😱

[wispy nest]

It gives you 10k symbols per month for free

[dark flare]

aahhhhh

[wispy nest]

so...end of the month

[dark flare]

How many symbols were used in creating those 2 examples?

[wispy nest]

last one - 251, previous idk , a bit less maybe

[dark flare]

Ah not bad

[wispy nest]

That one was funny attempt to make it read lyrics 😄

kay back to sd talk before mods invade

[tired peak]

ty. ill tryt o find it later

[wispy nest]

how are you guys specifying the style in the bots??

[smoky oak]

we aren't using bots

[wispy nest]

well about 10,000 other people are

so this question is probably for them

[smoky oak]

Why not ask in #🗣｜artisan-support-feedback ?

Most of us here have our own installs, not bots

[wispy nest]

I have my own install too

but obviously "most" people are using the bots

I'm looking for the option to specify style that everyone else is using

[wispy nest]

because it's insactive asf

[smoky oak]

good luck then, I guess

[wispy nest]

thanks for trying your hardest

[smoky oak]

A majority of people in this chat don't touch the bots, so you still have a higher chance of getting help in a dead channel, than in this channel where your message will be drowned out by actual conversation

[wispy nest]

help im drowning

[smoky oak]

Being a condescending jerk about it also probably won't help your chances of getting helped lol

[dark flare]

aw man

[wispy nest]

how did your feelings possibly get hurt

[dark flare]

I thought it was funny 😛

hehehe

[wispy nest]

I'm asking a question, you don't know the answer, and yet you're here flapping your mouth

[dark flare]

HEY LOOK WHAT I MADE YESTERDAY

[dark flare]

Wait ... which is the one where you can train it with your own voice? The same one?

[wispy nest]

ohhhh it makes sense

[dark flare]

🤦‍♂️

[wispy nest]

lmao

cool image bro

[smoky oak]

Imagine thinking that telling people you are a dude is indicative of anything other than basic courtesy lol

[smoky oak]

What a mischievous little fellow lol

[kind quartz]

@tired peak should be called somehow majicmixrealistic or similar

[wispy nest]

Training only paid here

[dark flare]

Ah! Thanks!

[wispy nest]

you're still mad but nobody can figure out why

And speaking of 2.1, I've not been able to get any horror monsters made with it, same with a lot of 1.5 models as well to be honest. Most results are just silly faces with too many teeth :P

[tired peak]

might be good for horror monsters

for 1.5

[smoky oak]

I'm sure there are some models that are good at that, it also depends greatly on your prompting abilities

Although, I do find that 2.1 tends to generate things closer to monsters than people most of the time, so maybe that works? Haha

[wispy nest]

hehe, I might need to add that I wanted the "no cool monster" types. I was able to get what I wanted, but using 2.1 models were a real no go for this type of thing. And yes, the image is using a 1.5 model with 2.1 default size because 1.5 does it better as well ;P

[smoky oak]

I'll see if I can give some monsters a shot when I get back to my PC

[smoky oak]

Reminds me of the mind flayer

[wispy nest]

my monsters are always so loud…no wait, I mean my neighbours ;P

[tired peak]

i have thousands for monsters i think

[wispy nest]

yeah, those results are typical when I tried in most models, you get a bodyless head type of thing that "hangs around" ;P

[smoky oak]

Hmmm, our wifi is having some big issues for whatever reason

Speed is definitely slower on this side of the house, but 150/250 should be plenty fast lmao

Not sure what's happening

Maybe it's discord

[wispy nest]

I'm trying to get a sort of "what even is that" feel, but yeah, it's easier said than done ;P

[smoky oak]

Like the images, they take like 10 seconds to load

I'm not sure why... Hmm

Restarted discord, let's see if that helped

I'm gonna have to assume it's discord, not our internet

[shut sinew]

Discord has been kinda slow for me as well, uploading an image sometimes takes 20 seconds or more right now

[wispy nest]

I got no issues with images, but yesterday I had some weird thing where discord lagged like nothing before, then it suddenly restarted by itself :/
Too bad about the flying thing in the corner

[shut sinew]

we doing creepy stuff huh?

[wispy nest]

I've been doing that the entire day...wait, no not like that! :P

[smoky oak]

ok, that at least reassures me that its probably discord and not my connection

[shut sinew]

yeah i got a 500k connection here and a 3 MB images takes a good while

[smoky oak]

500k?

[smoky oak]

500 Kbps?

[shut sinew]

500 Mbps

[smoky oak]

ah, confused me with the k lol

we have 500 as well, so its likely discord is just goofin

[shut sinew]

[wispy nest]

ooooo, real spoopy!

[sterile temple]

pics take a while on my discord too

[smoky oak]

alright, cool, reassuring for me haha

we had our modem kill itself like 2 weeks ago, and had no internet for 6 days cause of it lmao

so I am a little cautious haha

[sterile temple]

I just upgraded to 5G internet, because the best you could get here is 100/40 megabit

[smoky oak]

5G internet is dope if you are in a place that supports it

[shut sinew]

got a head holding one too

[smoky oak]

we are lucky, our rural town go reworked with full "Fivber" lol

[sterile temple]

head scratches

[smoky oak]

which is soooo needlessly fast for where we are

IDK who the hell needs internet that fast for residential lmao

[sterile temple]

people find a way to use it don't worry

[smoky oak]

The thing is, I don't know any hardware for consumers that can even used fivber speed

[sterile temple]

I wish they hurry up and upgrade to fiber here, was announced 2 years ago my suburb was supposed to be getting it

[wispy nest]

yet ;P

[smoky oak]

even with "yet", I just don't see a need for a long while, my 500/500 is plenty fast

even with me downloading TB's of data every month

[shut sinew]

Its always good to download those huge games faster these days

[wispy nest]

I wish I had the same up-load as download 😭

[smoky oak]

that is the biggest improvement since we went fiber

with old cable, we would have had 1000/20 lmao

[wispy nest]

I remember, what was it, 26k speed :P

[smoky oak]

and it would have cost more than 3x what we have for 500/500

[shut sinew]

slowest i ever had was 256kb/s in a huge city

[smoky oak]

but man, I just don't know what the hell I would do with internet 10x faster

[wispy nest]

and I remember when you had to install the internet with a cd ;P

[smoky oak]

most places I download from have a speed cap below 500mbps

[shut sinew]

CD lol, 💾

[smoky oak]

I genuinely wanna ask somebody what they use their fivber for

OH, wait

[static tusk]

I preloaded Diablo 4 today. 84GB took me 5 hours

[smoky oak]

I have been using a slang term lol

"fivber" is what they call it here

Its 5 gigabit

[wispy nest]

silly, the internet never fitted on a disc, it had to be those huge 6-800 MB cds! >:D

[smoky oak]

they used to call it fiber five, but they just call it fivber here now lol

also tried V-ber, but that sounds too... Naughty lol

[shut sinew]

Ill refuse to pay blizzard any more money

Even as a huge diablo fan

[smoky oak]

but really, what the hell would I do with 5000/5000

[static tusk]

diablo, must have.

[shut sinew]

host a datacenter

[wispy nest]

by the way, I just noticed some messages here stutter into view, so there's probably some issue, or someone using up all the internet…dust?

[smoky oak]

but thats their residentialinternet

Their business plans go up to 100 gigabit

[shut sinew]

Idk man, where im from were living 5 years in the past it feels

[smoky oak]

or well, thats as high as I can see when I browse their website lol

They do have a section that says something like "Need more speed? Message us and we'll figure something out"

but as a consumer, I have 0 clue what somebody would use 5 gigabit for

the latency isn't any lower, cause fiber is fiber through and through

[wispy nest]

I feel so small, and not only because of the image :P

[smoky oak]

interesting, didn't know that Singapore has the highest average fixed broadband speed at 242mbps country wide average

[smoky oak]

"How it gum to chew 5 feels"

https://tenor.com/view/how-it-feels-to-chew-five-gum69-gif-22974393

[shut sinew]

Feeling small you say

[wispy nest]

that person need to turn around :P

[shut sinew]

"Don't be afraid"

[wispy nest]

mine's not as tall >:I

[wispy nest]

I'd say it looks good, but I don't know what "it" even is, so I'll just say: good! :P

[sterile temple]

Monstrous entity attempting to consume larger cosmic horror, struggle, vibrant energy colors, style inspired by the unsettling organic forms in Zdzisław Beksiński's work, detailed textures, swirling chaos of the cosmos backdrop

[wispy nest]

I never learned if the ai understood none "standard" latin letters 🤔

[sterile temple]

it seems to be understanding it to get the style

[wispy nest]

yeah, but I always wrote that artist name without the weird letters, I also noticed you only needed to write his last name and not the first :P

[shut sinew]

[sterile temple]

dunno why my brain thought FLCL when I saw that

okay that's enough Eldritch for now 😄

[sterile temple]

treehouse of horrors

[shut sinew]

and enough shrooms to make this up

[shut sinew]

cyborgs now

[sterile temple]

bleep blorp!

[shut sinew]

got another one

[sterile temple]

purdy

[shut sinew]