#💬｜general-chat

1 messages · Page 41 of 1

vestal dew
#

trying to add objects is weird

#

I wanted to do an image of a guy with his tongue stuck on a cold pole...no can do

#

so then I tried a person licking a sucker..nope. How about just sucker in mouth? nope. Holding a sucker? kind of

regal snow
#

I'm having slightly better success with Controlnet but still, the details are all wonky

vestal dew
#

throwing or kicking a ball - often get a giant soccer ball, or a giant baseball, or multiple balls

#

or a hat that looks like a baseball

#

or a soccer ball as a hand

potent fox
#

I don't know if sd-webui has the same features like InvokeAI, where you can mask out a specific part and then recreate it. That is still something i am missing or am i wrong?
It is the unified canvas.

vestal dew
#

a11111111 does that

normal coral
#

Hey I'm in the middle of setting up Stable Diffusion for the first time using Automatic1111 as the webinterface while following along a youtube guide: my question is how do custom models work exactly? Do I still first have to install e.g. Stable Diffusion 1.4 before then putting custom models into the folder or are they effectively the same?

#

Like can i skip downloading 1.4 and just replace that part of the video with for example Realistic Vision 1.3 or does Realistic Vision 1.3 merely tell the base model of SD 1.4 how to behave?

potent fox
vestal dew
#

I dont think you have to install those originals

#

models are self contained

normal coral
#

yeah i had a hunch itd be like that because otherwise surely the realistic vision page on civitai would specifically mention for what version its made right?

vestal dew
#

usually they do mention

normal coral
vestal dew
#

Realistic Vision is a 1.5 model

normal coral
#

so am i good to ONLY drop the realistic vision.safetensor in that folder and then follow the rest of it normally?

vestal dew
#

but I dont think models use other models. It works on its own

potent fox
#

Here is something strange: i get the best results for my needs using a NSFW model. Even i don't create NSFW images. It depends so much...

vestal dew
#

I think so Good

normal coral
#

im a bit spooked about the installation to go off a suspicion

#

does anyone know for sure before i potentially waste 40 minutes of installing the wrong thing

#

also @potent fox i think its mainly because it's a lot easier for an AI to figure out human anatomy by training on naked images

#

even if it clothes them after

regal snow
#

Anyone tried both Dreambooth and LORA methods? Wondering if I might have better success generating objects if I try DB method instead

vestal dew
normal coral
#

main reason why id rather skip 1.5 is because theres seemingly no safetensor version of that one available

vestal dew
#

you load the model in the UI

#

if it doesn't work, then add the 1.5 base

normal coral
#

it has to download a bunch of stuff first no?

#

in the video i linked he said it took him 20 minutes on a really good pc

vestal dew
#

a model based on 1.5 has 1.5 in it

normal coral
#

im not sure what its downloading at that point exactly

#

all of this is rather confusing

vestal dew
#

yes, but the models are self contained files

potent fox
#

I even used 3 different Lora's at the same time.

vestal dew
#

the models are individual files that go in the models dir

#

unless you're mixing models, you select one at a time when you generate

#

if you generate with the vision you mentioned, you aren't using the 1.5 base

normal coral
#

hmm alright ill give it a go then

brisk bloom
# normal coral im not sure what its downloading at that point exactly

what it downloads are all the dependencies for SD to run and to enable easy update afterwards. You chave the web-ui which is updated nearly daily which loads checkpoint models, which can be based on 1.4, 1.5 or even 2.1. the models are technically the version of SD you are using. There are hundreds of models you can use

potent fox
#

You can use all your Lora's in your prompt even without the additional networks settings, which do not show up for me lately.

normal coral
#

as far as security goes since you said the web ui updates nearly daily

#

is that an avenue for someone to inject bad code or is it safe so to speak

brisk bloom
normal coral
#

is there a way to force updates off regardless?

vestal dew
#

don't add git pull in your bat

brisk bloom
#

unless you manually enable updates it wont update

potent fox
#

Just use <lora:name of your lora file:weight>

vestal dew
#

then you update only when you want

#

even when you update, you can roll back to a previous version

normal coral
#

i can set it up first for the install then remove the gitpull right

acoustic tangle
#

shhhhh, just enjoy the newsletter

#

and the metaverse

brisk bloom
#

Your safest route is always use known checkpoint hosts, dont go to shady sites and get safetensor files if possible

vestal dew
#

you dont do git pull when setting up

#

you use that command when you want to update

potent fox
#

I don`t know why SD is connected to the Metaverse. Really no clue at all.

vestal dew
#

if you want to git pull each time you run, add it to webui-user.bat

normal coral
vestal dew
#

No , literally "Git Pull"

brisk bloom
#

git clone is used for the first install. git pull is to update

normal coral
#

ah i see

#

alright

vestal dew
#

by hand, open a CMD window in your stable-diffusion-webui dir, in that CMD type git pull

brisk bloom
#

if you dont git pull, the web-ui doesnt update

vestal dew
#

you can also add it to the BAT so it updates each time you run it

cedar quiver
# acoustic tangle shhhhh, just enjoy the newsletter

nope, I'll just sit here and watch some of the preeminent pioneers of this technology empty a glock-19 full auto into their feet for no reason at all while the actual technology experiences a meteoric rise in popularity and development around them while they. . uh, do something with the metaverse apparently

acoustic tangle
#

lol

vestal dew
#

the meeeeeeeeeeeeeeeeeeeeeeeeeeta veeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeerse!

potent fox
#

Be sure to have full write access to the folder recursively. I had problems using git pull because of that.

vestal dew
#

when you enter the metaverse, the part of your brain they remove is given to you in a glass jar

normal coral
#

the only time the webui would have to be updated is if a model im using pushes an update right?

vestal dew
#

no. all sorts of things update

#

you may also have to update extensions

brisk bloom
vestal dew
#

if the person you go out with is taller, you will also have to up date

normal coral
#

alright ure getting banned

potent fox
#

I think that frank has frankly some kind of issue here...

normal coral
#

btw all this stuff installing right now is contained within the AI folder i created right

#

its not installing somewhere else as well?

brisk bloom
#

the web-ui gets regular updates because improvement, new features etc are happening all the time. SD is moving like lightning and the A1111 webui needs to be updated or people complain or move to another UI

normal coral
#

No module 'xformers'. Proceeding without it.

#

uh oh

potent fox
#

The git pull works fine. You just add it every now and then to the webui-user.bat

normal coral
#

ok well naturally it says unable to connect after it gives me the local url

#

dummy me expecting it to work first try

potent fox
#

I also add --no-half to the commandline. For some reasons.

normal coral
#

ok but how on mothers green earth do i troubleshoot this now

#

i have no idea whats going on at any point how am i expected to figure this one out now

potent fox
normal coral
#

ah wait i got an error message below the url

#

Interrupted with signal 2 in <frame at 0x000001629373B720, file 'C:\AI\stable-diffusion-webui\webui.py', line 173, code wait_on_server>

brisk bloom
normal coral
#

Terminate batch job (Y/N)?

normal coral
potent fox
#

That's strange. Please verify you installed Python 3.X and the comment of pure_fire.

brisk bloom
#

which python version did you install?

#

preferably you want 3.10.6

potent fox
#

open a cmd and type "python --version"

normal coral
#

a cmd anywhere?

#

C:\AI\stable-diffusion-webui>python --version
Python 3.10.6

C:\AI\stable-diffusion-webui>

potent fox
#

Can you update to python 3.10.9?

normal coral
#

i did install git before i installed python

#

does that make a difference

potent fox
#

Type git --version

normal coral
#

git version 2.39.2.windows.1

brisk bloom
normal coral
#

so what are my options here now

#

was kinda hoping itd go smoothly

brisk bloom
#

paste what you have on the line I mentioned earlier

#

set python etc

normal coral
#

set PYTHON="C:\Users\Steve\AppData\Local\Programs\Python\Python310\python.exe"

brisk bloom
#

try remove the ""

normal coral
#

what do i do after i remove those?

brisk bloom
#

relaunch the bat

normal coral
#

okk it works

brisk bloom
normal coral
#

the guy in the video had the " but oh well

lavish anchor
#

Where might i ask about dreambooth?

potent fox
#

Strange. It seams the "" are needed when you have spaces in the path. Or maybe they also need to be escaped then.

lavish anchor
#

"" are normally needed for spaces

potent fox
#

But i didn't think they make problems if there are no spaces.

lavish anchor
#

Soo... Anyone know where i can ask for some help with dreambooth?

normal coral
#

thank you everyone for the help by the way

#

much appreciated

brisk bloom
normal coral
#

hmm alright another question as a test i just copied peoples prompts/settings to see if my setup generates the same image

#

but it does not

#

is there another thing im missing perhaps

#

ahh id imagine resolution maybe?

vestal dew
#

model

potent fox
#

Probably the model u are using

brisk bloom
#

to get the same image you need exaclty all the settings right, but even then the GPU you have can change the output depending on how old your GPU is

vestal dew
#

right, the GPU might not generate the same details

potent fox
#

No, the GPU has no impact at all.

#

Only when it comes to render times.

normal coral
#

well yeah im just going down the realistic vision civitai page

#

where people post images they generated with it

vestal dew
#

link one

#

Ill try it

#

you wont get the exact same image generally

normal coral
#

idk dont judge for booba but this guy actually supplied in depth prompts

potent fox
#

Why would anyone create the same image as anyone else anyways? This is not art. Sorry.

vestal dew
#

what was the original image resolution?

normal coral
#

yeah thats waht imeant

#

i can only imagine resolution as the difference maker right

#

cuz that one isnt being listed

normal coral
vestal dew
#

that will make a difference

potent fox
vestal dew
#

it also doesn't mention if it had restore faces on, hires fix, what for restore faces

normal coral
#

what does restore faces do?

vestal dew
#

it fixes imperfections on the face...sometimes

potent fox
#

Only on low resolution generations.

vestal dew
#

this surely wasn't generated at the resolution of the image

#

that was upscaled

potent fox
#

You could give examples in the images channel.

vestal dew
#

not sure this image is within the rules

potent fox
#

What image? You can post it on my server if you dare.

vestal dew
#

i am generating now

normal coral
#

god theres so many different samplers

#

which one are generally the good ones

vestal dew
#

euler is good

potent fox
#

DPM2++SDE Karras if i remember correctly.

vestal dew
#

at 512x512 I got face only

potent fox
#

Euler-A is fast and good. But not so high details IMHO

vestal dew
#

I am now generating Thai Tina at 512 x 768

#

face only agaim

potent fox
#

The models are trained on 512x512 images usually, but you can go to higher resolutions using some tricks.

vestal dew
#

again

#

right. the image shown with this is seated, down to around the knees

#

I keep getting face only

#

there's nothing in the prompt about her pose, the clothes, nothing

potent fox
#

Come on. Get it done 😉

normal coral
#

is there a difference in generation speed in nvidia vs amd btw?

vestal dew
#

the prompt is "face of a thai woman with a happy face..."

#

"a face of a thai woman with a happy face, big beautiful lips, long lashes, extremely beautiful, 4k, 8k, trending on artstation, greg rutkowski"

#

that's going to make a face

potent fox
#

I can't test the prompt since i shut down my rig. But i expected something to show up somewhere 🍭

vestal dew
#

I dont think that image was generated with that prompt

#

could be the face, then more added with outpainting

normal coral
#

btw Loras are basically just adding certain generation training to the existing model right?

vestal dew
#

I get just the face every time

potent fox
#

Oh my guy, you are getting to it surely. But then, please show me your result!

vestal dew
#

I posted in general with images

#

it's just the face every time

normal coral
#

gonna take a while until i figure out a good balance of the negative prompts

#

is there like a staple that you should put in all of them

#

that dont impact generation negatively much at all

rain seal
#

I set up stable diffusion and its running!

#

omg!

#

I feel so psyched!

waxen pasture
potent fox
#

Okay, it looks like a portrait. Maybe you could outpaint it to include something else, but i guess you would get away better if you used a body shot.

waxen pasture
vestal dew
#

you can also try downloading bad prompt

vestal dew
potent fox
vestal dew
#

an image similar to that supposedly generated

#

but looking at the prompt, I don't think that image came from that prompt

potent fox
#

Can you show me the image which came from the prompt?

vestal dew
potent fox
#

What resolution did you use for generating the prompt?

vestal dew
#

512 x 768 and 512x512

weary hull
#

hello guys !

potent fox
#

@vestal dew Amazing. The given prompt is basically worthless and can't produce that output. Except they were using a specifically trained model.
Especially the amount of negative prompts is ridiculous.

lost marsh
#

used 3090 came in with 2 fans that are nicked, and scratches here and there kli

#

but its a 3090 for sure

potent fox
#

A 3090 is for sure a good thing for this.

past edge
#

how much u got it for

potent fox
#

Free 3090's for everyone.

#

NVIDIA should put them out cheaply and evolution gets faster 😉

normal coral
#

btw someone mentioned outpainting

#

where do i find that in automatic1111?

#

or is outpainting basically just changing aspect ratio and setting it to resize and fill

potent fox
#

Good question. i also wonder about that. It is easier in InvokeAI.

#

But i want to stay within sd-webui.

brisk bloom
#

Extensions tab > Available > press the load button, search for "openOutpaint extension"

#

you will need to add --api to launch arguments

surreal perchBOT
#
FAQ: What is Stability AI?

Our vibrant communities consist of experts, leaders and partners across the globe. They are developing cutting-edge open AI models for Image, Language, Audio, Video, 3D and Biology.. AI by the people, for the people. Learn more here stability

normal coral
#

where do i find the launch arguments?

brisk bloom
#

in the webui-user.bat

#

same place you put --xformers

lost marsh
#

ok i checked 3090 gpu z, all numbers are good, now i just gotta run it and see if any issues arise

lost marsh
#

this speed is enjoyable

#

still amazes me how the person who used directml to make it run for AMD was able to do it, with the issues that it had

normal coral
#

btw how do you increase or reduce the weight of a prompt phrase?

vestal dew
#

some versions of a1111 also open a launch window

#

there's a box in there for launch arguments, but you have to press SAVE after adding them

#

so many people have 3090's

normal coral
#

anyone know?

#

in novelai it used be {{}} to make the prompt phrase have more weight and (()) to reduce it

#

how does it work in automatic1111

brisk bloom
normal coral
#

ahh

#

ty

vestal dew
#

for some reason outpaint made the canvas and image tiny

#

I closed, loaded a new UI window, and now it's fine

warm pasture
normal coral
#

so if i understand this correctly only a few models are capable of doing outpainting?

brazen stream
#

Hi there! Where can I find models with safe tensors checked or maintained by Stable Foundation? Also, are there the same for upscalers?

burnt kelp
#

anyone knows how to get controlnet?

vestal dew
#

What in the world is going on?

waxen pasture
#

amazing

vestal dew
#

I run A11111111, try to generate something. Before generating, all the settings change to some hires fix, and then cuda error

nimble stump
#

I am new to this. Any resources available that I can use to get started with SD?

waxen pasture
nimble stump
#

Awesome. Thanks a ton:)

vestal dew
#

how can there be an ETA for Generate Forever?

astral goblet
#

bleeding edge code. thats how

vestal dew
#

It's odd. When I ran it earlier today, it didn't show any ETA. Now it's treating each generate as separate, even though it keeps generating

blissful vault
#

can someone one help me with this?
I tried changed the width & height to 512x592 but it gave me this error
Image
then, i tried not to change the size of the image i am trying to generate but still gave me this kind of error
can anyone help me out?

#

RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 148 but got size 74 for tensor number 1 in the list./

vestal dew
#

Looks to me like a tensor is too small

fervent thunder
#

ControlNet is absolutely crazy. especially with OpenPose. if you haven't messed with it yet, do yourself a favor and give it a try. just mentioning this because someone did the same a few days ago and that's when i learned about it

alpine bear
vestal dew
#

changed, BTW, from bigger or smaller?

fervent thunder
#

I just added the "open pose editor" other extension... my day is gone into this already I think

astral goblet
#

you'll be delighted to know that microsoft paint 3d provides a lot of easy shapes for building cool compositions off. some fun to be had there

fervent thunder
#

great tip, I hadn't though about it yet

#

I'm no 3D designer, blender and others are too far from my understanding for now, so a simple tool like that is great

astral goblet
#

I do know blender and such but they're such a bulldozer when sometimes just a shovel is needed

fervent thunder
#

yep, like starting visual studio to edit an XML file

astral goblet
#

lol yeh!

#

it's also easy to onboard kids with too

fervent thunder
#

Michael, that's a no go for here

#

you've been blocked once, warned once

#

next I kick

#

I didn't see, he was already spamming

#

banned

vestal dew
#

Instead of shapes, at least for people, what about images of people made in DAZ or Poser or Blender?

astral goblet
#

paint 3d has lots of other 3d models in a huge community gallery. but yeah others are good too. theyr'e less "accessible" but work just as well

#

i only throw paint 3d out there because its free for windows users

vestal dew
#

UM

#

As I was generating images, I made an edit to one in photoshop, added a tattoo. The next generated image had a similar tattoo in the same place.

wide juniper
#

Any news on SD 2.2 or 3.0?

lilac reef
#

3.0 should be 1.5-2 years

wide juniper
#

lol

lilac reef
#

2.2 by summer

#

2.3 around fall I believe

#

give or take 2 months for each

wide juniper
#

Think they’ll be popular? Or do you think people will stick with 1.5

fervent thunder
#

didnt they just announce 3.0 is coming soon/next

lilac reef
#

Depends on the tools

#

If they did then they are sucking the pipe so hard. They can see how much of a failure 2.x has been they may have just up and shat the bed to ignore 2.x and jump straight to 3.0. WHICH, by the way, is when 1024x1024 was originally supposed to come.

fervent thunder
#

2.x was definitely a bomb

lilac reef
#

I have no idea as jumping to 3.- this fast would show they utterly failed with 2.0

fervent thunder
#

they were like almost anti-hype levels of release

lilac reef
#

Basically admitting their failure. Don't get me wrong because they had to do what they did.

fervent thunder
#

this tweet is what i was referring to

#

guessing from that 3.0 could be out by end of march/early april

#

I seen some also say maybe Deep Floyd = 3.0 and thats what hes talking about, but either way something hot and new coming in haha

lilac reef
#

Yep, looks like this is why he has been so quiet because of the utter disgrace of 2.0 amongst the community. Too bad

fervent thunder
#

Deep Floyd they said originally would be early to mid feb but it kept getting pushed back

lilac reef
#

Yes

#

I though DF was just a way to finally get legible text

fervent thunder
#

as long as the phrase is like 3 words haha

#

it struggles with like an actual long sentence from what ive seen

lilac reef
#

Oh, damn. 😦

#

Anyway if DF is just for text then why would there be an opt out from 3.0?

fervent thunder
#

they had a test bot certain people could use to test it, but its been down for about a week now (DF) so people been wondering if that doesnt mean its release is imminent

#

DF isnt just for text, its a whole new bigger model

#

text is just something it can do

#

but i think controlnet yeets some of its features

lilac reef
#

Well, I can barely run SD now

fervent thunder
#

yeah people are saying youll need about 16 gig to run DF, but thats just some users guesses

#

and seen lots say would need a new ui too wont run in auto

lilac reef
#

If DF is 1024x1024 I can see that, yes

#

This is why I am due for a new card (every 3 gens) but Nvidia is too greedy for me so I sit here with a 1060 6gb. Next card has to have a min of 16gb or I will not even look at it.

fervent thunder
#

yeah

#

24 would be ideal to future proof

#

and even that is probably only a few years future proof haha

lilac reef
#

3 gens 24gb will be like my 6 or 8gb is now.

#

If things keep going as they are.

#

Sad thing is that no matter how great the 4090 is it is not worth the street price for it. 1200. 4080 800. 4070ti 600. 4060 399-449.

#

Sad thing is AMD is just as bad and Intel can't get its stuff together. I have hopes they can since they are Intel but never good with gpus. Not bad cards for a real first go though and priced right.

#

Now if they gained marketshare you bet they would be gouging us too.

steady arrow
lilac reef
#

Those prices I mentioned is what the cards should be at. 4090 out the door should be no more 1399.

#

I am accounting for inflation not their greed.

fierce steeple
#

wow

undone marsh
#

Does anyone know what the last tweet of Emad can mean? He says "Some fun AI-generated media dropping Sunday... Future is coming fast, time to nap..." Text to video maybe?

fervent thunder
#

well I didn't see it. Now I'm teased

undone marsh
#

Would gladly share a screenshot of his tweet but it isn't possible to share here

fickle kraken
#

How does selling art generated with dreamlike diffusion model works? I’m a bit confused by the license and copyright of it even after reading them. I thought if you generated it from SD, you would own them, but generating with a specific model card would prevent you from selling / gaining profits?

fervent thunder
#

yep I saw where you shared it

#

(Im not good at that, I prefer not to answer, I don't want to say stupid things about picture rights...)

fervent thunder
#

i know that's a -really- dumbed down version of the whole explanation, but it explained enough to me

nocturne python
#

Can someone give me a quick guid on how i can generate images using this server?

#

I’m totally new here

brazen stream
fervent thunder
# brazen stream Is there any curated list by Stable Foundation with safe models?

we have a forum on this discord with lots of models, #1047197565365538826
but what do you mean by safe ? no malware ? try to use models shared as .safetensors, those are made to prevent the risk ckpt can have.
If you use automatic for example, there is an automatic test on unpickle to protect you, and it should, unless you manualy disable it.
Outside of that, there is Hugging face and CivitAI that have implemented a scan of ckpt that are uploaded, and they flag dangerous things too.
https://huggingface.co/
https://civitai.com/
But no, we don't have a curated list like you are asking, those options seemed fiting for now

fervent thunder
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.

fervent thunder
#

hey

brazen stream
fervent thunder
brazen stream
fervent thunder
#

no problem, happy to see people care for their safety !

fading quarry
#

yo, so does Auto1111 have a guidance scale option for Img2Img generations?

#

i switched from cdmr2's easy diffusion to auto1111 recently so i can use LoRAs but it seems its missing the guidance scale feature

#

the CFG scale is an option but that doesnt really have the same effect on the image as the guidance scale in cdmr2's

fervent thunder
#

it should be the same, the CFG Scale that you found, yes. it's strange that it doesn't act the same way, this is a base torch paramater that isn't really processed more by the tools, it's sent to torch as is through the Stable Diffusion pipeline...
What difference do you see between the two features ? Maybe there is another feature in Auto that covers it too

fading quarry
#

the guidance scale in cdmr2's easy diffusion changes how close the img2img result is to the image you provide, while the Auto1111 CFG scale adjusts the effect the prompt has on the original image, which sounds similar on paper but if the guidance scale is set to max it'll literally just give you the same image with minor changes, and the CFG scale on 1 just gives you a noisy mess that vaguely resembles the image you provided

#

easy diffusion has two options for this same thing, one thats prompt strength and the other thats guidance scale

bleak matrix
#

Goof morning, everyone! How are we all today?

fervent thunder
#

it's quite sensitive, I mostly keep it between 0.4-0.6

fading quarry
#

aaah i see

#

thank you!!

fervent thunder
#

no problem !

fading quarry
#

this is so useful

fervent thunder
#

hum

#

wait

#

I inverted my thoughts

#

it's the other way around

#

the way to remember is this : the noise is the input picture

#

the more you denoise, the further from it you can get

lofty parcel
fading quarry
#

right i see

bleak matrix
lofty parcel
#

I gave up on the whole "SD making every chunk of a giant game world" thing and went for a more pragmatic approach

#

Sometimes to best way to solve problems is to not have them in the first place

bleak matrix
#

Plus, you game will be more structured, and a more playable experience with more cohesion

#

So, what is your new approach?

fervent thunder
#

yeah but... SD + GPT ?

bleak matrix
#

It can create game dev documents lmao

fervent thunder
#

🤣

lofty parcel
dire quail
#

Hello i'm wondering, is there anyway to merge the content of multiple images in the SD WebUI? I have one image that I really like the artstyle of, and another that I like the content of, and i'm wondering if i can mix the two, because i saw an article of a similar ai doing it but they never had any links or references

abstract pivot
#

What would y'all say is the best general model right now? like say instead of just using basic 1.5 and you want something with that and 'generally' more, is there one? i mainly use unstablephotoreal but it's obviously tilted towards photo or realism, is there any general 'upgrade' all rounder model ? There's a ton of specialisation models but its getting a bit much

bleak matrix
#

laughs

#

What does that m e a n?

#

🤣

lofty parcel
#

It just means basic character portraits and scrolling backgrounds
For now

bleak matrix
lofty parcel
#

ControlNet looking pretty spicy ngl

bleak matrix
#

I haven't used it much myself tbh LOL

#

I have so many things I am working on that I just c a n ' t

lofty parcel
#

I might want to train my own ControlNet but it sounds like I'd need 50-100K images and 1-2 weeks of compute

bleak matrix
#

I don't know the exact numbers myself, but like anything, it's a lot of work

foggy bone
#

hi guys , trying to fine tune the protogen model , where and how do i find the path from huggingface?

fervent thunder
#

if I change the filename of SD 1.5ckpt to model.ckpt webui comes up with an error and says it cant locate model.ckpt. What am I doing wtong?

gentle otter
#

why are you renaming it

fervent thunder
#

some older tools required to rename the model, but I don't think any of the current UI demands that

#

you should be good with the first name. Though, renaming it shouldn't pose that problem either
this is strange, you reloaded the list of models in the UI, or it doesn't even start anymore ?

past vortex
#

hmm, my Textual inversion embeddings aren't loading anymore

#

after a recent 1111 update

#

anyone else ever have this issue?

fervent thunder
past vortex
#

I have a 3080

#

but I guess it's the same issue

#

mine straight up says on launch that they were skipped

fervent thunder
#

yep, but I'm not sure the other guy diagnosed correctly the origin of the problem, I doubt it's linked to the card type itself

#

ok so not the same

#

it just says it skips them

#

and no why at all

past vortex
#

I can't post a screenshot

fervent thunder
mint imp
#

Generally, what's the best sampling method?

next oar
fervent thunder
#

Im renaming it becuase webui seems to look for model.cpky by defaault

wide juniper
#

My img2img using controlnet with a blurry image used to sharpen the features, but recently It's been outputting images of the same bluriness as my input. I forgot what helped me sharpen the features. Anyone has ideas?

astral goblet
# mint imp Generally, what's the best sampling method?

hard to say. i used to jump straight to DPM++ SDE but now i'm checking out 2M and DDIM instead a bunch. i can get great results from low step counts. Though to be honest, i dont really know what the best is.

Karras are mostly identical to their counter parts but do it in less steps. i've never seen a good explanation of what the karras samplers do

mint imp
#

hmmm ok

astral goblet
#

DDIM seems to get the fastest results i've seen around 20 - 25 steps

mint imp
#

maybe I need better prompts 💀 my results are kinda messy

stable pawn
#

Why do AI still suck when it come to make hands, it can make most body parts and even the sexy bits rather okay, but whats up with 12 ginger hands?

astral goblet
#

also these models are barely 6 months old. controlnet helps

fervent thunder
pastel fern
#

guys i have the perfect solution for ur hand needs, its called mittens.safetensors

fervent thunder
#

lol yeah, I resorted to never making hands again too

pastel fern
#

to be fair

#

almost every artists have struggled with fingers

#

its part of the journey

fervent thunder
#

could we say we are .... mid .... journey ?
looks for the door

dusk crown
#

hey guys, quick question - would I be better off getting a 16gb graphics card or would a 12gb suffice for most SD related stuff?

humble comet
#

hi is there SD artist here who can do a quick commission, i need to see portfolio to start. realism portrait

fervent thunder
#

Cause to me it looks hella AI but people say im wrong

dusk crown
fervent thunder
fervent thunder
fervent thunder
fervent thunder
#

And look at the weird details, this doesnt look at all like an 2023 preorder bonus picture, harley has a weird hole in the stomache

fervent thunder
#

or it's just a screen capture in a moving video

#

Ah maybe yes, somethings clearly wrong but the overall quality reminds me very much of cheaply generated AI pics

pastel fern
#

guys how do you prompt ugly ppl

#

why would they use ai art?

#

its obviously them using their low quality assets

mild ridge
#

hey, how do you install and then use checkpoints?

dusk crown
pastel fern
#

we need a chat bot

#

to help ppl onboard lol

fervent thunder
#

hey @mild ridge

mild ridge
#

@fervent thunderhey there Guizmus

fervent thunder
#

it's quite easy usually

#

you download the model (.ckpt or .safetensors), you put it in your models folder, you reload the dropdown menu at the top left of the UI and BAM, the new model apears there. you can select it and then use it to generate pictures

#

some precisions

#

prefer safetensors, they are.... safer ^^

mild ridge
#

thank you

fervent thunder
#

some models do require doing something more, downloading .yaml file, those are the 2.X models

#

it's usualy indicated on the model

#

there are some other things but it's already enough to start 😉 I'll loose you if I continue

pastel fern
#

whats even the diff between 2.1 and 1.5

#

and are most ppl just using 1.5 still

astral goblet
# pastel fern we need a chat bot

thats what microsoft thought for bing but then people are just using it to confirm their extreme political beliefs instead. chat bots might have some unintended social consequences. i think "her" from director spike jonze (sabotage music video), is just the tip of the iceberg of unintended uses

#

sure i'll ask "are chatbots needed?" and i have concerns about them, but i'm pretty sure we're on rails here and theres no getting off this ride. no sense fighting it. would one be useful here to help onboard people? probably not. it'd get immediately abused and is no more useful than #1072229020520947753

peak orchid
#

"Artist Eva Toorenent advocates for legislation and rules for what she calls 'unethical technology' with her organization, European Guilt for Artificial Intelligence Regulation (EGAIR). 'While Midjourney makes a lot of money with this software, the artists and creators whose work is involuntarily included in this dataset see nothing in return,' Toorenent said to the Volkskrant. 'Without the work of human artists, this program could not generate any works at all. The higher the quality of art in the dataset, the higher the quality of the AI art."'
"Toorenent called it bizarre that the Mauritshuis singled out an AI creation. 'That is quite something. With this, the museum is actually saying: we think this is okay.'"
Apparently this individual also has her own equivalent of the Concept Art Association that's trying to lobby the EU, called the European Guilt for Artificial Intelligence Regulation (EGAIR)

#

"Whilst artificial intelligence may be useful for some professions, Toorenent believes that within the art profession, AI should be a no-go."

#

She might be even more extreme than the Concept Art Association, as she wants a total ban rather than the CAA's goal of a 2% max limit.

pastel fern
#

European Guilt lul

dawn plume
#

Hey guys, I have a friend of mine who's a tech guy from Silicon Valley, he would like to chat and interview a few people that are interested in ai generated art, like people who have used stable diffusion.. Anyone interested?

vestal dew
vestal dew
vestal dew
cunning cosmos
fervent thunder
#

what

regal snow
#

What is the ideal size for a Lora preview image?

flat verge
#

heads up not to click on shady links btw

sleek raven
#

guys I am getting noise image for my classficiation image

#

is it normal?

trail wing
sleek raven
#

I think something is VERY wrong

#

help

crystal patrol
#

Hey guys, for the same price is getting a used 3090 or a new 4070ti for the same price better?

halcyon granite
#

regarding model training, how should I prepare the source images and what quantity is ideal?

hallow portal
crystal patrol
#

For gaming, under 4K they are equal. Although 3090 consumes quite a bit more power. (400w vs 250w I think?)

hallow portal
#

In general, used cards are fine, but if it's the same price then it's not the same price because the used one will be worth more with the warranty.

#

the new one*

#

Plus 3000 series cards are old now so there's no real reason to keep supporting that ecosystem. 4000 will be the last of this generation (PCIE 4.0) so you'll be able to ride out the driver updates for many years.

crystal patrol
#

I also read that there were transient spikes on the 3090 which can kill your psu. I don’t know if that’s fixed now or not

#

I just got a Response from the seller to get the deal on the 3090 for $150 less than what I pay for the 4070ti. Should I take the bite? If so I will return the 4070ti I bought

hallow portal
#

nah

#

Just re-sell your 4070ti just before the warranty is up for close to full value and upgrade to a 4090 then.

crystal patrol
#

I see, thanks for the input, hopefully 12gb vram will be sufficient for the next few years

summer marlin
#

hi

limpid grotto
#

hey ya'll -- i'm about to post a video2video workflow i've been experimenting with on twitter

#

and i wanted to know which of these two videos i should put in the first tweet to make for a more impactful hook to read the rest of the thread

#

ah, can't post videos here 🙂

digital lake
#

hi

vast ingot
grave copper
#

guys ive made a good picture but ive used random seed -1

#

how do i now use that seed to try and fix the image

limpid grotto
vestal dew
stuck marsh
#

hi guys

#

can anyone draw a portrait?

fervent thunder
grave copper
#

i got it

wraith night
#

I know there is a lot of instruction on building poses, was wondering if there is model that does "more" simple poses rather then the full on OpenPose

sleek raven
#

Jeez i havent slept

#

Ffff

#

Guys if I want to teach somethint like woman looking left, I need to do dreambooth right?

magic citrus
#

is there an easy way for auto1111 to do the same img2img settings run for 100+ images in my foldeR?

pastel fern
#

i see gen chat is just questions now huh

sleek raven
#

Like we need more tutors

#

Jesus hire more tutors for us

#

Raaahhh!!!

#

Hahahah

#

Anyway I am so tired now

pastel fern
sleek raven
#

Sure

#

Minimum wage

#

Lets go

rancid sleet
#

Is any way i can train preexisting model and retrain how i am totally messed

last merlin
#

check batch tab

magic citrus
#

ty

last merlin
#

people make video using those batch + control net

sage plume
#

Does anyone know any AI to generate music?

lusty verge
#

Hello

astral goblet
#

i got a 4080 so if you are legit with that information, i'll go hit up my copium dealer rn

minor pawn
#

E

astral goblet
#

i think too if a psu died because of a spiked load, that psu was on it's last legs anyways. They really shouldn't be that fragile

vestal dew
crystal patrol
astral goblet
#

yes but vivy was saying 4070ti beats 4090 and 4080 somehow. makes me curious

crystal patrol
#

I never said that

#

I said it beats the 3090

#

🤣

astral goblet
crystal patrol
#

The 4070ti is faster……besides 4080/90

#

🤔️

astral goblet
#

oh yeah okay, you meant its faster than all but those two

#

mystery solved gang

#

also, the 4070ti has dlss3 for those sweet free frames

crystal patrol
#

Yea there’s that for gaming. Without a doubt it’s the better buy for that purpose. But since I often hear vram is king in air training this making me curious

coral totem
#

If you weren’t to count the subscription part, is Novel ai still worth it? I’ve used control net with the anything v3 model on stable diffusion but the constant unfixable memory issues are making me consider novel ai. Although I’m not sure if it’s nearly as good as stable diffusion with anything v3 and control net or if it will even save storage and work faster.

#

Has anyone here ever used Novel. Ai?

vestal dew
lilac reef
little yoke
#

Can someone point me in the right direction with a post or online resource about a problem? I trained models using photos of myself through a couple different google colabs and when I test them right there in google colab the images generated look like me. When I copy the model into my locally run (on my pc) stable diffusion and select that trained model, the images generated don’t look like me anymore. I’ve looked here and haven’t seen this question asked yet but I’m sure it has. Thank you!

vestal dew
heady dawn
#

tips on getting a different character into the pose that a character in the 'initial image' is in?

#

I'm trying to use 'midjourney diffusion' to get a midjourney style

astral goblet
#

good 14 14 14 14 ddr4 dimms that max my alderlake are 2.5x the price i've found

#

still got me thinking though

lime bane
#

Guys how do i even generate a specific anime character

#

People who use stable diffusion in tiktok are generating detailed anime characters

#

I use the same model

#

But whats the prompt

astral goblet
#

danbooru tags, loras, etc.

lime bane
#

Whats a danbooru

#

Loras to?

astral goblet
#

danbooru is um, its an anime porn site it turns out. i asked the same question, and it's a huge rabbit hole

#

butttt, it's well organized and all of the anime models are trained on sets like it

lime bane
#

Yk how to like generate a specific anime Character

#

I legit comment on whats the model and prompt and settings but they won't reply

astral goblet
#

so you use the booru tag for that character, cause all the images of it are trained with that tag

#

also loras

lime bane
#

U mean prompt?

astral goblet
#

you put the tag in the prompt yeh

lime bane
#

Which tag

astral goblet
#

the booru tag

lime bane
#

Like this?

#

Masterpiece, danbooru, loras?

astral goblet
#

no

lime bane
#

I cant comprehend alot about computer stuff only a bit

astral goblet
#

anyways, i've given you enough breadcrumbs to explore and discover! glhf

lime bane
#

So yeah you can get annoyed by me

lime bane
#

Well time spam there comment section again

vestal dew
#

got a lot of no help, huh?

nova silo
#

Hello. Im not sure what it's called. Are there freely available models that are able to preserve face, clothes, body of actors in generated images between different prompts? Preferably for anime. Maybe some prompt helper that adds very verbose descriptions of generated 1boy/1girl

vestal dew
#

controlnet, I think

#

I haven't used it, however

astral goblet
nova silo
vestal dew
nova silo
# vestal dew custom actors?

Imagine your actor has blue hair, certain face features, some estimated height etc. I want to show this actor in different situations

vestal dew
#

Controlnet may do it

#

same seed, same description, especially if you use some mix of existing characters from used tags like the booru tags

#

I am no expert on controlnet because I cannot run it

grizzled harness
#

What do ya'll do with your generated images collections?

vestal dew
#

that link is specific to your topic, not a double dog link to unrelated info

vestal dew
grizzled harness
#

I curate out the ones I like, but have amassed so many at this point it's sort of silly...

vestal dew
#

I do too, and then I'm left with hundreds I like, but none are perfect

limpid grotto
#

It’s a few years old but has 24gb of vram and should hold me over till a Ti or Titan drops

#

Prolly not worth upgrading to the 4090 for me. But that’s the card I’d get if I was buying today

vestal dew
#

right

#

the 24 VRAM is important

astral goblet
uneven heath
vast ingot
#

Hey SD gang ! How is it going this fine time of the day ?

crystal patrol
#

I went for the 4070ti because I read db can be used with 12gb now, but weiui uses 4gb just having it open

lilac reef
#

That first one going from 14 to 18 is a bit big but in the real world not noticeable I have been told, and if it is not by much.

astral goblet
#

doesn't them both down to a common compatible speed?

#

this stuff is the limits of my hardware know

lilac reef
#

dual rank is slower (twice as many chips) than single rank so the single rank has to bow down to the dual rank.

#

due to more chips timings has to be slower. We are speaking of nanoseconds

#

adds up though. Funny thing is a 3600mhz CL18 is exactly the same speed as a 3200 CL16

#

I was just lucky. TBH, 48 gigs I will trade any imperceptible speed loss for.

#

When I go to bed I get to ram test all night.

astral goblet
#

i've got 32 and my usage has never had issues. i'll probably just keep that bank size and worry about high performance ssds and timings next

#

i figure i'm throttling my system performance by 15% with the old dimms i've got in the board

lime bane
#

Does anybody know how to generate a specific ANIME CHARACTER

#

The people who post ai generated anime characters from tiktok wont tell me how they do it

#

I have the same model yet they idk how they generate a specific anime character

vast ingot
#

Anime is one of the biggest community around here, you should ask there, you'll get answers easily. Sorry not sure, outside of using a dedicated model or some embeddings for that character.
#1072013871730131004

lilac reef
astral goblet
#

gl

lilac reef
#

Yeah, I think the fastest this board can do is 34xx but if I had a second gen ryzen I could get the next bios and do 3600.

#

It does see it as 3600 and has the xmp for it but this board can't handle that without the newer bios. All good.

#

Now when I gen and it goes into ram it shouldn't be eating my nvme m.2 anylonger

uneven heath
#

you can use Waifu Diffusion 1.5 and generate Textual Inversion embeddings

exotic acorn
#

Can i use controlnet with my face module ?

vast ingot
#

Depends what you mean with your face module, but from what I saw for now, there isn't anywhere where you can't use controlnet in auto, except in upscaling or training

exotic acorn
#

Btw .. i have been using colab sd 1.5
On 1200 steps i get a great model but on 2000 i get bad

#

13 picture of my face with iphone 13 pro max

#

And suggest ?

#

Any *

vast ingot
#

ho ok. you meant a model you trained

#

well yes, you can use controlnet on those, no problem for that

#

as for your question on training itself, there are multiple things to know there

#

the quantity of steps you need will mostly depend on how many pictures you used, and on the "Learning Rate", those are 2 of the main training parameters here.

#

after a point, yes, training strats to get worse

exotic acorn
#

1a-6 learning rate

vast ingot
#

it's called "overtraining", and there are some things you can do to mitigate that, but not a lot

#

mostly you can try and use more varied pictures of your face. think different lights and backgrounds

#

second you can add regularisation data, or class data, it's called different things depending where you train

exotic acorn
#

Wheres the best place to train

vast ingot
#

it's another "concept" than your face, it's complementary data to train on so that the model doesn't go full "tunel vision" on you

exotic acorn
#

Sadly i got only 6 vram on my pc so i use colab

vast ingot
#

I said "where", I meant "how", what tool you use. even on collab there are multiple options

exotic acorn
#

This one

vast ingot
#

and it's hard to say what is the best in a generic sense, but given what you are saying, the easiest to get how all this works is the best right now, this can be quite coplicated sometimes. I use TheLastBen's version of dreambooth for that personally

#

nope another one

#

but it's about the same

#

Shivam is really good too

#

I made most of my models in it

exotic acorn
#

U train on sd 1.5 to ?

vast ingot
#

mostly yes, having a hard time to get it right on 2.X

#

depending on the subject I'm training, I can also train on other things

#

like waifu diffusion is a nice base if you want to train anime

exotic acorn
#

Btw .. more then 10 pictures is needed ?

#

How many picture for best result ?

#

Of my face

vast ingot
#

as many pictures as you can while keeping quality and diversity high.
The way I make a dataset is, I start from an empty folder and I add 1 by one each pics I like from my selection, starting from the best.
Each new pic I add, I must answer myself "what does it add? what is in this one and not in any others ? ". If nothing is added, I don't pick that new one
Same for duplications. Everything that repeats too much gets trained. so for your face it's cool but if it's always the same wall behind ? it gets trained too

#

10 is enough

#

for face, I go from 6 to 12

#

and I train for 100 steps per picture as a base, and add up to 50% more steps

#

so if you select 10 pics, that would mean training for 1k steps first, start doing checkpoints from there and train 500 more steps

#

compare the different models it gives and keep the best one

signal trail
#

so I’ve been looking into SD a bit, and Automatic 1111 keeps showing up everywhere. What exactly is it?

warm junco
signal trail
#

oh

#

so it uses my pc power to generate images essentially?

#

hm ok

#

do I need to train it to make images?

#

only stable diffusion I’ve used is the one through novel ai

warm junco
#

Nope, there are so called model files, these are pretrained

#

You can download and use them

signal trail
#

but if I wanted to I could make my own?

warm junco
#

Yes depending on your GPU vram

signal trail
#

hm

#

Well my laptop isn’t too powerful, but I’ve been really wanting to make an ai model I can keep consistent

warm junco
#

With 6-8gb vram you can train a lora that can help for consistency of a character

signal trail
#

oh god

#

is a Lora a trainer thing?

warm junco
#

Its a file usable in the webui (its a light model or an Addition to it)

#

There are a lot of loras for well known Anime Chars for example

signal trail
#

is it possible to have a Lora that can make multiple characters?

warm junco
#

Yes this is possible

signal trail
#

alr ty

#

u just made everything so much simpler, I appreciate it loveyou

warm junco
#

No problem, if you want to try the local webui (automatic1111) i can link you a tutorial and for technical Problems there is #🤝｜tech-support

signal trail
#

hm

#

would u say it’s the best one I can use?

#

or is it more of like a preference thing

warm junco
#

There are multiple Webuis, Automatic1111, InvokeAI, SD webui V2, Sygil webui, and Shark.

The most used and the most feature rich of those is Automatic1111,
so i would go with this it also can handle gpus with less vram pretty good.

signal trail
#

ok sick, I’d totally love to try it

warm junco
#

@signal trail Nice make sure your GPU has more than 2gb vram for it to work and has to be from nvidia, if your on Linux it will work with amd too or Shark.
Here is the tutorial for Automatic1111 for nvidia gpus:
https://m.youtube.com/watch?v=VXEyhM3Djqg

alpine mica
#

HI guys(these days it means Gils as well) 😜 , I saw some info on the internet, but could not find anything concrete on a API for Cinema4D, using SD for rendering. Anybody know where to look?

cobalt cliff
#

anyone tried sd generation and training on a 4090 laptop gpu?

fervent thunder
#

hey i wanna write text with controlnet how can i do that ?

#

specific one

vast ingot
#

they didn't use the same mode, you can get your text with different effects depending on the mode

#

this one is a great example too

maiden swan
#

how do i run run the programm lol?

#

with the webui-user.bat?

#

i have already installed it and used it once

blissful pendant
#

can someone controlnet a photo of a rainbow? i wanna see what happens

ionic copper
#

yooo wus new

faint plume
oak depot
bleak matrix
#

Good morning, everyone!

#

How are we all today?

cerulean halo
#

hello, Sunny! I am great

#

tired, though xD

#

a lot of exciting research

#

how about you?

#

(had a meeting with advisor to finallly release my paper.... he said he didn't understand anything and now I have to rewrite it after a long-ass meeting)

#

why is it so much harder to explain things than to make them xD

vast ingot
#

Hey Sunny !

bleak matrix
#

I'm awake. And h u n g r y for breakfast! (Suddenly, all I feel like acting like it's a 90's breakfast commercial. "I AM A KID!" She yells as the nabs the box of Trix.)

cerulean halo
#

HAHAHAH

vast ingot
#

little better :p happy to see you around ❤️

#

hehehe

bleak matrix
#

You know. I have been writing research. And I personally decided that professionalism was never going to be a part of it from the beginning. That idea was straight tossed out the window like the hot garbage it was...cuz ADHD 👏

cerulean halo
#

what HAHAHAHAHA

bleak matrix
#

But if you need help reading, I'd be happy to read

flint pulsar
#

How can I generate things here?

cerulean halo
#

not sure if I understand, you mean professionalism is writing research is hot garbage? That's an unusual take 😛

bleak matrix
#

No, I mean my ability to write professionally in one go is hot garbage from ADHD and I'll die from boredom

cerulean halo
#

unless you mean generating in your machine, then search for install instructions

bleak matrix
cerulean halo
bleak matrix
#

I write pieces, usually. Typically I write bulletins to organize the information and gradually string it together. It looks like a hot mess during the process, but once you thread it together, it works

#

And if I stop mid sentence because my brain just dies because it decided it's going to work on another sentence, I just do that, and resume when my brain is ready for that sentence

#

It looks like an apocalyptic landscape while writing. I'll color code or highlight things to keep organized.

lofty parcel
#

You know, I think AI like this will soon lead to an explosion of personalized media and we won't have to wait for companies to produce it anymore
But knowing how capitalism works they'll probably try to lock it down somehow

#

But what if I really want the Mario Infinity War crossover

bleak matrix
#

I don't know the future. All I can say is that, imho, we can continue to build and do cool things, and show those cool things to other people and innovate and help people

#

Which, frankly, ya'll be cool

#

So keep rockin' on ❤️

cerulean halo
#

I just sit down for 6 hours and write the whole thing continuously

cerulean halo
bleak matrix
#

So, what is your paper about?

cerulean halo
#

as in novels I wrote to learn

#

what's your novel about?

potent spire
cerulean halo
#

so like suppose you want to do that yellow bird among white flowers example from last time

#

you could generate an embedding for a yellow bird and then put it with white flowers and the model wouldn't mix up the colors

lofty parcel
cerulean halo
#

but you don't need previous images in my method

lofty parcel
potent spire
#

@lofty parcel oh, well im fully on that side then

cerulean halo
#

kinda like this

#

oh I can't put images here, you send as DM

mossy ridge
#

is there any guide for mixing models?

cerulean halo
#

will send*

cerulean halo
bleak matrix
lofty parcel
exotic acorn
potent spire
cerulean halo
lofty parcel
#

Oh

bleak matrix
# cerulean halo what's your novel about?

It's a fantasy, somewhat dark fantasy, I suppose. But not traditional, per se. Like me, very different. Each character has their own issue to overcome, but you get to see their thought patterns, reasons for being, as if you were that person. It's about learning to reconcile with the difficult things in life, especially the unfair things, and choosing to do good anyway. Also, there's many dragons in it, a lot of magic, somewhat of a steampunk setting, and characters of all ages.

lofty parcel
#

Personally I like the part of the Internet where people can make fanart and memes because it would be too much trouble for companies to shut them all down

cerulean halo
#

though it doesn't really read like a suspense thing, but rather like idk a heroic fantasy novel xD and, on the core, it's philosophically about apathy and a lost capacity to dream being restored. (e.g. "why a future?" and that question being resolved as the story progresses)

bleak matrix
bleak matrix
cerulean halo
potent spire
bleak matrix
cerulean halo
bleak matrix
lofty parcel
#

I guess you could argue that once we can make things indistinguishable from official sources we could have some sussy Mickey Mouse propaganda psyop

bleak matrix
cerulean halo
potent spire
#

Glaze is a start, but not the end

cerulean halo
potent spire
#

Oh, i meant STOP them from become a reality

cerulean halo
#

suuuuuure.... 🤣

lofty parcel
#

I mean I just want the freedom to make my Among Us Infinity War Super Mario spinoff personally

potent spire
#

You can, but you cant showcase it as your own creation and cant profit from it

bleak matrix
lofty parcel
#

Media will become so saturated that there probably won't be any more profit to be made lest companies lock the tech behind a paywall and copyright holders gain a share of profits

bleak matrix
#

I have plenty of self doubt otherwise, hahaha!

cerulean halo
#

also so curious to read it now, especially to check if brag matches fact! And also because it sounds good

potent spire
#

I hope for Adobe to do their job on this

#

And others too

bleak matrix
#

I don't mind sharing most things with ppl lol

cerulean halo
lofty parcel
#

The thing is, content never has to be published for people to consume it for their own enjoyment
And once that's possible, those companies won't make any money from their own creations if the generator is good enough
That's why they'll want to stop people from getting technology that powerful

#

Artificial scarcity would be required to maintain the current system

bleak matrix
potent spire
#

AI at some point could become a godlike entity

potent spire
#

A radical change of order

lofty parcel
potent spire
#

If AI becomes this powerful it should be used for areas some people wont like it to be involved to

lofty parcel
#

I don't think it requires AGI, I think we're already coming up on it this decade

#

A longform video synthesizer would be nothing short of incredible, and I don't think we're too far away

potent spire
#

I want AI to be used in several areas before AGI anyway

#

I dont talk about some "fluffy" stuff only tho

lofty parcel
#

I like fluffy stuff
SD can make some very fluffy cats

charred mesa
#

I can't wait for UniPC sampler to be in AUTO1111

bleak matrix
vast ingot
#

sure, what's the problem ?

#

and what's the program I guess :p

cerulean goblet
#

yes

pallid leaf
#

where i can generate pictures from text

true knoll
#

a

vast ingot
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.

slender field
maiden crystal
#

Question how do I uninstall SD ? I plan to install it again later under Invoke Ai.

#

I'm using A1111 currently.

vast ingot
#

it's a program that "runs", but doesn't install in the classical sense, not A1111

maiden crystal
#

I'll start with deleting the folder.

#

Is it installed into Python ?

#

should I uninstall that also.

vast ingot
#

I can confirm to you that it is enough for it to disapear, no need to delete python

#

plus you need python for invokeAI too I think

maiden crystal
#

Yea.

polar summit
#

sorry to bother everyone but where do i put my lora file in stable diffusion

vast ingot
#

in the LORA folder, I believe it's inside the Models folder by default

reef onyx
#

Vb-n

maiden crystal
slate spire
#

Hello all! I'm starting to use ControlNet and it's cool! But i dont really understand thé difference between canny pose, normal, hed etc. I understand thé difference in the process of creation but dont really understand when WE have to use one of them rather than another. I dont really Saw différence between each results so i dont see how i Can decide one or the other? Have you Seen cases where one IS more useful than the other? If yes, un which situation?

vast ingot
#

french ? 🙂 damn autocorect :p
It's a hard question, people are still discovering it. We see lots of uses in for example :

fervent thunder
#

Hey I had a question regarding prompt creation how exactly do stuff like this work farm, ranch, ((night)), cloudy, what exactly does (( )) do?

vestal dew
#

(()) means the weight is 2x

#

it makes it more likely to express that, but it also means less likely to express other prompts

vast ingot
vestal dew
#

prompt order also matters

fervent thunder
#

Got It thanks I appreciate it I would always see it and never know what it did

vestal dew
#

if you're using A1111 you can do it like this

#

(prompt:xx) where xx is the weight

#

(night:2) makes the weight of the prompt 2

fervent thunder
#

Alright

vestal dew
#

you can also do [night:day:xx] where xx is the number of steps done with the first one, and then the rest of the steps are done with the second one

#

so you can mix things

fervent thunder
#

Alright once again thanks I appreciate it

vast ingot
#

I like to link relevant wiki parts

vestal dew
#

a lot of prompt info there

#

I was just about to post it

vast ingot
#

the wiki is one of their strong suit on AUTO

#

it's always been up to date or almost

vestal dew
#

also there are tricks like if you do "blue pants, long hair" vs "long hair, blue pants" - the result will be different

little yoke
#

You guys are tons of help. Loving reading all this

vestal dew
#

you can also mix these

vast ingot
#

you're on the server with groupies for this tech, it helps

little yoke
#

Is the seed value what’s important when you want the same image but want to slightly tweak it?

vast ingot
#

I count myself in at least

vast ingot
#

if you vary on one, you can check what it does

vestal dew
#

([fred:bob:12] wearing a blue shirt:1.5) weights the whole thing 1.5x, and then Fred and Bob would be maybe Lora tags you're using to get a certain character, Fred expressed for 12 steps, Bob the rest

#

resolution also changes the image, as does seed, sampler, and of course model

vast ingot
vestal dew
#

you can also mix models in the "checkpoint merger" tab

little yoke
#

I just started last week learning all this. It’s been a little overwhelming but I keep telling myself to just go one step at a time, build my skills and knowledge one brick at a time and be patient. It’s been exciting learning

vast ingot
little yoke
#

I will share for sure. Right now I’m dividing my time learning more about prompts using models I’ve found and then learning to create my own models. I really want to create my own models

vast ingot
#

it's a really fun part of the thing for sure

#

but it's best to try and understand the most of image making first, since it will be hard to understand what you do right and wrong when you start training without that

little yoke
#

Thanks for the advise

vestal dew
#

chatgpt works the same way - many don't realize.

#

You have to consider order, weight of prompts, styles

#

I dont understand why Tortoise TTS never progressed like this...with different models etc.

wide juniper
#

do you think it is good practice to change the LORA's file names to their trigger words?

vestal dew
#

if it works for you, then do it

#

but how do LORA names make you remember the trigger word when you're using the AI?

#

UI, I meant

#

why not put it in a spreadsheet?

fervent thunder
#

Hey, so I've been having trouble rendering stuff with stable diffusion. It loads all the way, but then I just get a blank screen instead of the image. I don't think it's because of too many prompts since I tried it with just 4-5 prompts on both negative and positive. Do you think it's because of my computer not being strong enough or a memory issue?

vestal dew
#

what are the specs?

fervent thunder
#

2070 super 16gb ram intel i7

vestal dew
#

I do it with an i7 and a GTX 970

#

so it isn't your system

#

Is the 2070 AMD or NVIDIA?

wide juniper
fervent thunder
#

Nvidia

vestal dew
#

your card and system are better than mine, and I can generate

fervent thunder
#

that's weird then

#

It happens quite a few times

vestal dew
#

maybe you're tryuing too much

#

too high resolution, hires fix

fervent thunder
#

maybe

vestal dew
#

try it 512x512 without hires fix checked

fervent thunder
#

alright honestly I think that might be the issue cause I've rendered almost like 9 thousand images all high res 😅

vestal dew
#

but you said you can't render, it just stops

fervent thunder
#

Yeah every now and then It happens at random

#

like Ill generate lets say 6 perfectly fine and then maybe the next 2 fail or something like that

#

Usually I just ignore it but since I'm here now I thought I might as well ask

vestal dew
#

then it renders fine again?

#

sometimes mine stops working, but then I have to start it again

fervent thunder
#

yup it's weird 6 will render fine then maybe the next 2 fail and then it renders fine

#

and then once again every now and then the render breaks but then goes back to being fine

#

I've only had one instance where it just completely doesn't render

vestal dew
#

Unstable Diffusion discord has a technical support channel

slate spire
# vast ingot french ? 🙂 damn autocorect :p It's a hard question, people are still discoverin...

Thanks for the answer and yes my laptop's auto-correction plays tricks on me 😉
So it's interesting and the logo is really top. On the other hand, I personally don't run the AI on my own computer but I use Replicate. So I can't mix several models at the same time. If I use Replicate it's because I have a website and I propose to my users to test the AIs via Replicate's API.
So I tried on Replicate to make a logo, I went on this page : https://replicate.com/jagilley/controlnet-depth2img (Ithink it's the good model?)
And I tried with a text written in black on a white background.

When I launched the modèle I got this error message;

Not enough memory available to process your request. Try reducing the size or number of any file inputs or outputs.
More details
CUDA out of memory. Tried to allocate 3.06 GiB (GPU 0; 14.76 GiB total capacity; 9.11 GiB already allocated; 462.88 MiB free; 12.26 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Do you know where this can come from?
With other ControlNet templates I get either the same errors or less sharp images... or I can't see the letters.

vast ingot
#

you'll need to increase the virtual machine's vram if you don't want to reduce the dimensions

finite junco
#

Hi

storm plover
#

How do I diffuse?

vast ingot
#

hey new people 🙂

slate spire
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.

vast ingot
# slate spire It's weird because the image is only 20.7ko and I already made the template work...

the image is small, but to make an image in SD, it goes through latent space and loads the model, and this costs VRAM,lots, GBs. Adding Controlnet loads a second model. there is a different one depending on the mode you are using, but all those up the VRAM you need to run this. So it could explicate the difference.
Still though, you have quite enough vram here, if you are in 512x512 it should work smoothly. Do you have xformers running to help with the vram ?

bold yacht
#

/start

vast ingot
slate spire
# vast ingot the image is small, but to make an image in SD, it goes through latent space and...

I don't know if you know Replicate?
The main thing is that users put models online (here ControlNetDepth) and then you can use them via an api, because it is exactly Replicate that runs models in the cloud at scale. So it doesn't run on my computer but on the contrary it runs on a Nvidia T4 GPU which should do the trick. On the other hand, maybe the model that has been deposited on Replicate has different settings than the one used by AVEROK ...

vast ingot
storm marsh
#

does anyone have problems opening sd via сolab notebook?

real ridge
#

hey guys, how to install on ubuntu OS?

latent flower
#

wow, Im back and lots of things changed...

#

thats cool

latent flower
ivory wind
#

how do i get to Dreambot-archive role

latent flower
#

new emoji! showlove noice

latent flower
real ridge
potent fox
#

Just another (hopefully helpful) hint from me: if you changed something and then generate the image, generate it once more. For me it oftens changed from worse to perfect then.

ivory wind
#

can anyone help get me to the dreambot archive roles? I would like to retrieve my images there

sleek forge
#

Does anyone have gradio live forum or Discord? its offline on google colab

normal coral
#

hey guys quick security question: how safe is installing controlnet as it isn't a safetensors file?

fluid plaza
#

Em que sala eu posso criar imagens?

potent fox
lilac bear
#

hello everyone, i would like to start with SD but i am so overhelmed with all the information so apologize upfront if this is a silly question. please can someone advice how can i know what is the official most actual version of SD?

latent flower
latent flower
glacial slate
#

Is anyone using an Nvidia tesla card? Are they worth it? Are they more difficult to use in any way?

#

Considering something like the P40.

latent flower
# lilac bear hello everyone, i would like to start with SD but i am so overhelmed with all th...

no problem, feel free to ask. SD is just a method to generate images. If youre refering to the SD models, then the last version is 2.1 and you need something like the WebUI to work with it. If youre planning then, to gen some imagens, Id install the WebUI, since its the most complete interface to use SD. You can use any model you want with it. Theres also InvokeAI, it used to lack lots of features but its really good these days. Most ppl just use the WebUI tho. You can get it here: https://github.com/AUTOMATIC1111/stable-diffusion-webui

Then follow the instructions for you build. If you get an error or something just ask for help in #🤝｜tech-support (but remember to search the wiki and the issues on GitHub first, most of the time the errors are already solved there)... So yeah, its pretty much installing the WebUI, downloading some models from hugginface/civitasai/models-embeddings channel and having fun...

lilac bear
latent flower
sick python
latent flower
#

lul

oak depot
#

?

#

you cant be serious

sleek raven
#

I am curious

#

How many images were processed by sd2.0?

#

Thousands?

#

Millions?

latent flower
sleek raven
#

So that means

#

I can throw billions of images to my stable diffusion

latent flower
#

laion 5b

sleek raven
#

And it will survive

#

Right?

#

Perfect

latent flower
sleek raven
#

You drank my milkshake?

#

Dont drink my milkshake please

latent flower
#

I did! 🧋 waow

sleek raven
#

Gggggaaaaaayyyeeeeeee

latent flower
#

😎 slurp

sleek raven
#

I am blipping 6400 pictures

#

Then... I am going to throw it into the machine

little yoke
#

Question for everyone. Other than 512X512 sized renders what other sizes have you all had good success with? 512 seems to be the most successful for me

fervent thunder
# little yoke Question for everyone. Other than 512X512 sized renders what other sizes have y...

Depends on what resolution is model trained on , most of them mention which resolution works best.
For 1.5 it's mostly 512x512, yea..

Making it wider or higher is inconsistent, but sometimes works...
There might be a trick to try to pull higher resolutions out of model, but eh. who knows.

I did some runs with 1280x700
No idea why this numbers, but it it did okay-ish somewhere around 33% of the times
(maybe I got lucky and it's actually less)

Anyway...depends from model.
Look what author is saying, what people prompting, if there's nothing , you can look at models it made with to see it's resolutions...

little yoke
#

I finally figured out what I had been doing wrong with the model I trained - novice error / omission in my prompts. Im very happy with what it’s doing now. How do people generally share their work here in this group?

fervent thunder
warm junco
hollow jetty
little yoke
past edge
#

You can increase ur luck by training a model on 1024 if u have the vram for it

#

768 models suck ass for the most part anyways, so until we get a decent model, 512 is the norm which can be stretched consistently to about 678 before it starts doing weird shit

#

Keep in mind that all outputs are high fidelity, any output will look way better than any jpeg out there when resized 2-4x its original size.

near silo
#

Hey guys, anybody know how to run a publicly accessible SD without using Gradio and their crappy services?

past edge
#

Wut

#

Use cli ig