#💬｜general-chat

1 messages · Page 42 of 1

past edge
#

But gradio is a godsend

near silo
#

I don't really know what that means

past edge
#

Command line

near silo
past edge
#

Well the only alternative is command line so... I would never even consider switching

#

I can guarantee whatever issue ur having is not gradio but the repo ur using

#

I've never honestly heard anyone complain about gradio here

near silo
#

I'm using a1111

#

The public server it creates is very very buggy and annoying through gradio

past edge
#

Thats strange

near silo
#

The last 10 or so times it's made a server, I keep getting error 504, bad gateway error

past edge
#

I don't think that's Gradio

#

I have seen people having that error but I think it's an easy fix but since I've never had it, I wouldn't know how to fix it. Sorry.

near silo
#

And all the other times it's "worked", it would randomly disconnect and lead to "there is no interface running at this time"

#

If it's a problem on my end, then it's untraceable, IDK what to say

#

I wonder if it's because of our new internet. Maybe something on the router is preventing the launching or something

potent fox
#

Another hint for users of ControlNet. It seams to be very sensitive to shadows in the image. You can only make such adjustments in a 3D pose tool. It has so much impact.

#

Of course, the openpose model is not sensitive to shadows, since it only detects the pose of a character.

last merlin
#

In my local gradio works well and predictable

fervent thunder
near silo
brisk mesa
#

After searching for an hour I've found that Gradio is just down.....

near silo
#

Ok, so it's not just me

#

I'm getting error 502's, gateway timeout

shadow birch
#

Good evening everyone, I know that someone must have already mentioned this here, but recently they updated the automatic 1111 code, now when we install it on Google Colab, the public link to access the app does not appear. how are you doing now?

brisk mesa
#

yeah I'm getting the same thing. Someone suggested switching to LocalTunnel. I'm not sure how that works, completely.

near silo
last merlin
#

Why not hosting on local

brisk mesa
#

504, you.

near silo
#

I wanna find an alternative way to host my SD permanently on my own PC

#

Where my PC just functions as the web server

brisk mesa
#

I'm using paperspace. My pc is trash. And yeah getting that 502 bad gateway.

hollow jetty
last merlin
#

🤔 is becoz of gradio.live?

vestal dew
#

This is why the Cloud isn't a good idea - it gets taken away

glossy shoal
#

Yo does anyone know what diffusion_pytorch_model.bin is? I got them from making LoRAs and they take up a of space

#

If I don’t need them I want to go ahead and delete them

brisk mesa
#

but my stuff is working...odd

#

been trying to get it to work for the last 4 hours

#

anyone else?

deft crater
#

oi

vestal dew
#

Now it's that time of generating images when because Im only getting close to what I want not exactly I try a different model. This usually results in disaster.

#

it did

waxen mirage
#

Is gradio.live working again?

leaden mural
#

Is the best anime model still orangemix, or did something overthrew that when LORA became a thing?

spice quest
astral goblet
#

nice. while i'm image genning using my gpu, i can still play some cheaper games on my intel integrated graphics

pliant sequoia
#

hello everyone, I need some resources for mac. I couldn't use stable diffusion animation with my mac, is there any good resources? thank you

lament steeple
#

Data aquisition takes awhile waits

eternal oar
#

any good prompts?

lament steeple
#

"Two pigs fell in the mud, 3d render, high def, 8k resolution, trending on artstation"

#

That AI Prompt Bible thing popped up on my face book feed. I guess they deleted my comment because I said they were selling google search results

near silo
#

I really really wanna be able to run my PC as a constantly available SD portal that doesn't rely on another company to work

vestal dew
lament steeple
#

poke a hole in your firewall

#

using --share will only give you a 24 hour link, you have to port forward if you want to make it perminant

near silo
#

do you have any info on how I could do that? I really wanna be able to just always access it, and not have to deal with the new gradio links and bugs all the time

near silo
eternal oar
#

also is there any realistic Lora or Embeddings?

#

becuase mostly all of them are nud*s

near silo
eternal oar
near silo
lament steeple
#

go into your router settings and go to port forwarding. send port 7680 on the router to port 7680 on your system running automatic. Do NOT forget to set a username and password or the world will be using your resources

#

port 7860 tcp. forgot to mention that part. you can do tcp/udp and it will work, but just doing udp won't work. you have to have the tcp

#

if you want, you can IM me when you get into your router settingss and I can talk you through the settingss

near silo
lament steeple
#

Yes. Jusst need to know your router make and model

near silo
#

I am always very worried I will mess this stuff up

near silo
lament steeple
#

I have mine set up so I can access from my phone while I'm away 🙂

#

so It

#

so it is something I have actually done to

#

too*

near silo
#

fantabulous

#

hopefully that will get rid of all of the bugs that come with gradio (specifically errors and timeouts when doing really big batches or high res gens)

lament steeple
#

It helps

#

you may want to look into dynamic dns if you don't have a static ip address too.

near silo
#

IIRC my IPS changes

eternal oar
#

Also, how to use stable diffusion's API? I am trying to convert it into a Discord Bot where my friends can access. I do not want people going into the website it created publicly because they can change the settings.

lament steeple
#

Then you will want dynamic dns. Your router should have a setting for that too. You just need to tie it into a service like noip

near silo
#

so, does what I am gonna do only allow me to access it when I am on this network, or will it work from anywhere with the correct IP?

near silo
lament steeple
#

I have it set up on a desktop in my office, I'm using the api from my laptop right now, I can reach the wep page from my laptop or my phone

near silo
#

sorry, not sure if that answers what i asked or not 😅

#

Would I be able to give the link to a friend and they could use it as well?

#

I have a few friends that use this a lot, but they always have issues with gradio

lament steeple
#

the answer is yes

eternal oar
#

yes, the api isnt working for me, even after i used --api

near silo
lament steeple
#

there is also a short api tutorial in the wiki

near silo
#

NOOO

#

MY NITRO ENDED

#

dies

#

Those scammers just HAD to steal my card 2 days before the billing cycle ._.

lament steeple
eternal oar
lament steeple
#

Nooooo. I had to write python code to access the api

#

for a discord bot, that's in javacript, right?

eternal oar
#

No, im trying to make a python one because this project is almost made of full python

lament steeple
#

basically, you access the url via get sending your arguements as json and it kicks back json that includes your image as base64 encoded text

eternal oar
#

Yes i did that but then i got mad and delete it all

#

because it keeps sending nothing

lament steeple
#

you testing on the same computer or is it not working in discord?

#

if it's discord, see above discussion about punching a hole in your firewall

eternal oar
#

same computer, because i often have problems in my internet

#

which limits my access to coding

lament steeple
#

ok I'm going to send you the link for the api tutorial. There is something I want you to do with your code that isn't listed there though. hold on

eternal oar
#

okay

lament steeple
#

in the request.post line, add the perameter timeout=30

#

and after it add

#

print(response)

eternal oar
#

ok

#

thanks

lament steeple
#

That way if you get an error, it will tell you what it is

#

if you get a wall of text, it worked

eternal oar
#

k i will tell you after i run this code 😄

lament steeple
#

OK gl

near silo
#

This router is SOOO OP lmao

eternal oar
#

<Response [200]>

eternal oar
near silo
#

Through 6 walls and over 80 feet away, I still get a speed of 480/360 on our 500/500 5GHz network

lament steeple
#

change print(reponse) to print(response.json())

eternal oar
#

ok

#

it sent alot of text

#

is that working?

lament steeple
#

and if you pasted the code, it should have saved your image

#

ye

#

ye

#

yes

eternal oar
#

AYEEEEEE thanks bud

near silo
#

alright, I have the router setup

lament steeple
#

kicks the "s" key

#

did you set a username and pasword?

#

and ddns?

near silo
#

I forgot 😅

#

I might have to factory reset it again

#

lemmie see

#

ok, i am in

lament steeple
#

set the user name and pasword for automatic1111 and ddnss for the router

near silo
#

sorry, I am very tired and not very smart, could you explain what you mean just a litttlleee more? 😅

lament steeple
#

then when you have that set up, pm me the url. keep the un and pw to your self. I'm just going to make sure you have access from outside your network

#

ok in your webui-user.bat you need to add a command line arg

near silo
#

ok, navigating

#

ok, I'm there

lament steeple
#

--gradio-auth USERNAME:PASSWORD

#

that keeps the world from using your SD install and bogging down your gpu

near silo
#

ok, nice

#

so for example --gradio-auth Sytan:test123?

#

username is Sytan, pass is test123?

lament steeple
#

yes

near silo
#

ok, cool

#

where to next?

boreal maple
#

Sorry to interject

lament steeple
fervent thunder
#

Is there an easy way to save all gui txt2img settings?

boreal maple
#

But I saw stability making an announcement of releasing something on Sunday

lament steeple
boreal maple
#

Was there anything?

lament steeple
#

I missed it

near silo
lament steeple
#

Yes, but you need to pick a name and sign up. it will give you a perminant name for your network

#

from outside

near silo
#

oh, this is gonna give me ads, and won't work on my phone?

lament steeple
#

No ads. I was using them until they dropped me for not updating (static ip)

near silo
#

interesting...

#

I was going off their breakdown

lament steeple
#

I've never gotten an ad from them... maybe in email

near silo
#

ah, I see

#

alrighty

lament steeple
#

there are other ddns services out there. they just were the ones I was using

#

but because you have a dynamic ip, you need a way to get to it

near silo
#

alright, sounds good

lament steeple
#

then when you are set up with ddnss, you need to set it up in your router. There should be a setting for it

near silo
#

ok, I am in the port forwarding section

#

@lament steepleDo you know what IP binding is?

#

would that help me?

lament steeple
#

not off hand

#

what brand is your router?

near silo
#

Its an ASUS

#

ASUS RT-AX82U to be specific

lament steeple
#

binding is for when you have more than one network connection

near silo
#

ah, alright

#

@lament steepleoh, I can enter a manual IP instead of DHCP

lament steeple
#

Check with your isp before you do that. You can break your internet connection

near silo
#

ah, alright

#

I'll just use no ip for now then

#

alright, sorry, where were we?

lament steeple
#

The only reason I have a static is because my isp set me up with a static

#

you nave noip set up in your router?

near silo
#

I am downloading the DUC

lament steeple
#

ok

#

let me know when everything is setup and send me the url for your automatic and I will try to connect and make sure it works from outsside

near silo
lament steeple
#

Yeah. Not driving 3 states over unless my mom really needs help

near silo
#

oh god haha

hidden sluice
#

hello chat

near silo
#

@lament steepleok, truth be told, I am not too sure what IP I am supposed to be forwarding, and I am not sure what type to use. Sorry, i can be really useless sometimes, but I am a fast learner!

#

oh my dumb ass, nevermind

#

the video says later lol

lament steeple
#

hehe ok. You work on that. I have to make my room mate a bagal and step outside

near silo
#

sounds good, though some of these things hes saying are not showing up for me

#

here's hoping

lament steeple
#

I'll try to be as quick as possible in case you get stuck

near silo
#

thank you <3

#

I got stuck lmao

unique patio
#

Chat gpt is an absolute BEAST when you want to fill in gaps or detail your prompt. It literally ampifies your prompt by 10-20 times the clarity before you actually ask it, its so good

lament steeple
near silo
lament steeple
#

I know nothing about that

#

lemme try something

near silo
#

this website tells me my IP

#

But I don't know what I am supposed to do with my IP tho

hidden sluice
elder cradle
#

Holy crap, you... disclosed ur IP address!?

#

@hidden sluice delete that message now, your IP is compromised!

hidden sluice
elder cradle
#

This is your PUBLIC IP ADRESS

hidden sluice
#

cool

lament steeple
hidden sluice
lament steeple
#

Because you don't want the world to have your ip address

elder cradle
#

Your ip address is literally exposed

hidden sluice
#

why not

elder cradle
#

So you want people to trace ur location

hidden sluice
#

yes

#

unfortunately it only works under a 30 mile radius

warm junco
#

One Router restart and he has a new one

lament steeple
#

h i don't know. my town isn't 30 miles big and it will tell people where I live

hidden sluice
warm junco
little yoke
unreal eagle
#

how do i make chatgpt amplify my prompter?

vast ingot
# little yoke How do you format your question(s) to it when you want help with your prompts?

here is the one I use to do the same thing. you just need to add a prompt idea at the end after the word "IDEA : "
It may not be the same technique as Army1123 was talking about, no idea

Stable Diffusion is an AI art generation model similar to DALLE-2.
Below is a list of prompts that can be used to generate images with Stable Diffusion:

  • portait of a homer simpson archer shooting arrow at forest monster, front game card, drark, marvel comics, dark, intricate, highly detailed, smooth, artstation, digital illustration by ruan jia and mandy jurgens and artgerm and wayne barlowe and greg rutkowski and zdislav beksinski
  • pirate, concept art, deep focus, fantasy, intricate, highly detailed, digital painting, artstation, matte, sharp focus, illustration, art by magali villeneuve, chippy, ryan yee, rk post, clint cearley, daniel ljunggren, zoltan boros, gabor szikszai, howard lyon, steve argyle, winona nelson
  • ghost inside a hunted room, art by lois van baarle and loish and ross tran and rossdraws and sam yang and samdoesarts and artgerm, digital art, highly detailed, intricate, sharp focus, Trending on Artstation HQ, deviantart, unreal engine 5, 4K UHD image
  • red dead redemption 2, cinematic view, epic sky, detailed, concept art, low angle, high detail, warm lighting, volumetric, godrays, vivid, beautiful, trending on artstation, by jordan grimmer, huge scene, grass, art greg rutkowski
  • a fantasy style portrait painting of rachel lane / alison brie hybrid in the style of francois boucher oil painting unreal 5 daz. rpg portrait, extremely detailed artgerm greg rutkowski alphonse mucha greg hildebrandt tim hildebrandt
  • athena, greek goddess, claudia black, art by artgerm and greg rutkowski and magali villeneuve, bronze greek armor, owl crown, d & d, fantasy, intricate, portrait, highly detailed, headshot, digital painting, trending on artstation, concept art, sharp focus, illustration
  • closeup portrait shot of a large strong female biomechanic woman in a scenic scifi environment, intricate, elegant, highly detailed, centered, digital painting, artstation, concept art, smooth, sharp focus, warframe, illustration, thomas kinkade, tomasz alen kopera, peter mohrbacher, donato giancola, leyendecker, boris vallejo
  • ultra realistic illustration of steve urkle as the hulk, intricate, elegant, highly detailed, digital painting, artstation, concept art, smooth, sharp focus, illustration, art by artgerm and greg rutkowski and alphonse mucha

I want you to write me a list of detailed prompts exactly about the idea written after IDEA. Follow the structure of the example prompts. This means a very short description of the scene, followed by modifiers divided by commas to alter the mood, style, lighting, and more.

IDEA:

little yoke
#

That’s very clever

lilac chasm
#

Hi guys, I am counting the time consumed by SD to generate images. Do you know whether SD must render images every time it is inference? (Because when I use WebUI, a picture will be rendered at each step, maybe rendering pictures is time-consuming?)

fervent thunder
lilac chasm
unreal eagle
#

very clever stuff

#

it makes some good prompts

little yoke
golden tartan
#

Wasnt Emad eluding to something be released on Sunday on Twitter

#

Or did I eat too many space cookies

undone wraith
#

Hello people. One question. I saw that the bots are offline. But i did not see in the announcement if they will get back or no more. We can still generate pictures?

sleek osprey
#

How much faster would inference be if I had a a100 compared to a 3090

#

rented obvs lol

vast ingot
sleek osprey
#

saw that already, pretty sure it doesnt mention a a100

vestal dew
vestal dew
scarlet barn
#

Does anyone have a script to convert a standard stable diffusion model which is trained on custom dataset to an inpainting model, you help is appreciated

#

Rather than Automatic 111

sacred badge
#

hello,I want to use the original design image of my product with a white background, without any changes to the product's design (including style, color, details, etc.), and have an AI generate different background images of the product in various usage scenarios or angles based on my textual prompts. It is important that the product in the original image remains unchanged. Please contact me if you can fulfill this request.

surreal perchBOT
#
FAQ: How do I generate variations?

There’s no explicit way to generate image variations, but you can “DIY” it a bit, by re-running the same seed with either slightly different prompt text or slightly different settings. Try shifting the CFG scale value by a few points, reordering the words in your prompt text, or adding another new keyword or two.

pallid flame
#

can anyone tell me how "Inpaint batch mask directory (required for inpaint batch processing only)" works i cant find any tutorials on youtube

daring verge
#

with all these LORA model keeps popping up, is it necessary to upgrade to 2.1?

mossy river
#

What are yalls thoughts on LDSR? Is it worth how long it takes?

stable pawn
#

@mossy river I often want to test different models and now I use just "Bunny" followed by a random word for the model I test, but what would you say is a good standard test prompt? The thing I want to test how safe/unsafe it is (sexually) and how it make clothe, hair, fur and what style it preferer.

#

I tested LDSR and was not happy, I think I get better result with Lanczos.

#

The thing I did not like in LDSR was that it could warp images a bit, and in a old photo I had and tried to enlarge it made a very good job, but it removed thin lines in a tree and turned a shape of a small girl in the background into what looked like a fire-hydrant.

#

But it is not useless, it can be good to add details in some images, so if you have the time and GPU power is is always a good option to test.

#

So I say it is not worth the wait if you want just a upscale that is close to the original, but it may be good for trying to add and restore in some images.

broken gull
#

Anyone here heard anything from Guizmus

vast ingot
#

yep

#

he is sad

#

hey @broken gull

#

I got hacked ^^

broken gull
placid comet
#

Wow. This is possibly the most correct prediction I've made in my entire life lol.

vast ingot
#

it won't stop for quite some time I think, this is still moving so fast, and new types of AI grow here and there. I'm really teased for what the mix of all those will be capable of

placid comet
#

Yeah. I made that prediction like two weeks before stable diffusion even released. It's surpassed even my optimistic expectations at the time tenfold.

#

With how much progress has been made, august 2022 feels like a lifetime ago lol

vast ingot
#

Now on making an AI that predicts the future

#

yeah, it's kind of magical already

#

even chatGPT, for SD, has been a great tool

#

very good at making prompts

warm junco
vast ingot
#

I'm not proud of it

#

I was contacted by a user here (be careful of PM guys) wanting help/feedback.
I listened, answered, and eventualy got a link to download a ... .exe
Even if virusTotal told me 1/52 with a very low alert on the 1, I thought I could take the risk.
Don't ask me why I though I could, I shouldn't have...
The rest is history

warm junco
#

Oh wow, hope you did reset your PC and changed the Passwords off all accounts

#

Also use 2fa whereever possible

vast ingot
#

but the virus took over my f2A

#

I'm still trying to get lots of things back

warm junco
#

Oh crap wtf

vast ingot
#

yeah...

#

even had to block banking stuff

warm junco
#

Make yourself a Keepass Database and use an 2fa APP now

vast ingot
#

I do have all those, and I did have

#

but I made the greatest error possible

#

I opened a .exe on my main computer

#

I'm a dumb dumb

#

what can I say

warm junco
#

Yea xD

vast ingot
#

I'm really not proud

#

I'm in contact with the different supports, it will take some time but I'll get back on tracks

warm junco
#

You should open such stuff only (if nessasary) on a Locked down Sandbox, there is also the Windows 10 Sandbox which wipes everything when its closed

vast ingot
#

that error made me switch to paper for password keeping

#

yep, I should have. I shoudl also have at least cut internet

#

I cut it directly when I saw what was happening, but like, before ? better

warm junco
#

Did it encrypted stuff?

vast ingot
#

I don't think so, but I didn't take any risk. I shut it down after cutting its internet, got on my other computer (that was off while I got infected, nice), and managed the other things, like making sure I got banned

#

since I was a mod it was dangerous for the server too

#

and all the other hacks that continued, since they had time to get to my email and could continue on their side

#

and I booted to factory reset directly, not taking any additional risk

warm junco
#

Okay good, hope you didnt loosed to much files at the reset

vast ingot
#

well

#

I wiped

warm junco
#

Also how did they get your login details in first place, if its all in keepass

vast ingot
#

I didn't try it

warm junco
#

Mostly they check the Browsers for cached logins

vast ingot
warm junco
#

Hmmm okay yea if you was logged in that time :/

#

And how did you saw whats happening?

#

Did the exe opened more scripts?

vast ingot
#

I saw discord disconect almost instantly

#

I cut it at that point

#

I didn't see that the hack was still going on until a few minutes later when the mails started pouring in

warm junco
#

Damn

vast ingot
#

yeah... well, it was a lesson in gullibility on my part. and first time I get hacked in more than 35 years feels like a good score still

#

in the end I lost quite some time, 25€ in Nitro, and a lot of hairs from stress

#

not too bad

warm junco
#

Yea if your banking stuff is save most is fine xD and if you get your primary mail back

#

But that stress wow, cant imagine

vast ingot
#

yeah. primary mail should be OK, it's my main microsoft account for more than 15 years and all my family is on it, I have all the proofs I could ever need

#

discord, I can't start getting it back until I get the mail

warm junco
#

Ah okay

vast ingot
#

I tried, but the hacker can just cancel the request from my mail (they did)

warm junco
#

:/

#

Do you use chrome ? And saved auto logins ?

vast ingot
#

yes and no

#

(yes to first, no to second)

warm junco
#

Okay so no Browser passwords

vast ingot
#

nope, not those at least

#

I still changed all the important stuff in case

warm junco
#

@vast ingot Hope you get your whole stuff back, hate it when nice people who try to help get victims of scammer kids

vast ingot
warm junco
#

Guys pls leave this discord out of political discussions

vast ingot
#

yes, this isn't the place for this

#

hey, thanks sunny 🙂

warm junco
vast ingot
#

there is #🌶｜off-topic for things like that, this has no place in the Stable Diffusion discussion, no link to AI, and politicied to a point attacking groups of people. We are not here for politics like that

uneven heath
#

Man easy.

#

Okay, good for you. Please stop thoughm

#

There are amazing political servers on discord, you can check them out on top.gg

last merlin
#

Taiwan number 1

#

China abuse human rights

#

😭

vast ingot
#

there are some, I'll ping I think

#

no but I do have google translate

last merlin
#

I din even want to translate 🤣

vast ingot
#

werll, I'm a currently powerless mod, so I try to get what's happening to report accordingly

#

like that

#

thanks Aether

tidal bough
#

Keep your discussion civil and within the rules everyone.

last merlin
#

And Taiwan number 1

vast ingot
#

still inapropriate. damn

last merlin
#

Wow I just say Taiwan number 1

vast ingot
#

can we just go back to being civil and not call people dicks ?

last merlin
#

Some uncivilized people

#

🤣

vast ingot
#

sometimes, I don't remember why I want to try after a while

#

the toolbox seems open, i'm missing a hammer here

last merlin
vast ingot
# last merlin Wow those hackers are extraordinary

I wouldn't say that... They had a good tool with them, but tbh, they could have done a lot more damage if they had wanted to. They were discord hackers, mainly after a discord account, and I can see they destroyed it quite fast, it shows as spammer already. But like, they managed to get hold on paypal technicaly, if they knew what they were doing.
I was the error there, everybody knows you don't do what I did. But yeah, sweet words and a nice person can make stupid happen

stable pawn
#

Question. If you use Stable Diffusion v1.5 (Euler a) and just write "erotic art", how many images do you think you need to create before you see a male? I am at image 120 now and it have generated more couches than males 😄

lethal venture
#

Hey guys, so I’ve been working on this consumer-friendly app that “turns your thoughts into selfies”. The goal here is to make it super un-technical for everyday people so that they can express themselves visually and share moods, memes, and ideas.

So I’m looking for people to join our closed beta and play around with it. If you're interested, feel free to dm me!

bleak matrix
#

Good morning, everyone!

#

How are we all today?

vast ingot
#

grrreat ! and you ?

pallid flame
#

hello

vast ingot
#

Hello Ryan The Gray

pallid flame
#

does anyone else missing openpose hand model

#

i dont find the model

vast ingot
#

I don't think I have seen a hand model to download on controlnet yet, but I could be wrong. I can't seem to find it either

cunning saddle
#

Hey everybody! I'm new here but I have only 1 question. Is it possible to use a Black/White (no greyscale) image (like an inPaint image) to generate random things ONLY into the White shape. Like a mask where we could ask Stable Diffusion to create something but only IN this area? And if so, HOW? Thank You SO much in advance!!!

pallid flame
#

there was a hand model with open pose before

#

i can see it in toturials

#

im pretty sure i had that few days ago

vast ingot
pallid flame
#

i will send a screen shot in general with images

vast ingot
vestal dew
cunning saddle
vestal dew
#

I have seen everything now - got an image with a hand with no palm, just fingers

vestal dew
vast ingot
pallid flame
#

do you know what Guidance End (T) and start does ?

warm junco
# pallid flame thanks

The open Pose Hand is a preprocessor and not a model To download,
it was available but i will check later if its still there

vast ingot
#

not yet, I'm just still playing with the different models to get it for now

pallid flame
pallid flame
warm junco
#

Make sure you run the latest extension version

pallid flame
warm junco
#

Okay 👍

cunning saddle
vast ingot
#

but mostly, there is a new tech in town

#

a new extension to try

#

something that will help you tremendously here

#

controlnet

cunning saddle
vast ingot
#

yes, adding this to your current inpaint will help a lot I think

#

it will preprocess your image, make an image out of it (different images depending on the mode you use), and use that image as conditionning for the image making process

#

even without mask, it would keep the shapes like they are in your original image for example

#

using a mask still stays nice, because without it, it would modify colors and details inside the currently masked area

cunning saddle
vast ingot
#

Have fun 🙂

pallid flame
#

i was wondering

#

what do you guys have for hardware

#

and laptops or pc

vast ingot
#

there are lots of types of that around, from 4GB up to 48GB VRAM you'll see screenshots floating.
Personally, I started on a laptop but was too limited for training, I mainly use a desktop with 3090TI, 24GB

pallid flame
#

thats a beast

#

cool

vast ingot
#

yes for sure, runs quite well

pallid flame
#

i have a pc but i have only a 1070 or 8gb

vast ingot
#

still can do things, and there is also dreamstudio and collab for some more free fast stuff

pallid flame
#

i am using local Stable diffusion fine but at a slower pace

#

but i am moving to italy in 6 month and i cant take my pc with me

vast ingot
#

😢

pallid flame
#

i was thinking of getting

#

a very good laptop

#

any suggestions ?

vast ingot
pallid flame
#

thank you

#

by the look i think a laptop with 3070 will work just fine

little yoke
vast ingot
#

I think aiming for 12GB currently is the most optimal, it lets you train at home on almost anything with some optimisation. After that, the faster the better but it all depends on budget, 8gb can also be the good point for you

pallid flame
#

what do you mean by training models ? im a little noob

vast ingot
#

putting your face into your SD model, to make it short

#

but so much more

little yoke
pallid flame
#

cool , i use something similar on deepfacelab

vast ingot
#

but collab works great for that too

pallid flame
#

by i dont want to train models

#

i just want to create videos as fast as possible

vast ingot
#

the more VRAM you have, the higher size pictures you can make, and/or the more pictures you can batch make at once. There is a gain there.

vestal dew
pallid flame
#

i know the bigger the better but budget is important for me

vast ingot
#

doing batches of 8 pictures at once on 24GB for example takes about half the time it would to make 8 times 1 picture in a row

#

yep, it's the middle ground you need to find

#

8 is great

vestal dew
#

right, but not everyone can afford a 24GB card

vast ingot
#

I know, was just putting the pieces out to understand

pallid flame
vestal dew
#

and?

vast ingot
#

new laptop incoming

vestal dew
#

you cant bring a computer over the border?

vast ingot
#

not a tower, harder ^^

#

well, takes space

pallid flame
#

there i will be moving around so much and i cant use my pc like that

#

i need to be mobile

late sluice
#

hey!
I am curious how stability use embeddings in production, do you use any database like weaviate, milvus, pinecone, etc. or just maybe postgres + pgvector? that would be very helpful 🙂

fervent thunder
#

Hi

carmine reef
#

hi, if i have a question, can i make it here?

hexed rivet
#

Hey, I am a student starting university society and I want to get in touch with stability AI on collaborations and potential projects within my school. We are based in London, does anyone know how best to go about this?

hexed rivet
vast ingot
hexed rivet
warm junco
sleek raven
#

guys

#

anyone know how much vram I am using?

#

I posted a pic on general with images I THINK i can do more vram

alpine mica
#

Who here can help me? On the search for a SD API for Cinema4D the one that was opensource is now a paid one and removed from GitHub. There is this other one but it only explains integration with blender. Can anyone help me? https://github.com/blender/cycles

violet grail
#

Will the dreambot come back?

vestal dew
#

no

lean laurel
astral goblet
#

i think it will make it's way back. i've seen officials alluding towards that

#

heres where i saw soemthing that caught my eye. while it "isn't active" hmm right?

chrome spindle
#

Whats the best channel to ask about training my own model?

astral goblet
sleek raven
#

How can I assist you user

stable pawn
#

I some time do lovely images, then the AI go berserk and my fairy scene is full of fire and flies... Oh, dyslexia "fireflies" is just one word.

ripe girder
#

where i can generate

wild prawn
#

why do people piss their pants when it comes to ai art

stable pawn
#

When I send a image to Extra to upscale with LDSR I see this in the command window, "Downsampling from [768, 512] to [384, 256]", why do it do that?

wild prawn
#

everyone seems to despise it

stable pawn
#

It is new, new thing will always kill art or music, for atist is always a artist and can not take any other job to get along.

#

I think it was during the 80th "Home tapeing kill music", but we still have music.

wild prawn
#

spend an hour arguing with some schizo shill if ai art is real art or not but it apparantely steals art from real artists

#

i mean literally people were going apeshit when photography was introduced and it was "not an real art because all you do is take a picture with one click"

stable pawn
#

Ai art do democratize art, some artists may get pushed off, but other AI-artists will take their place, and art will live on, just cheaper and give folk what they love.

#

No need to get crappy mass-produced art on IKEA now whan all can have their personal art.

wild prawn
#

yea

#

I mean look at ai art now when it only began

#

imagine what can it do in like 20+ years

#

not all people can draw and ai art is a great alternative to express their creativity and ideas

stable pawn
#

Folk are afraid more will cheat with AI, but we just need to learn to test knowledge of folk we hire rather than just put all out trust into a paper saying folk can something.

astral goblet
#

a lot of people who bait arguments online are just trolls and don't actually stand by what they're saying

#

its dangerous out there

vast ingot
#

yep, I can vouch for that

#

I didn't use to think like that, but man, being mod, I saw some disturbing things. trolls will be trolls

velvet plover
#

Where can I find some lists of good negative prompts for landscapes in SD 1.5? 🤔
I've seen a lot for 2.0 or 2.1 but haven't seen much of those for 1.5

lament steeple
#

Ha ha one of those "companies" selling "prompt libraries" on face book blocked me. If you don't want to get called out for selling google search results, don't sell google search results

astral goblet
#

its okay to sell things and if people think they have value, that's fine too

lament steeple
#

landscapes can be tricky and the best way to deal with them is on a case by case basis

velvet plover
#

Ohh, I see, yeah, makes sense since using a list could probably remove a lot of things that maybe a person would like for their image, but other won't

lament steeple
#

or you may want in one image, but not in another

#

just make sure if you have one you like to hit the recycle button so you don't loose it

#

lose*

astral goblet
#

i've stepped up my wallpaper gens to 1440p resolution, but my poor 4080 gives an out of memory error on that 2nd upscale step right? kind of expected tbh. BUTTT, then it just keeps going after the error. like the UI looks like it failed too, but when the terminal finishes, it updates the UI

#

task failed succesfully

lament steeple
#

I'm going to start training a 16x9 TI when I get home on 16:9 images that have been squished to square

#

"I want a 16:9 image, but I want you to use the whole network. I'll unsquish it when you're done"

astral goblet
#

does it work that way? won't that just destroy information?

lament steeple
astral goblet
# lament steeple did you submit an issue report?

i never bother because i don't have a comp science degree and am told everytime that it's not filed correctly. FOSS has a massive beauracracy infestation that slows down community engagement from the likes of me

lament steeple
#

render at 512x288 or render at 512x512 and stretch to 910x512. Which is more losssy?

astral goblet
#

well i shouldn't say that, i do bother sometimes. like when i was getting NAN errors for a month. it was because of python 3.10.6 causing it but nobody in any official discussion wanted to hear me. even when i recreated it reliably

lament steeple
#

I'm working on a theory. SD 1.5 was trained on badly cropped images. That's no secret or mystery. What if instead of cropping them all to square, we cropped them to the nearest standard aspect ratio and passed the aspect ratio on the prompt. We may get better reults

astral goblet
#

sd was trained on millions of images. badly cropped are still fine. training a base data model is a lot different from refining a base model

lament steeple
#

you have more data with the 512x512 that you stretch so it's less losssy

#

I'm downloading the original dataset

astral goblet
#

it's goal is to denoise accurately and too do that it needs a LOT of examples. notjust perfectly composed images.

lament steeple
#

The ti is just an experiment to see if it will understand what I'm telling it

astral goblet
#

no i get it. you're going to stretch the output to your aspect ratio. but... i mean.. i don't think it'll lead to better images. your ti will still be trained on destroyed data

#

worth experimenting, but don't get your hopes up i'd say

lament steeple
#

I'm not going to fine tune the model in the end. I'm going to brain wash it

#

If the experiment works, I'm planning on retraining the model from scratch

astral goblet
#

textual inversion training is sooooo much differetn than base model training, but okay. good luck. i dont think that should be your next step and you should adjust your power curve a little for better acceleration

lament steeple
#

This step is because I'm bored. and need to write the code to scale the images. The laion_improved_6 takes forever to download LOL

astral goblet
#

aspect ratio bucketing i've seen murmers about. it's supposed to help with wide generations from what i udnerstand. i think it may be more for dreambooth, but maybe the code is worth looking into

lament steeple
#

This experiment started because I want to retrain it based on certain data being synthetic. This just became a side part that I want to see if will work to improve the quality based on an attempt to improve the quality of the input

astral goblet
#

may the ai gods bless your tests

sleek raven
#

Hey

#

Can we train our own CLIP?

vestal dew
#

clip?

#

will a 4070TI do some good LORA training?

vestal dew
#

wish I knew

limpid tusk
#

ayo, in the x/y/z plot, i cant find a way to mix subjects with my prompt at the top. For example, i want to type "beautiful photo of" at the top, then have it run through different subjects like "beautiful photo of a man/woman/etc", i tried prompt s/r and prompt order but those are something different

#

ah i got it, you have to put "(<subject>)" in your main prompt, then in x/y/z plot you select prompt s/r. In the x value put something like "(cat), (dog), (woman), (man), (tree)", it will replace whatever you put in parenthesis in the main prompt 🙂 prob general knowledge but thought id share Edit: u dont need parenthesis, but could be useful to keep track

exotic acorn
#

may i ask.. if traning on the same step is useless and waste of time ?

#

or i can get a better result ?

astral goblet
# limpid tusk ayo, in the x/y/z plot, i cant find a way to mix subjects with my prompt at the ...

close. all you need to do is "beautiful photo of a man" in the main prompt. the s/r is search and replace. it looks at the first in the list, and replaces it with the others. so in s/r you'd put "man,cat,woman,dog,tree" and it'd first do the main man prompt, then replace man in each try after. () is for adding emphasis to a token in a prompt, and you can do s/r with "man, (man), (man:1.2), tree, (tree)" and it'd work too

high falcon
#

I have a super noob question.. I'm new to this space and curious about making my own image/caption dataset. Is this just a CSV with an image in column A and a caption in column B?

vast ingot
high falcon
vast ingot
high falcon
#

I was looking at using Dreambooth through google colab to train the model

vast ingot
#

the google colab notebook should tell you then. Some even have their integrated naming process

#

with auto caption that you can then fix

high falcon
#

The examples I've found don't even have captions.. it's just a bunch of pictures of dogs with the names "dog1, dog2, dog3, etc."

vast ingot
#

yep, those still are captions, in a sense. If you are training a single subject like a dog, it's usually what you do

high falcon
#

You won't get a very specific dog, but you'll get something in the vague shape of a dog 😄

vast ingot
#

full caption comes in handy for training styles, or more than one concept at once, or even finetuning in a larger sense with thousands of concepts

vast ingot
high falcon
#

I didn't realize the auto-captioning was intrinsic to some of the colab notebooks. I was under the impression I'd have to seek out auto-captioning on my own before training. But the goal is to manually caption the images though. I've seen a few similar datasets with automated captions and they're wildly off base to what the image shows 50% of the time.

vast ingot
#

yep, automatic captioning isn't enough, that is for sure, and some datasets are just that, with a simple replacement sometimes in the caption

high falcon
#

The goal is to create a small insect dataset. The auto captions keep referring to caterpillars as "toys" and random stuff like that.

vast ingot
#

using a local autocaption, and then using something like "Bulk Rename Utility", you can upgrade the captions quite easily though

high falcon
vast ingot
#

but still, you may go faster by hand, depending on what tokens you want to train

#

lots of approach for the same dataset in the end

high falcon
#

Hey I appreciate you taking the time to answer the question(s).

vast ingot
#

no problem, going to bed now. I how you manage to have a good training session !

high falcon
#

Thanks! take care and goodnight!

stark thicket
#

noob question but are automatic and gradio providing the same thing? webui?

magic citrus
kindred vessel
#

what ever happened to the old achive channels?

#

of the dreambot

wild steppe
fervent thunder
#

@vast ingot hi, I see you're well informed about training in dreambooth, I used Shivam's like months ago and I'm wondering if it's still the go-to solution quality wise. I'm on a 4090 so I don't need compromises, all I want to do is to train myself into SD 1.5 (if it's still the GOAT) but I have to get a grasp on the updates (dreambooth-wise) that I missed in these months. Could you please point me to the most insightful and up-to-date tutorial in your opinion? Thanks a ton

gaunt radish
#

what the hell happened to the categories that used to be on here????

#

that was the coolest part

smoky thorn
#

Is there a good way to load stable diffusion or other safetensors models from python? I'm trying to make my own gui. I've looked at the diffusers library but that looks like it just lets you download from huggingface.

#

I want to load models from the local filesystem.

#

Maybe I should go look at the automatic1111 source code and see what they're using...

peak orchid
peak orchid
high falcon
peak orchid
#

But sometimes you really have to play around with the model and test it fully to determine how well it learned the content

high falcon
#

Like, Stable Diffusion already knows what a pokemon is... if I try to train a model on a dataset that only includes starter pokemon (excluding pikachu specifically) but then prompts given after training show parts of pikachu in the finished custom pokemon.. how do you know it's not just ignoring your data altogether.. or at least not biasing its results based on your dataset

#

This is assuming I've used SD 1.5 as the base model

peak orchid
high falcon
peak orchid
# high falcon cool, ty for explaining that

Also note that finetuning will modify the knowledge that the model has, and this can lead to it loosing concepts. The goal is generally to balance the inclusion of new information while minimizing the loss of the existing information

high falcon
unique patio
# little yoke How do you format your question(s) to it when you want help with your prompts?

honestly I just ran into that problem right now and its really hard to pass. We would probaby need a copy and paste standardized prompt for it to be taught every time we start a new convo with it that makes sure it knows how to format the prompts. Itll probably be a very long essay to explain to chatgpt but it probably will work. Since this isn't a built in feature from what it knows (to teach stable diffusion to create an image)

#

It still does things it doesnt know how to like create prompts like that, that's why it's artificial intelligence and not google, but it will just lack unless properly taught how to do it

astral goblet
#

for chatgpt prompts, i like to have a conversation with it about prompts first. i show it a few and explain how they work

#

once you got it creating prompts in that style and it understands what the purpose of prompts are, correct it as it starts making it's own

graceful kelp
#

how do you get higher quality images?

fervent thunder
#

hello evey one~

vestal dew
#

Just ordered my 4070TI system parts. Now I will generate and train like nobody else can.

#

this is likely new ground for SD. Nobody has such POWER

fervent thunder
vestal dew
#

try to get a system close to mine

void citrus
#

Hey newb here but very excited

#

My first attempt was "Monkey farming on the Moon" and it was awesome

vestal dew
#

about what?

void citrus
#

I'm trying to get the 1111 installed but running into some hurdles

vestal dew
#

if it's not installed, how do you generate images?

void citrus
#

It's installed, I was using the default web interface and I'm trying to upgrade. I'm almost there

#

Had to make sure of the proper version number and I'm almost done

#

I'm very amateur but I kinda know what I'm doing

vestal dew
#

what is your GPU

void citrus
#

old nv 1080 but in good shape

vestal dew
#

I have gtx 970

void citrus
#

I might be able to fix this I'm breaking it down

vestal dew
#

but 4070 TI on the way

void citrus
#

It's good that prices came down I'd totally buy the 4x range but I have to watch my money

vestal dew
#

well, the SD world will soon have me with all this power

#

once I have it, I will get SD to render a new UI

void citrus
#

so anyway it seems like the installation happened but running the .bat isn't working. I'm so close

forest kettle
#

@high falcon

vestal dew
#

someone helped me to get it to work

harsh mural
#

Hello, im new here aqwave

pure arrow
#

Hi I'm new. Can someone help me start using stable diffusion

vestal dew
#

Hello, was it me you're looking for?

pure arrow
#

?

harsh mural
#

i guess alot of people got in to stable diffusion after that one model/checkpoint from china exploded

vestal dew
#

what model was that?

#

when did this happen?

harsh mural
#

ChillOutMix in CivitAI

#

i got pretty big since last week in tiktok and facebook, atleast in my country

vestal dew
#

um

#

no

#

that chilloutmix only has 7000 downloads

tame bone
#

I took a break from SD until I saw Asmons video on the current capabilities, then last night I found Photorealistic nsfw model and boi

vestal dew
#

SD was big long before that

void citrus
#

Dancing on the Ceiling underrated party song

tame bone
#

Still a pleb with not enough VRAM so have to wait more

vestal dew
#

This is not a NSFW channel, whatever that means

harsh mural
#

thats probably true, i just joined so im just probably spouting dumb stuff from my keyboard

vestal dew
#

at the Pepsi company, wearing a COKE hat is NSFW

tame bone
#

To do: generate pepsi logo with coke color scheme

vestal dew
#

I doubt chilloutmixs is driving popularity. It makes images of the same Chinese girl

tame bone
#

Last night I found a nice prompt, but it kept generating the same pose no matter what, sadge

vestal dew
#

that's how SD works

tame bone
#

aye

vestal dew
#

also lots of portraits

tame bone
#

then I fixed the pose but the realism of the picture dropped

vestal dew
#

you lose details when trying to generate full body

#

controlnet supposedly helps with all that

#

but I cant run it on the crap system I have

harsh mural
#

ngl, i jumped in SD because of it, but currently looking to how i can get more interesting things with it

tame bone
vestal dew
#

you can add such things in the prompt

tame bone
tame bone
vestal dew
#

right

#

same thing I get

harsh mural
#

does adding more sampling steps on it help ?

vestal dew
#

to some extent

tame bone
vestal dew
#

it depends on the sampling method as to ideal number of steps

#

for example, euler A is 28

tame bone
#

Less steps just makes it run faster but result is of lower detail

#

Some samplers achieve great results with less steps

vestal dew
#

right, but beyond a certain number no extra details are added

tame bone
#

you would have to compare it pixel by pixel to claim that

harsh mural
#

ive tried running some on maxed out sampling steps

#

sometimes i dont even see the difference

vestal dew
void citrus
vestal dew
#

it's running

tame bone
vestal dew
#

you have to add a command to auto open a ui

tame bone
void citrus
#

ok gotcha that sounds right

vestal dew
#

or localhost:7860

vestal dew
void citrus
#

Just trying to finally boot the ui

vestal dew
#

you now connect in a browser

#

use the address I mentioned

void citrus
#

Will do thanks everyone that's why I asked 👍

vestal dew
#

that's why we answered 👌

#

millions of dollars in training, SD can't do better hands than the emojis have

#

SD can't match the emoji hands

void citrus
#

If I could make my cat into an emoji I would do it in a sec.

vestal dew
#

🐈

void citrus
#

One of my motivations for learning this and getting this thing to run locally is to make up silly things related to my pet. Why not

vestal dew
#

it's fun

#

my goal is to at last have revenge on the Jedi

fervent thunder
#

hiii

vestal dew
#

hi and byeeeeeeeeeeee

#

now I must depart

fervent thunder
#

u may

#

hey, uhm im new here, discord and shit, someone tell me how this works HAHAHA

tame bone
tame bone
fervent thunder
void citrus
#

still working on that. localhost numbers aren't working so I have to make a sandwich and go back to the drawing board

tame bone
#

you didn't close the cmd.exe did you?

void citrus
#

Sure but I don't want to clog this channel up with my problems. Maybe another channel

tame bone
#

You can DM me if you wish

void citrus
#

I closed cmd

tame bone
#

That's the problem then 🙂 have to keep it open

void citrus
#

And installed with Git

#

I'll reboot

tame bone
#

no need to reboot

#

you have to keep that window open

void citrus
#

ok

tame bone
#

and then go to localhost

void citrus
#

ok cmd is open

tame bone
#

should see the webui

void citrus
#

refused to connect

tame bone
#

if it still doesn't connect, screenshot your cmd.exe again

void citrus
#

that worked

#

nice thanks

#

I thought it might be a port but I'm just an amateur

tame bone
#

All good, have fun

void citrus
#

ty so much

tame bone
#

Just remember you need sleep smugshroom

void citrus
#

nah that's for the other amateurs

tame bone
#

Stayed up too late, woke up too early, queue a headache facepalm

wide juniper
#

I read from a YouTube comment that on LoRA training, if I increase the batch size, I should also adjust the learning rate. Can somebody confirm this, and how do I adjust it? Proportionally, or inverse proportionally?

tame bone
#

Just trial and error yourself

final gorge
#

I have ControlNet installed but have no models in the dropdown box in the webgui

#

not sure how to fix this

wide juniper
tame bone
void citrus
#

I can't say enough how much I'm thankful you guys helped get my UI working I'm having a blast

tame bone
tame bone
void citrus
tame bone
void citrus
#

It's all good it seems like I should generate a bunch, pick the best, and then go from there for fun

#

the filters are interesting

tame bone
#

Never used filters

void citrus
#

the things that make your photo look like famous stuff

tame bone
void citrus
#

Where's the best place to see what StableDiffusion has done and are you hiring?

tame bone
#

It highly depends what you are looking for, though.

void citrus
#

I went with 'lizard growing tomatoes on the moon'

tame bone
#

Good start, going with different models could make same prompt produce a humanoid lizard, a robot lizard or a lizard lizard, just from the model alone, without changing the prompt

#

A model can interpret 'lizard' in myriad of ways.

void citrus
#

gotcha totally agree

#

this was my fav of like 5

craggy blade
#

SAY MY NAME

void citrus
#

gotta make breakfast first

#

come play switch golf. it's fun

craggy blade
#

sizin ben anennnenizi

#

toplayıp

#

daglarda surerek koyunlara siktirtiyim oruspu cocukları

#

dagda s,k,len aneannei

#

alıp yerde hoplatıp parif kulesine sokiyim

#

paris

#

oruspu cocuklarıütop

#

babanı yerdeb yere vurim oc cocugu aslanlara yedirirken babandan cıkan kaval kemigini tutup anenin amcıgına sokim

#

aq kodugumun evladı

void citrus
#

anyone want to step in and translate?

craggy blade
#

nop

#

its not that bad

#

i try to find sryan guy

#

or smthin

vestal dew
#

It's Turkish.

craggy blade
#

yeeees

void citrus
#

yeah I'm out gg

vestal dew
#

He said he wants to have sex with your mom, kill the men, rape all the women

craggy blade
#

nahh

#

bro

#

what dedin gülüm

craggy blade
vestal dew
#

I'll beat your father to the ground, rip off his shin, and use it to rape your mom - that's what he said

craggy blade
#

ı dont say that

#

you lyin

vestal dew
#

oruspu cocuklarıütop = sons of B's

craggy blade
#

naaaaaaahh

#

bro

#

oruspu cocugu is

#

every one brother

#

if anyone say that

#

he say every one brother

vestal dew
#

nope

#

it's child of a whore

craggy blade
#

yep

#

bro

#

you dont know turkish

vestal dew
#

I know enough to know what that means

craggy blade
#

ı say that

vestal dew
#

don't come in here spouting that nonsense

craggy blade
#

its not that bad words

#

in turkey

#

every one say evryone

#

sory

#

my friend

vestal dew
#

Not here

craggy blade
#

ı am in school

#

i open that in school

#

acc

vestal dew
#

this channel and discord aren't your school

craggy blade
#

bro

vestal dew
#

talk like that to your pals. NOT HERE

craggy blade
#

ı open that acc on my school

vestal dew
#

that is a non sequitor

#

Me: "Stop the nasty talk" You "BRO, I open my account at school"

#

spend more time on your work, less on potty talk

lilac reef
#

wtf?!?

craggy blade
#

bre are free now

vestal dew
lilac reef
#

How is that tame?

vestal dew
#

What kind of school do you attend? Is it like a pot school where you learn to get high? You speak as if high or drunk.

craggy blade
#

yes

#

drunk

#

cigaretes

vestal dew
#

then go sleep it off

craggy blade
#

evrytng

#

you know

vestal dew
#

stop putting poison in your body

craggy blade
#

its not poisen

#

ı dont dring

#

ı sell

vestal dew
#

do you use Stable Diffusion?

#

this channel is about that, the discord

craggy blade
#

do you want some eroin?

vast ingot
#

true dat

craggy blade
#

ı can sen you home

vast ingot
craggy blade
#

send

vestal dew
#

I dont take poison

craggy blade
#

its not posion

vestal dew
#

it is

craggy blade
#

its bale

vestal dew
#

my father lived in Turkey

craggy blade
#

realy

#

where

lilac reef
#

Well, I see the intelligent life landed so I am out.

craggy blade
#

ı can send him eroin

#

or anything he want

vestal dew
#

I am trying to remember the name

craggy blade
#

ıf you remember say

#

ı am afk now

vast ingot
#

man, sometimes I think I saw a lot of things, but that's a first :p

vestal dew
#

He was in Cyprus for a time, but also Turkey

#

he was a radio operator there

craggy blade
#

do you know where he livin?

vestal dew
#

He lived in a hotel when he was in Turkey

#

I don't know about Cyprus

uneven heath
#

Damn Elon Musk in the AI race is kinda nice

#

things blew fast from last 6 months

void citrus
#

Hear me out - maybe Elon shouldn't just post dumb memes all day while he's on his plane. Just a thought.

#

Lots of money at play

vast ingot
#

yeah but trolls gotta troll

void citrus
#

yeah I get it too it's very good trolling

#

very expensive

vast ingot
#

feels like it too

#

I'd be less jagged about it if I didn't got a tesla, I liked the guy at some point. youth's gullibility

vestal dew
#

he has become more likeable recently

#

He claims he's going to assemble a team to make a chatGPT rival

craggy blade
#

franktutor

vast ingot
#

well, his claims have not the same weight at lots of other people's claim when he just changes direction with the wind. I hope he does nice things, but I'll wait on the results and not the tease from him

craggy blade
#

is your dad name ali?

vestal dew
#

he's accomplished a lot

craggy blade
#

ok

#

mehmet?

#

Cemaliye Aligüllü?

craggy blade
#

hepinizin aq oruspu cocukları

stable pawn
#

I still rather new to Stable Diffusion and I often use Automatic1111 web UI, and there is a number in the prompt going up to 75, some time I pass this number and it increase, what is that, what is a Token and what difference do it do for the images?

tame bone
vast ingot
# stable pawn I still rather new to Stable Diffusion and I often use Automatic1111 web UI, and...

Tokens are parts of words. A token is around 0.75 word in general. Tokens are what your prompt gets broken into, and it's passed to the model to make your picture.
Tokens also are what the model has been trained on. it's how it manages to find the right weights to apply to your image so it can follow your prompt.
Tokens go up to 77. 2 are reserved, so it lefts us with 75 to use. Automatic uses a nice trick to let you go over that limit, at the cost that each token is a little less considered.

stable pawn
#

Can I turn that off, I have a bit of what once called Asperger's and when I see that number I get a stupid urge to optimize my prompt to get as low number as possible 🙂

vast ingot
#

I'll check, I don't think there is directly an option for that, but it's quite custumizable

stable pawn
#

I was in looking at the settings but my language skills do so I don't want to click randomly.

#

and I use "skill" in qotation 😄

vast ingot
#

ok

#

so open the file "style.css"

#

and add this line :

.token_counter {display:none}

#

then just do an F5 in the UI

#

all token counters should be gone

stable pawn
#

@vast ingot Cool, and thank you.

vast ingot
#

this may prevent you from doing "git pull" though

#

since you'll have modified files :/

#

I'm checking if there is another way

#

yes there is !

#

put that line in a new file you make that you name "style-user.css"

#

instead of the main style.css

#

it will act the same, but it will also keep the "git pull" possible

toxic perch
#

Where do I start as a complete beginner?

lament steeple
#

other work around is to commit your changes and then it will pull noormally

#

and let us know if you get stuck

vast ingot
#

yep, the #1072220168534642768 , the #1072229020520947753 .
Once you get access to the AI, you'll want to try it, and ask different channels in around here depending on what you are trying to do.
I'll link the FAQ on how to access the AI itself

surreal perchBOT
#
FAQ: I'm new here, how do I generate images ? Where is the bot ?

Welcome ! There is no bot currently to generate your images on discord. You may want to start by taking a look at the #1072220168534642768 channel. You can access Stable diffusion in different ways : 1️⃣ the official website, https://beta.dreamstudio.ai/. The easiest and fastest way to access Stable diffusion with 200 free credits. For any question on it, you can find help in the #1025467151206854736 channel. 2️⃣ Installing Stable diffusion on your computer. There are numerous projects that let you do that, and you will find help in the #🤝｜tech-support channel. 3️⃣ Running Stable diffusion in the cloud, through rented GPU services, using notebooks. You can find lots of them shared and discussed over in the #1011228442399883294 channel.

stable pawn
#

if I have a prompt like "small studio apartment Nordic style, one squared window, 250cm ceiling height, color and design idea", what model would be recommended to get variated 'inspirational' images, v1.5 and v2.1 often get stuck on same color (gray, blue, wood).

lament steeple
#

Are you looking specifically for color ideas?

#

or do you have some in mind?

#

try adding as many colors as you can to your prompt. sometimess sd will pick

stable pawn
#

Good idea, there is some few colors I do not like, as purple, but the idea is just to get random rooms, but not to wild in colors. I witsh there was like a IKEA model or design Pro model that is uptodate with some trends but also can go retro.

tame bone
young trench
#

Haven't you fixed the installation of automatic1111 yet?

stable pawn
#

@tame bone No, not yet. I can look into it, but I have a slight problem if it is to complex. I had a smaller stroke a bit over a year ago and that affected my logical skill and I do relearn a lot.

tame bone
#

I assume you are using webui?

stable pawn
#

Yes, Automatic1111

tame bone
stable pawn
#

I sometime use the web, but then my spelling is not the best and I need to test me forward word for word I often came to a limit or it was just to long wait. The Web UI and all models eat tons of resourses but was better for my usage.

warm junco
tame bone
#

But I'm trying to explain to @stable pawn how to locate it first.

stable pawn
#

Oh. Yes I used that scrip part, I use it to upscale in Img2Img.

tame bone
#

Ok, so select the prompt SR

#

then you take your prompt "small studio apartment Nordic style, one squared window, 250cm ceiling height" add at the end any word, so like "small studio apartment Nordic style, one squared window, 250cm ceiling height, red"

#

then in prompt SR you start with that word, in this case 'red' and you fill it with everything else, like "red, green, blue, emerald, sapphire" etc. just remember to keep batch count and batch size at 1

stable pawn
#

Sorry if I sound stupid now, I may have a logick block again, I cant find "Prompt SR" or connetl logical where it should be. Sorry.

tame bone
stable pawn
#

ty

tame bone
#

It's amazing to find out which triggers work and which don't, for example blue dress doesn't (results in white dress) but emerald and green produce two different shades of green

#

Trial and error I've found out SD doesn't really know colors.

stable pawn
#

It feel a bit stupid when I still just forget things randomly, not long ago I did try to lia that I understood, but that did not help me, so th for take the time and do the simple reminders.

cursive rock
#

hi guys, im using latest version of Automatic1111 webui, is there any way to bypass the [Warning: too many input tokens; some (11) have been truncated:] error?
i remembered that i dont have this warning last time in December when i have a long long prompt

tame bone
tame bone
zenith bobcat
#

yall know a way to more conveniently move stuff around in the colab right side file view

#

i am poor my pc is made of mud

#

i am forced to use colab 😭

spring cedar
cursive rock
#

thanks anyway

tame bone
cursive rock
tame bone
#

then just try to omit something from prompt

cursive rock
#

okay

sleek raven
#

where to post generated pictures

fervent thunder
#

general with images

sleek raven
vast ingot
small cloak
#

I have a problem that maybe someone of you have also encountered before. I use a Mac Studio M1Pro 32GB RAM and I was going to train a model with Dreambooth. Now, the problem is that it seems like Stable Diffusion Dreambooth doesn`t see the VRAM? In terminal I see only VRAM0 and it takes like 8 hours ETA for training? Is it a way to tell Stable Diffusion that I have 32GB RAM?

tame bone
#

you have RAM not VRAM

small cloak
#

yes that what I was worried about. So it means that my Mac Studio M1PRO doesn`t have any VRAM right?

tame bone
#

VRAM stands for Video RAM

#

which is in your gpu

small cloak
#

yes, now I inderstand. I thought for a moment that VRAM was Virtual RAM

tame bone
#

Is that a laptop you have?

small cloak
#

so it is Video RAM. Thats strange that MacStudio like I have where I edit big 8K videos and process big video files doesn?t have a VRAM

tame bone
#

I think it only has integrated video card right>

small cloak
#

no its desktop. Mac Studio is a desktop professional Mac

tame bone
#

If it uses CPU for processing, it uses RAM

#

does it have a dedicated GPU?

small cloak
#

oh I see

#

I can check, in the specs, one second

surreal lance
#

Hello, did someone manage to make rocm work with radeon 7000? I've read that there is a sort of workaround to make it work but i tried and it was unsuccessfull

small cloak
#

here are the specs

#

Apple M1 Max chip

10-core CPU with 8 performance cores and 2 efficiency cores
24-core GPU
16-core Neural Engine

#

32GB RAM

#

this is my configuration

tame bone
#

seems like it's a single chip

#

I'm surprised it works on mac at all, or are you using windows ?

small cloak
#

on Mac yes. I ahve used Stable Diffusion on this Mac for a long time. It creates images really fast, but I was surprised that when I tried to train I discovered that about the VRAM

tame bone
#

yeap, training is different from generating

small cloak
#

yes right? That what I was worried about 🙂

tame bone
small cloak
#

yes thanks anyway for the response 🙂

tame bone
#

I only have 8GB of VRAM myself, need to wait so I can train myself

#

wait for 5th series nvidia card, that is haha

warm junco
tame bone
vestal dew
#

Some day maybe you can have what I have arriving - a 4070 TI and all the components for a new system. That's the most powerful GPU, and I will be training with it

cobalt cliff
#

anyone running sd on the new 4090 laptop gpus?

fervent thunder
#

regular 4090 not laptop why

#

you having issues

cobalt cliff
#

no just wondering how it performs

#

i really wish stable diffusion was part of the stock benchmark test suite such as 3dmark etc

#

it's a fantastic gpu+cpu+ram workout

fervent thunder
#

Anyone around that uses controlnet?

pliant iris
#

Does anyone has link to some good new models. I did not generate for some time, and want to try again, I saw video about the Lada models and their are impressive.

fervent thunder
pliant iris
fervent thunder
#

your welcome

vague dove
#

hey everyone

vast ingot
#

Heya

carmine reef
#

some model ideas to make anime style?

vast ingot
#

#1072013871730131004 should have your back on this, there are so many, I wouldn't even know what to recommend today

hoary igloo
#

Hey! Did anyone try to train using 1024 to make high res images?

carmine reef
#

ok, thanls

wide juniper
#

Just chanced up on this extension: Kohya-ss Additional Networks
script, models Allows the Web UI to use networks (LoRA) trained by their scripts to generate images.
Added: 2023-01-06

How does this do something that isn't already possible? Aren't all LoRAs already usable?

fervent thunder
#

Unrelated to diffusers but does anyone know how to fix overfitting?

steep mortar
#

I wish there was a way to have our outputs named via certain tags that are in the txt prompt. such as: "girl wearing a <red dress> at a coffee shop." the img would be named after what is in the < >. so 0023_red_dress.png

lime violet
#

might be a stupid question, but do I need to install SD2.1 on auto1111 to run custom SD2.1 based models?

warm junco