🏞｜general-with-images

[wispy nest]

logo

[sinful mirage]

mixture of inpainting and outpainting, kinda obvious in some spots

[dense tapir]

@smoky oak Well, I was right again as I said over six months ago that Google banned Deepfakes in 2018 as they were taking too many resources and I expected them to do the same to this. Sure enough free accounts (for now) are banned from using gradio. Use it and get a warning that will eventually get your account banned. They will do it too.

[smoky oak]

I can't say its surprising, but it is really scummy IMO

Why provide free accounts if you are gonna limit what you can do with them

Just cut down the time or limit the GPU's they have access to if its that big of a deal

[dense tapir]

Well, yes to all of that but I do understand in this one respect. I say that because assholes were hosting 3rd party FOR PAY shit on their servers that was free.

[smoky oak]

Oh, thats even scummier lmao

[dense tapir]

yep

[smoky oak]

Yeah in that case I can understand

[dense tapir]

https://old.reddit.com/r/StableDiffusion/comments/12t8tc7/is_colab_going_to_start_banning_people_who_use_it/

[smoky oak]

Alright I can get that

[dense tapir]

I left them a couple to three weeks ago as it was zapping my will to live. I saw it coming when 12h became 6 then 4 then 2

[smoky oak]

so they are saying that the sheer amount of free web UI usage isn't able to be floated by the paid users, which I understand

[dense tapir]

They are saying they can't float it.

no more free rides

[smoky oak]

Yeah, I can understand that

[dense tapir]

again, because asshole companies were hosting APIs for pay that was using Google's free

[smoky oak]

it was good while it lasted, but too many people cashed in on their frankly generous offer and it is starting to hurt them

[dense tapir]

Yeah.

[smoky oak]

Google collab is an insanely useful and infinitely beneficial platform, and I can understand why they are moving away from this, even if its a detriment to those who were using it responsibly

call it another example of trashy people playing the system ruining it for everybody else

unfortunately, I have a feeling this is gonna make the active SD community a decent bit smaller considering the amount o people I have seen that run SD on collab

[dense tapir]

Correct on all accounts

[smoky oak]

I am just as fast to jump and call out companies on capitalist behavior, or when they purposely try to make things sound worse than they are, but genuinely when it comes to Google Collab I can honestly say that I think that what they do is extremely beneficial, and extremely reasonable. So it's very sad to hear that they are having their generosity of allowing people to use that powerful service for free take it advantage of, and subsequently ruining it for everybody else

[dense tapir]

I wonder if there could be anyway they could allow it for others while cutting off the BS bad actors?

I wish they could, but remember in 2018 this happened too but it was NOT companies it was just sheer amount of users training deepfakes. I honestly believe that is the case now too as it was draining their resources.

SD became too popular

[wispy nest]

Sooo.....

first kaggle bans, now google...

[dense tapir]

yes

[wispy nest]

we got no more good free services

[dense tapir]

nope

Well, there is still free but I will be damned if I ever could get a card so gave up. Might as well say nope.

I think SD over there too sucking up all their resources as they gave faster cards than T4

betcha next will be banned there too

[wispy nest]

Last time I used it there was only T4's

[dense tapir]

not colab, I am talking about AWS

[wispy nest]

isn't aws pain in the butt to register?

oh...it's cause I'm from Russia, I recall now , nvm 😄

[dense tapir]

nope, they let me within 10m but I never could get their fancier card to even try it

Yes, that would be why

Their service was so fancy and professional but a lot of good that did me

[wispy nest]

need to buy another temporary phone then...that would also let me try their coding assistant thing

[dense tapir]

Well, good luck as they never have free cards to use. A lot of people left colab to go there since it is a superior service but I bet they stop that

[wispy nest]

no more SD for me

[dense tapir]

well, no more gradio app

that is what they banned

[wispy nest]

so...you are saying I can still use collab without UI?

That's limiting tho 😦

[dense tapir]

https://preview.redd.it/l6mk996t94va1.png?width=1917&format=png&auto=webp&v=enabled&s=fb6945f57993c04dfc390b793202834fe26fe477

automatic1111 notebook

I see what they did

Using the Wayback Machine, I compared the FAQ page today with the page last archived. The changes are underlined in red.https://preview.redd.it/r2xurpey94va1.png?width=1164&format=png&auto=webp&v=enabled&s=5074bcddde850d6dde61aa789014e62c108c755a

That was from the reddit post.

[wispy nest]

yea just webui then, but can't do alot without it

[dense tapir]

I know

ffs

:/

[alpine dragon]

Beautiful view

[dense tapir]

I typed in ooga booga and got this

[hasty nova]

imma do that next

booga ooga

[hasty nova]

so I put in "ooga booga"

and

um

Im not sure if I should share the results

Not nsfw or anything but slightly offensive probably

[dense tapir]

ut oh, lol

I am not easily offended but I getcha

[hasty nova]

Yeah I don't want to make anyone else mad.

tho I spose you can guess what I got based on what I am saying

[dense tapir]

yes

depends heavily on the model used

[hasty nova]

yeah im currently messing around with a 2.1 photorealism model

ill stick with my pretty forests

[dense tapir]

which model?

[hasty nova]

realism engine

https://civitai.com/models/17277/realism-engine

[dense tapir]

yep, I knew it

my fav

[hasty nova]

yeah it's very nice

I don't usually do pr

gonna try one word prompt

[dense tapir]

I mixed that one with an anime one and yeah, it was cool

[hasty nova]

I wonder if you can guess my one word prompt

[dense tapir]

interesting

[hasty nova]

why does one of the banans look like an eggplant

yet another one word prompt

[dense tapir]

the one on the left looks like a summer squash

[hasty nova]

yeah a little bit

some sort of gourd

[dense tapir]

okay, now this is amazing

"birth of a nation"

[hasty nova]

yo crazy

I wanna do that one

[dense tapir]

that was realism engine btw

[hasty nova]

these are really cool

[dense tapir]

see

[hasty nova]

(they are a wider aspect ratio, discord is making them squares with the gallery images thing)

[dense tapir]

Aye

[hasty nova]

a really furrowed brow

thats a word right

[hasty nova]

doesn't seem like the right word

gonna try a 3 word prompt

meh kinda sucks

[dense tapir]

Illuminati

[hasty nova]

personally from the images I made I prefer realism engine, the black and white is cool

[dense tapir]

earlier I did a b&w

1 word illuminati

[hasty nova]

trees I guess

[dense tapir]

lets try realism

[hasty nova]

anyway I gotta go to bed

gn!

[dense tapir]

night

realismengine 1 word

[quasi pike]

Eid Mubarak

[wispy nest]

judging from this logo on the exosuit, i feel like GPT is infiltrating my SD install lol. I did not prompt anything with GPT and that is really well done writing on the suit

[glossy herald]

Metadatas for image shared through discord are baaaaaaack !

(you still need to download the pic/open in browser)

[glossy herald]

@livid ermine #🏞｜general-with-images message

without training at all, you can use some creativity and do good things with base visuals too

[livid ermine]

🤔

[livid ermine]

🤣

Original:

[glossy herald]

lol

maybe a wrong setting there

[livid ermine]

Predictions:

[glossy herald]

x)

[livid ermine]

hahah

[glossy herald]

ho

[livid ermine]

yeah 🙂

[glossy herald]

yeah I though your face was the output

that seems more like it

canny mode ?

try mlsd with architecture 🙂

[livid ermine]

😄

Sorta did just that, plus some GIMP-stuff afterwards.

I actually work with the people who designed that building, www.snohetta.com/projects, many claim they're the best architect firm in the world.

[glossy herald]

let me try it :p

[livid ermine]

😄

You gotta know the secret formula though lol

If not I'll tell ya afterwards 😄

[glossy herald]

https://media.tenor.com/tzXPnnWpB0MAAAAd/felicity-arrow.gif

[livid ermine]

"parametric" 🙂

[glossy herald]

wait

don't spoil the sauce

[livid ermine]

parametric architecture, the future of human habitations

has an enormous effect on the outcome

my bad 🤣

[glossy herald]

depends on the intended result anyway

[livid ermine]

Yeah parametric architecture doesn't look like that, the Oslo Opera wasnt parametric to begin with

But it's worth checking out!

Brilliant visuals

[livid ermine]

Wow, eerie!

Really nice

[glossy herald]

I'm trying different settings

a little more "artistic" than photorealistic though

[livid ermine]

Indeed

[glossy herald]

I'll change model

[livid ermine]

Btw now that you have the negro version of me, here's the viking version 🤣

[glossy herald]

please though

no racial slur

[livid ermine]

Oh that's a slur?

I'm mixed race, I didn't know

[glossy herald]

from what I got last time someone said it around and a drama started yeah, I'd prefer if you replaced it with black for example

[livid ermine]

Indeed, I'll keep that in mind.

👍

[crystal delta]

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!

[wispy nest]

How dose this Violate their TOS

[glossy herald]

it seems they nuked the webui on google colab free tier last few hours :/ https://twitter.com/thechrisperry/status/1649189902079381505

[leaden night]

how can i upscale and get more detail, without changing the prompt and wrecking the image. this is the upscaled image with noise reduced up to the point it doesnt majorly change the image.

[tawny forge]

I am running it on google colab now, should I stop then?

[glossy herald]

no no, continue ^^ it gets stoped by google themselves, lots of people report they can't run the webui anymore

this just targets webui though, using other notebooks without UI doesn't seem to be a problem either on free tier

also it just affect free tier

[tawny forge]

alright

so if I only use commands to generate images everything will be fine?

thanks

[glossy herald]

yep, original diffusers library is currently safe

no idea what they change tomorrow though

[tawny forge]

it is kinda sad they are banning the web ui

I finally built a code to run the colab and then I read this

[glossy herald]

yeah... it's a fringe use to colab though. it's supposed to be meant to run code snipsets, not power your own WebUI, so I do get why, especially if it's impacting their cluster since everyone does it now

[tawny forge]

yea

I am using it only because I dont have a powerful GPU

if that wasnt the case I would not be using colab

[coarse skiff]

same here, I was asking yesterday about this and now I have this problem

is there's any alternative to running in google colab?

I was training a model

[simple venture]

@glossy herald first generated image (upscaled)

[gloomy elbow]

this tile-able generation with the simple prompt "numbness" is appropriately uncomfortable...

[wispy ether]

couldnt you have censored that

[glossy herald]

@wispy ether

[wispy ether]

how

[glossy herald]

I use a base prompt "a painting of a girl with a hoodie on her head, intricate details"

[glossy herald]

and a dynamic prompt on top

[wispy ether]

can you copy paste the prompt

[glossy herald]

first I use different settings in there

I lowered the threshold

so it keeps more details

I used that scribbel_xdog because the preview was the best of the scribbles

it needs a pure black and white format, can't just use levels of grey

[wispy ether]

🗿

[glossy herald]

the example I showed when using no preprocessor at all what pure black and white

not grey

it's your problem, because of it you need to use a preprocessor

[wispy ether]

which

[glossy herald]

a painting of a girl with a hoodie on her head, an abstract sculpture by Aleksandr Ivanovich Laktionov , [:anime line art:0.2], Thunder and astral themed, epic splash art , harlem renaissance, , realistic
Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 264555375, Size: 512x640, Model hash: c4947fa2c4, Model: DB_mix_realbiter_v10, Clip skip: 2, ControlNet-0 Enabled: True, ControlNet-0 Module: scribble_xdog, ControlNet-0 Model: control_v11p_sd15_scribble [d4ba51ff], ControlNet-0 Weight: 1, ControlNet-0 Guidance Start: 0, ControlNet-0 Guidance End: 1

[wispy ether]

🗿

[glossy herald]

#🏞｜general-with-images message

set the controlnet like this

lol

you forgot to check "enable" on controlnet

x)

[wispy ether]

doesnt show how i get one

nvm

its broken

i get weird shit

[wispy ether]

how you made this

i get this

a painting of a girl with a hoodie on her head, an abstract sculpture by Aleksandr Ivanovich Laktionov , [:anime line art:0.2], Thunder and astral themed, epic splash art , harlem renaissance, , realistic
Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 264555375, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, ControlNet Enabled: True, ControlNet Module: none, ControlNet Model: control_v11p_sd15_scribble [d4ba51ff], ControlNet Weight: 1, ControlNet Guidance Start: 0, ControlNet Guidance End: 1

[glossy herald]

the base model

you are using 1.5

I'm on this one https://civitai.com/models/16592/realbiter

[wispy ether]

how do i get yours

ok

[wispy ether]

[glossy herald]

at last !

I liked this though

lol x)

[wispy ether]

[real sleet]

Hey what do y'all think?

[blissful wren]

https://civitai.com/models/46294/diffusion-brush-everything-sfw-nsfw-all-purpose-checkpoint-nuclear-diffusion-anime-hybrid holy crap guys im so proud did proper weights, try this <£

[dense tapir]

https://www.youtube.com/watch?v=AuaHNt1MY-U

[leaden night]

interesting. i gotta say though, theres not much margin left on any of these cards. I looked at my distributors yesterday and the cheapest i can get a 4090 is like $100 under retail. not even close to worth it

[dense tapir]

Yeah, and all of this is why I am jumping the sinking ship to AMD. Tired of Jensen and how he hates us with him saying, in print, that moore's law is dead and the days of getting more for less (or the same price) are over. No joy so off I go to team Red and Blue in a few years.

[leaden night]

being that im just starting out with AI, whats it like on the amd side of things?

[wispy ether]

nope

[leaden night]

i didnt think you could do SD and whatnot

[dense tapir]

getting much better. #1 do not go for anything less that a 7k series AMD card if you plan on AI

[wispy ether]

if you plan on ai do not ever go on amd

go on nvidia rtx

not gtx

[dense tapir]

Nep, stop being a stupid fanboy. That is so antiquated and wrong now. Previous gen yes

[wispy ether]

fp16, cuda and ai cores

[wispy ether]

what

[leaden night]

i mean, i just picked up a 1080ti for under $200

[wispy ether]

no ai cores

no tensorcores

[dense tapir]

TC I agree, that AMD lacks so if you use TensorRT AMD is DOA

[leaden night]

i mean, im not blowing a ton of cash to play around with ai

its doing what i ask of it so far

[dense tapir]

Pascal cards have no TC too

[wispy ether]

i spent 400 for a 3060 and its fine

[dense tapir]

I train so I am screwed as I need 24gb

[wispy ether]

look this

new york city explodes
Steps: 150, Sampler: Euler a, CFG scale: 7, Seed: 636382230, Size: 700x700, Model hash: 1a1cf9e766, Model: realbiter_v10

[dense tapir]

I am HIGHLY logical about this as I used Colab. Is the 7900XTX faster than the T4 and does it have more VRAM? Yes, and Yes. Done

[wispy ether]

took less than 5 seconds 🥂

[leaden night]

hmmm. i can get new 3060 for $346

[wispy ether]

bro wha

where

[dense tapir]

In the USA a 3060 is about 340ish on Amazon

[wispy ether]

get 12gb one

not 8

[dense tapir]

Yes

3060TI is over 400 which is stupid

[wispy ether]

or save for a 4070

cost double

does double

🥂

jk

[dense tapir]

fuck Jensen and the 4k series over priced turds. I hope Nvidia gets punished and their stocks fall to 10 cents per share.

[leaden night]

i mean looks like you can order them on PNYs site

[wispy ether]

pny

[dense tapir]

PNY does the pro cards for Nvidia so are good

[leaden night]

does what its supposed to, no?

[wispy ether]

weirdest gpu ever tho

[dense tapir]

All I know is I am done with Nvidia unless they magically fart out a 500 USD price reduction for the 4090.

[wispy ether]

pny design looks shit

[dense tapir]

A 7900XTX does about 20-22 it/s on 5.5.0 beta. On 5.6.0 should do about 30-35 while 4090 is about 40

[wispy ether]

my card 🥂

[leaden night]

can i ask why you care about how the card looks?

[wispy ether]

would you care

[dense tapir]

Nope, as long as it performs and is cool

[leaden night]

if it performs, no

[wispy ether]

thats a plastic cooler

[dense tapir]

I think it is safe to discount anything this one says

[wispy ether]

the image will look fire

my pc dying

[leaden night]

lol, i had to downclock the 1080ti i just put in because the psu croaks

[dense tapir]

Xetrov, do your home work as there are a lot of shills, and fanboys out there.

[leaden night]

im over here looking for budget gear

[wispy ether]

my gpu is a bottleneck

[wispy ether]

im saving for a 4070

[dense tapir]

7950XTX is due out and I hope it is a banger

lol, omg lol

[wispy ether]

[dense tapir]

4070 is their worst card for the price to date. Dismal sales and so many returns.

[wispy ether]

better than 3060

i need better card cuz i have new monitor

higher res

[dense tapir]

your money so you do you

[wispy ether]

what would you rather

i dont wanna buy a new psu just for a 4090

[dense tapir]

4070ti or 4080 tbh if you just have to stick with Nvidia and current gen.

[wispy ether]

cost lot more tho

[dense tapir]

you will regret the 4070

[wispy ether]

reason?

[dense tapir]

it can't even be overclocked. Not that you would but that says just how bad it is.

[wispy ether]

i dont overclock tho

[dense tapir]

not that you would I said

Not a single content creator has said it is worth the price and more horror stories are coming out about it since its release. As I said you do you but no fucking way would I spend my money on a 4070 base model.

[wispy ether]

most content creator buy 4090tho

[dense tapir]

no

[leaden night]

what about titans?

[wispy ether]

sucks

[dense tapir]

good cards

[wispy ether]

no

titan is weird asf

[dense tapir]

ada titan is 48 gb I think

[wispy ether]

now tell me

whats so bad about 4070

for a mid gamer and ai

[kind quartz]

https://media.discordapp.net/attachments/1018998769418784838/1070706457139486760/00567-1013785916-a_Photography_of_a_science_fiction_a_Highly_detailed_a_Photo_realistic_a_Fine_art_a_Full_screen_a_Full_image.jpg?width=960&height=640

[wispy ether]

how to i generate image above 2048x

[leaden night]

titan rtx for 800

[wispy ether]

fake

[dense tapir]

titan rtx for 800 usd?

[leaden night]

local to me

[wispy ether]

fake

price

[leaden night]

people dont typically attempt to sell fake shit and do it in person

[wispy ether]

answer my question

[wispy ether]

,

ok

[wispy ether]

moldy

[leaden night]

3060 12gb for $250

tempted to return this 1080 and pick that one up

[dense tapir]

Way better

[kind quartz]

https://cdn.discordapp.com/attachments/1018998769418784838/1095005976593190925/23022823-975302563-fruit-fruit_Highly_detailed_Photo_realistic_Fine_art_full_screen.png

[wispy ether]

@dense tapircan i generate 10bit images by any chance

[dense tapir]

I wish we could do HDR

[wispy ether]

i think my monitor is 10bit

some websites said 8bit frc and some 10

displayspecs website says 10

but brightness too low

LMOA

va panel tho

[wispy ether]

https://tenor.com/view/hog-rider-coc-clash-of-clans-gif-24053135

[wispy ether]

https://tenor.com/view/clash-royale-wizard-video-game-hair-gif-16221762

[wispy ether]

@dense tapircan u help rq

🤨

how to not

[dense tapir]

??

[wispy ether]

she got 3 arms

how to not

[dense tapir]

it is your negative prompt

[wispy ether]

what is it supposed to be

[dense tapir]

I don't know what you currently have.

[wispy ether]

i asked what its SUPPOSED to be

[dense tapir]

it depends as too much will destroy it as bad as not enough

[wispy ether]

[dense tapir]

there is no supposed to be is the fact of the matter

[dense tapir]

1.5?

[wispy ether]

?

[dense tapir]

which sd

[wispy ether]

wdym

bow

[dense tapir]

you know 1.4, 1.5, 2.0, 2.1

[wispy ether]

i dont

[dense tapir]

can't help then

[wispy ether]

this is kinda fucked up

some pokemons that doesnt exist

[kind quartz]

what is this?

[wispy ether]

half titanic

[kind quartz]

i realy dont know, seems to me like some sci-fi apartment. GN!

[wispy ether]

hm

[leaden night]

whats the key to not having mangled fingers?

[dense tapir]

example?

[dense tapir]

what is your neg prompt?

People really need to concentrate on their negs far more than they do

sometimes they can't even save it

[leaden night]

one leg,Amateur,Low rated,Phone,Wedding,Frame,Painting,tumblr,watermark,signature, hat, mangled fingers with (5 fingers:1.6), in prompt

[dense tapir]

I take it this is 1.5?

[leaden night]

yes

[dense tapir]

grab bad_prompt and v2 embedding

[leaden night]

ill google that up here in a bit

havent done embedding

is 2.1 just a model download?

[dense tapir]

stick one or the other in with your neg prompt

yes

I found v2 not as good when I use it

[leaden night]

so whats with all the models on civit.ai specifying 1.5?

if 2.1 is ust a different model itself

[dense tapir]

1.5 is easier to train and 2.x is a sob to train. I hate it

as far as end user is concerned it is just another download but to train the TE is a mofo. As an end user the clip is different so what works in 1.5 no longer does in 2.x

I honestly do not expect SDXL to be any easier to train and I know it will not be if they stick to the same CLIP and/or do not give us new training tools made specifically for it.

[livid ermine]

hmmmm

[dense tapir]

TE is the textual encoder and it fights you when you train. Immediately the same dataset in 1.5 just works and in 2.x forget it

[livid ermine]

Doesn't look realistic 😦

[dense tapir]

I no longer can even attempt to train as Google banned gradio

[livid ermine]

Why isn't that everybody's #1 focus?

[dense tapir]

What doesm

[leaden night]

i thought it looked pretty real

[dense tapir]

t look realistic about it?

[livid ermine]

Xetrov: It kinda looks handdrawn

[leaden night]

oh, well the artist its modeled after does lol

[livid ermine]

Shouldn't it be modelled after real photos?

I use "https://huggingface.co/SG161222/Realistic_Vision_V2.0/blob/main/Realistic_Vision_V2.0.ckpt"

prompt: "a #{ artist } man, detailed skin, perfect face, fashion photography, photo realistic, ultra realistic, 20 megapixel, 35mm analog camera, background is #{ ["party", "concert", "festival", "nightclub", "pool party", "rave party", "sunny beach", "misty mountain", "forest", "luxury aparment", "penthouse suite"].sample } with #{ ["friends", "family", "girlfriend", "dog", "a group of beautiful girls", "a bunch of random people"].sample } ",

[livid ermine]

negative_prompt: "(deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime:1.4), text, close up, cropped, out of frame, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck"

That's the best I got so far.

[dense tapir]

Try the first .pt and see what it does for you Xetrov

[livid ermine]

General Awareness: His photos don't look real

www.natgeo.com photos look real

[dense tapir]

I really don't give three shits if they look real or not he asked and I just gave him what might help with his issue.

[livid ermine]

btw gotta re-upload, was missing the instruction text:

Oh we're moody today too, huh?

I don't want no trouble, General!

[dense tapir]

I really don't care because looking real isn't what his issue was about. Was from an earlier question he asked.

[livid ermine]

so whats all the fuss about

im just saying it doesnt look realistic, maybe it'll make him think of ways to make it look more realistic

[dense tapir]

Does he wish for it to look more realistic?

[livid ermine]

i don't know, let's ask him

my man, do you wish your photos looked more real or more fake?

i'm out

[dense tapir]

Alright @leaden night back to what we were talking about. There are the two embeddings I mentioned to possibly help with your hands/fingers issue

[leaden night]

I'll take a peep. Appreciate the insight

[wispy ether]

@dense tapirare bloody images allowed

i got a complicated good one

[dense tapir]

Should be

I never managed a good bloody one in SD

[wispy ether]

[dense tapir]

Interesting

[wispy ether]

[dense tapir]

Still, I wish SD could do it better and more realistically

[crimson fern]

guys how do put my vae slider next to my model picker like this

[wispy ether]

@dense tapirthe feeling when u see the pre image

and the result looks ugly

[wispy ether]

wtf

[wispy ether]

that looks kinda ew

[dense tapir]

My 1060 is so slow

[wispy ether]

get gud

https://tenor.com/view/doctor-strange-the-ancient-one-portal-marvel-gif-16666856

crying

[wispy ether]

whats ur stuff

prompt n stuff

[dense tapir]

284 words for my neg

[wispy ether]

can you send a image that contains them

🥂

the expection

.

[wispy ether]

why doesnt yours contain metadata

@dense tapir

[wispy ether]

give prompt

[wispy ether]

@dense tapir

[dense tapir]

That is 2.1 and with some of my embeddings

and loras

[wispy ether]

still

promopt

[dense tapir]

@marsh lily Heyas

[wispy ether]

why tho

there was nothing uncensored

[wispy spindle]

I deleted it because it goes against the server rules, focusing on an entirely nude and bloodied character

[wispy ether]

where was the nude

it didnt show a single detail

and not nude focused tho

[wispy spindle]

Nude= no clothes
The character had none on. Thus, pretty nude focused.
Be mindful of this when posting is all! Thank you

[wispy ether]

the nude is random

btw how do i set priority of a detail

quality loss

ceres typing for 8 minutes compilation

9

nice

[wispy ether]

you were typing for 10minutes+

[dense tapir]

I had that error a while back and trouble ticketed on my 1060

[dense tapir]

I forget but it is not an AMD, or Nvidia, issue

16 it/s on 512 or 768?

try 768

now is that just raw or with embeddings, and loras?

[wispy ether]

[dense tapir]

My 1060 at 768x768 and using loras/embeddings

[wispy ether]

whats loras

[dense tapir]

loras, locons, lohas, oh, my.

[wispy ether]

tell me or ill hunt your dog in his dreams

[dense tapir]

they are like little models

[wispy ether]

imagine you see that while being outside

[dense tapir]

@wispy etherhttps://rentry.org/lora_train

[wispy ether]

who

[dense tapir]

With nothing just a prompt about 22s

k

[wispy ether]

3060

all

[dense tapir]

DDIM 20

[wispy ether]

2000x2000

batch size 8 is max

what seed

hm

i stopped

ye

[dense tapir]

hmmmm

[wispy ether]

@dense tapiris ur aspect ratio 16:9?

[wispy ether]

got a fancy wallpaper for u (720p)

[dense tapir]

16:10

[dense tapir]

As you see that was toad on a log and no neg or anything else

my 2s/it is your 5.25it/s. That is a really big difference

It does

DDIM should be faster

[wispy ether]

overclock gpu by 200mhz

jk

[dense tapir]

already is

[wispy ether]

bru

[dense tapir]

stock is shit, I mean really slow

[wispy ether]

pass

@dense tapircan u give me a good prompt

i wanna try making a good image

[dense tapir]

toad on a log

you saw it

[wispy ether]

not that

the fire one

LOL

[wispy nest]

Is she deepthroating fire?

Wow I need a woman like that

[wispy nest]

WOW

[wispy nest]

ya i'm beautiful

WAIT

CAN SOMEONE IMG TO IMG THAT

[wispy ether]

me

[wispy nest]

I WANNA SEE WHAT HAPPENS

[wispy ether]

i was gonna do that

[wispy nest]

I have the full image

wait a sec

[dense tapir]

Here is a toad on a log with all my juice.

[wispy nest]

WTF

[dense tapir]

Sometimes

[wispy ether]

excpectation

goofy ahh result

[wispy nest]

She's singing songs now

[dense tapir]

No speed gain so yeah, a bottle neck and I know where that bottleneck is coming from too.

[wispy ether]

im going to be funny

hm

LMAOAOAOA

[dense tapir]

docker

docker has a performance hit

are you using rocm 5.5.0 beta?

[dense tapir]

yes, for now

when done expect faster and 5.6.0 even faster. then they should be done

directml is even worse

[wispy ether]

NO

[wispy nest]

BRO

Damn she fine

it's so goofy but great

[dense tapir]

I have never had so many people tell me to not go with AMD and Nvidia is the best so shut up and spend your money as I do in this SD shit.

[wispy nest]

Why spend money when you can

s t e a l

[dense tapir]

I am interested in the training mostly

[wispy nest]

nvidia is the best true

[dense tapir]

Colab is no more as Google banned gradio for free accounts

yes, I am screwed

[wispy nest]

woah

[dense tapir]

worst case I turn my back on SD best case I get an AMD since Jensen is never going to lower the price of the cards where they properly should be.

how was your inference compared with 4080?

[wispy nest]

yo guys why nep muted for an hour

what did he do?

[dense tapir]

about right

That was in Jan though

neck and neck with the 3090ti

same

though these numbers will go up as ROCm matures and gets the updates

the one thing I will miss with AMD is the lack of tensor cores. I donot have them being on Pascal, but a lot of stuff is moving to TensorRT

Tensor cores is a game changer, ngl.

oh, for SD you ditched xformers and used its replacement?

7900

7950 is due now btw

I checked and I saw no difference so I am not sure

meh

what are your command_args?

@marsh lily

oh, speed gain incoming

--opt-sdp-attention

that takes the place of --xformers

needs pytorch2

that is non determinsitic as is xformers but there is a slightly slower version that is deterministic.

wtf?

lol

I am wondering how much of a hit the deterministic one is because if not much of one I will use that for sure and for training ALWAYS use it as xformers was problematic.

That is supposed to be on the comand_args in automatic1111 webui-user.bat

set COMMANDLINE_ARGS=--theme dark --medvram --opt-sdp-attention --disable-safe-unpickle --port 9000 --api --opt-channelslast --always-batch-cond-uncond

I can't use it because pytorch2 for my low life card is slower AND I can't even get it to compile now

for current gen and 3090 it speeds it up

Your it/s is a tad too slow so I knew you were not using it

No idea but I know if I get an AMD I am not running it in windows

I already have ubuntu ready

(dual boot)

I run automatic1111 there and it loads so much faster than on windows

FC my first Linux until 4

@marsh lily https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Optimizations

AMD users get a performance increase but not a single mention on how they run it. I suspect on Windows since 5.4.3 is now on Windows and these are 6700 and 6800 cards which is what it supports.

gn

[ripe cedar]

Did a little non-scientific testing. It seems like the Anime Line-art CN is better at backgrounds than the regular Line-art CN and Scribble. This is because it doesn't add extra doors/windows/etc in empty wall areas. However, Scribble makes better characters from an inputted line-art because the eyes are consistent better, less extra noses, mouth lines up better according to the mouth line, etc.

[wispy ether]

[wispy ether]

vvenom dog

[wispy ether]

@dense tapir

[leaden night]

how can i get images like this sharper?

[wispy nest]

what model? cool photo

[sinful mirage]

fed gpt4 some info on stable diffusion (only about 10000 total words), this was one of the first it made with no feedback from me yet, off to a good start

[smoky oak]

My day has been thoroughly ruined, so I decided I'd pop in here and see whats up

@dense tapir Any new info on AMD's GPU performance?

@glossy herald Hows the model coming?

[hasty nova]

I'm trying to work on a lora for 2.1 and this is the only good image I have gotten out of it so far.

starting to get better I spose

[dense tapir]

Due to docker there is a performance hit but looking good so far. "With euler_a, 512x512, I got around 19-20 it/s. So roughly 4-5 it/s ahead of my XTX." When docker is removed that will go up about 3-6it/s because docker can be substantial. Funny thing is his numbers are without xformers (due to AMD) and no optimizations at all. Due to docker I had him try the optimization but it didn't do anything so seems he either put it in wrong or, as he said, docker could be ignoring it.

[smoky oak]

So what I am reading is AMD is doing promisingly good, especially for no optimizations?

[dense tapir]

@smoky oakso about 15it/s. for 768x768 it was around 5.3it/s. We need it out from under docker and able to use that sdp optimization

yes

Up next will be to see how it does for training

@smoky oak Update regarding the 7900 XTX. Inference is fine, around 16 it/s. I couldn't get the training to work, mostly because of what I assume is a bug with the ROCm fork of Triton that's currently in development (https://github.com/ROCmSoftwarePlatform/triton). My expectations are rather high, though, considering the fact that the Triton fork even exists (forked by AMD themselves no less), so I'm pretty certain training will be feasible soon™. But maybe hold off on buying the card for now unless you're patient and want to gamble. For all we know these issues are what's holding up the proper release of 5.5.

What I am seeing for no optimizations and no Triton I am thrilled.

7950XTX is due out now too.

Never will be as good as a 4090 but close enough for me.

[sinful mirage]

hands :/

[hard wagon]

Totally missed the pow 😀

[gaunt heron]

👀

[glossy herald]

online already ^^ I'm sleeping for now, but felt good when I checked before bed #1099123538927026237

[smoky oak]

PEENG

Howdy

[smoky oak]

Awesome, I am taking a small break from AI at the moment for music prod, but I will check it out soon

[glossy herald]

I'm opening an open dataset collection of the community's "best of", to complete the "SDArt : Complete collection" model that I train on all the PoW (currently 600+ pics)
I'll leave this open for like 2 weeks before adding around 50 to 100 pics to the dataset, taken from there, if anyone wants to play some : #1099245477007990825
for reference, corresponding model https://civitai.com/models/44230/sdart-complete-edition

[cunning jungle]

Room with full dark engry

[simple venture]

@tired basin this is what my I got from it

can you see where I inpainted ?

[tired basin]

I'm guessing to the left of the clouds in the centre of the image, but otherwise my 140p vision cant see it 😛

[simple venture]

nope I actually inpainted the small waves on the beach for some reason AI decided to do this:

I dont know what prompted it to just drop a gem in there

[pallid snow]

((realistic)), ((warm)), ((earth tones)), ((impressionistic)), ((portrait)), ((historical)), ((high quality)), ((oil on canvas)), ((old man walking in the warm sunlight down a cobbled street of old stone buildings in a historical city)), scenic, masterpiece

[crystal delta]

@pallid snow

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!

[simple venture]

generating it rn

@pallid snow whatcha think

[wispy ether]

what

[wispy ether]

[silk tendon]

Requesting bigger ice cream

[pallid snow]

An old man of black origin sits cross-legged in the middle of a lush forest clearing. He has a serene expression on his face and wears a white robe with intricate gold embroidery. His hair is long and gray, and his beard reaches down to his chest. The sun rays filter through the branches and leaves, casting a warm glow on the old man's face and making the golden thread on his robe shimmer. The image is an oil painting on canvas, using a wide-angle lens to capture the vastness of the forest

[blissful wren]

made with https://civitai.com/models/46294

https://civitai.com/models/46294 ^ this model too

[blissful fern]

Hello how is everyone

[glossy herald]

hello there

[glossy herald]

DJ battle

[dry crow]

xD idk

[sinful mirage]

does anyone know of a script/extension that'll apply masks and inpainting to small regions automatically over the course of like 20 runs, eventually transforming the whole image? May need to make the mask files then run a batch inpaint? But id want it to feed the img back into itself each time. Looking to make images sorta like these but much quicker

[wispy ether]

[smoky oak]

naise :>

@wispy etherWould you mind if I give a suggestion? Just for future gifs

[blissful wren]

hey dude havent seen you in ages! its captainzero

https://civitai.com/models/46294/diffusion-brush-everything-sfw-nsfw-all-purpose-checkpoint-nuclear-diffusion-anime-hybrid

https://civitai.com/models/46294/diffusion-brush-everything-sfw-nsfw-all-purpose-checkpoint-nuclear-diffusion-anime-hybrid

[wispy ether]

not high fps ik

im currently crying

[smoky oak]

Oh not that, I was gonna suggest using a VAE to help with the really washed out colors/contrast

A lot of Anime models seem to have really drag colors when not paired with a VAE

[wispy ether]

how do i do tat

[smoky oak]

They make a huge difference as I am sure you can see

[wispy ether]

how do i

@smoky oak

ok

[smoky oak]

They are a feature included in A1111, you can find documentation on how to use them online depending on what you are doing

[wispy ether]

no

idk what vea is

[smoky oak]

You can search for documentation on how to use them online. There are different VAE's for different models, and a few different ways to go about applying them

[wispy ether]

@smoky oakis that right

[smoky oak]

If you google/research how they work, you will find out better than I can currently explain

[wispy ether]

you didnt try explaining it

[smoky oak]

You didn't try googling it.

[wispy ether]

i did.

on the website i found one

no

@amber leaf

i dont care

@dry crow

[smoky oak]

What happened to the mod mail react?

[wispy ether]

i dont need u here either

leave

L

[smoky oak]

@glossy herald Are you on by chance?

[wispy ether]

@shy coral

all are afk

lamo

lameo

lemao

[smoky oak]

The modmail Emote seems to be gone

otherwise we could have gotten the message deleted immediately

[dry crow]

hey

[smoky oak]

Scam link above

[glossy herald]

done

[wispy ether]

gone

[smoky oak]

Its still there for me

[dry crow]

thanks! its deletet

[wispy ether]

its not

[smoky oak]

For whatever reason I can still see it

ok, gone now

weird

[glossy herald]

bots takes time it seems

[wispy ether]

betterdisco~

[smoky oak]

What happened to the mod mail react?

[glossy herald]

but yeah, 1AM, not reactive to the second

[smoky oak]

I don't use better discord, thats against TOS

[wispy ether]

everyone breaks tos

without noticing

[glossy herald]

mod mail react ? ⚠️ react you mean ? also same as Applications => warn staff ?
Should be there. Sometimes, discord removes it from me for 24h, no idea why, while it's working for others

[smoky oak]

I remember there being a specific emote called modmail from this server that was like an envelope with an exclamation mark that I was told to react with cause it hides the message for review

[glossy herald]

(can it stop though lol, it doesn't at all tonight lol x) )

[glossy herald]

I'll be honest, I don't remember any

[smoky oak]

hmm...

[glossy herald]

but once again, 1AM, sleepy, not the best of guizmus rn

[smoky oak]

Understandable, thank you

[wispy ether]

smash or pass

[glossy herald]

both

[smoky oak]

I did use the ⚠️ emote

[wispy ether]

hand gone

[glossy herald]

yep, I have it in there. it's just that I was off and didn't get the notification until 10 minutes later

[wispy ether]

nvm

[smoky oak]

Oh no, I am not saying its not working or anything, just wanted to make sure that it DID

Like make sure that was the right move

[glossy herald]

and you can rest assure, it did 🙂

[smoky oak]

Alright cool, thanks

[glossy herald]

yeah

usually, it's a direct ping when I'm on for me

[smoky oak]

I saw it instantly disappear so I wasn't sure if it was the bot, or if they blocked me

[glossy herald]

the bot removes the emote yes

[wispy ether]

@glossy heraldcan you assist me with fp16 in #🤝｜tech-support

[smoky oak]

alright, cool, thanks for the info. I'll make sure to save that as a favorite

[glossy herald]

so you don't have your name displayed basically

[wispy ether]

fp32 and fp16 are same speed for me

[glossy herald]

no, I still don't know more than yesterday when you asked me on this, sorry Nep

[wispy ether]

that wasnt that

fp16 is always supposed to be faster on rtx cards

[smoky oak]

More than likely your speed is being limited by CPU in that case

the speed difference would be hidden if your CPU can't let it show

[wispy ether]

that doesnt have to do with the cpu

fp16 is half precision

[smoky oak]

Your CPU feeding the info to the GPU is still a bottleneck at some point

if you can't fully deliver the full info, then you won't see a difference

[wispy ether]

5700x+3060

[glossy herald]

yeah I'm still going back to bed, sorry. even if I was clear, I don't know those things and take what works for picture making, or what is optimal for training

[smoky oak]

Then I am not sure in that case, that GPU isn't particularly needy, however it does have a constricted bus width, so I wonder if that has something to do with it

[wispy ether]

i bet you dont even know what fp16 means

https://www.intel.com/content/www/us/en/developer/articles/technical/should-i-choose-fp16-or-fp32-for-my-deep-learning-model.html

not intel tho

[smoky oak]

You don't need to result to insulting my intelligence. I know what FP 16 and 32 are, I also know that there are far more factors than just GPU hardware in the equation of speed

[wispy ether]

no

my pc doesnt bottleneck

ive chosen what to put together to not

what cpu and gpu

which bottlenecks at certain resolution and which not

[smoky oak]

There are still far more factors than that

[wispy ether]

[smoky oak]

For example, bus width is what is currently making the 4070ti and 4070 frankly bad GPU's for image gen

[wispy ether]

i was about to buy one

lol

[smoky oak]

The bus width of them severely limit memory speed, which makes them very constricted

for example, current tests of the 4070 are showing around 3060ti performance (within 5% perf difference in AI)

And the 4070ti is closer to a 3070

[wispy ether]

next gen ai cores tho

[smoky oak]

They only pull ahead when more than 8GB VRAM is needed

[smoky oak]

Yeah, but they are useless when paired with constricted VRAM that can't keep up

which is exactly what happened

a 192 bit bus is way too small for GPU's that powerful

[wispy ether]

?

[smoky oak]

You were insulting my intelligence, but don't even know about how memory busses and bandwidth work?

[wispy ether]

bus doesnt have to do anything with intelligence

[smoky oak]

Memory bandwidth is far more important than core performance

[wispy ether]

people are learning

[smoky oak]

For future reference, the 4070 and 4070ti are too bottlenecked to be a good purchase for AI. You can get a 3080 12GB for not much more and its dramatically faster

[wispy ether]

fp16 still has to work

[wispy ether]

not in games

[smoky oak]

In games no, but I said in AI

well actually

the 3080 is usually faster than the 4070 in games as well

[wispy ether]

i do both tho

[smoky oak]

only point where it has a real edge is in DLSS3, but less than a dozen games support that

so generally speaking, the 3080 10GB/12GB is a better buy than the 4070 for around the same price, which has been the findings of tech reports and such

[smoky oak]

and that lead gets way bigger in AI

[wispy ether]

dlss swapper