💬｜general-chat

[winter viper]

hey, how to use stability ai, i am new in this

[pulsar carbon]

Hi all! Can anyone please help me with controlnet? It's no longer showing the generated maps in the output when finished, i only get the result image. How do i get the maps to show alongside the final image like before? I think this changed after new controlnet update. I've consulted reddit and huggingface and haven't got an answer in weeks, i'm tearing my hair out. Thanks in advance.

[broken gull]

It's not necessary faster right? It just allows you to generate multiple images at the same time which could make your generations faster.

But sometimes it is slower no?

Depending on how much vram you have and what res you use

[abstract quarry]

might depend on your hardware but usually it's faster

[broken gull]

I think there is some kind of treshhold to when size becomes faster but only when you have quite some vram and plan to do a lot of generations but I haven't played a lot with this I am mostly using InvokeAI

[abstract quarry]

I guess there is a point where you have enough vram to process but it gets horrible slow. Dunno why, maybe extensive garbage collection

[prisma horizon]

hello guys

i can ask her for some help with stable diffusion?

[steel garden]

Hello, does anyone know a good tool to convert video to ai animated video? The tool should be free and preferably work over the web so that I only need the video as input.

[broken gull]

#🤝｜tech-support

[broken gull]

Deforum

[steel garden]

Is there an online website of stability or is it only possible with a colab

And how high should the frames per second be and the video for it to work well?

[tawny sleet]

Can ChatGPT generate Stable Diffusion prompts?

[next haven]

Please tell me what am I being asked to do?
error: Your local changes to the following files would be overwritten by merge:
extensions-builtin/LDSR/scripts/ldsr_model.py
extensions-builtin/ScuNET/scripts/scunet_model.py
extensions-builtin/prompt-bracket-checker/javascript/prompt-bracket-checker.js
Please commit your changes or stash them before you merge.

[heavy flame]

Well it's pretty much what it sais, you have made changes to these files that would go lost and it wants you to tell it what to do with those changes

if you do not need to retain your local changes you can revert them to how they were and then re-execute what you were doing

[next haven]

I have no idea what I'm talking about. What I did, what needs to be returned. I won't touch anything.

[broken smelt]

Assuming you're using a1111, there should be a setting specifically for this in the Settings tab that you can look for.

[red skiff]

So apparently

Colab banned sd webui

So when people use it it will give red arrow and die

[tall furnace]

what’s up

[red skiff]

Pretty sad

Currenly i have a fix for it but when the fix blows up google will get to it

[primal storm]

Good morning all!

[next haven]

How to make it so that 2 "Styles" tabs are displayed in the interface?

[peak portal]

run my discord bot on colab 😄

it uses a custom WebSocket client to become a worker node for a master Discord bot controller that runs somewhere else, so, techncially not running disallowed code there

[copper crystal]

not banned. Premium users can still use it.

it makes sense. the vast majority of webui users weren't using it for academic/learning purposes

you can still use SD with a free colab if you use scripted ones without a ui

[young ivy]

When you rent a server with let's say an A100 80gb, does it start with a remote desktop or something? Or does it startt with a vu

a custom thing

and you can then kinda pick windows there and whatever

I'm wondering how much work I have to spend on it before I can install my trainer and upload the dataset

[peak portal]

windows is unlikely lmao

you will want to start small on a free Colab or Paperspace notebook and get your training stuff working before you go blowing a budget on an A100 setup

[feral loom]

generally just ssh access to a vm with some flavor of debian or ubuntu LTS

[radiant hollow]

When will we be able to use the newest models to generate images with API?

[floral kestrel]

hey there clip interogator on hugging face is off for everyone or just me ?

[hallow plaza]

Anyone use backblaze and runpod? I'm having an issue where if I try to upload a folder in my bucket I get an error. I that common? Is there a way to upload less than my entire bucket?

[sick onyx]

how to install BooruDatasetTagManager?

[analog lodge]

Got couple questions.
Anyone know the hardware requirement for highres pipeline?
Anyone has any success upscaling anime style images with diffuser python api?
Would like to learn more about the nature of models. If i have a model for t2i inference, can i use the same model for inpainting / upscaling?

[lilac reef]

Is there a tool out there, besides Gimp, that will do 2d images (.png) to depthmaps?

[floral kestrel]

nvidia gpu works better highres is pretty limited on AMD card. if I uderstood your question correctly and you talking about using highres upscaling. ( personnal feedback )

[meager nest]

Anyone gettign this error AttributeError: 'function' object has no attribute 'weight' when Im trying to use my lora

I just updated automatic after not using it for 3 months and now its borked

for loras at least

[lilac reef]

1.5, 2.1?

tbh, go to extensions and grab the lycoris extension.

[hallow plaza]

How much of a difference does it make in render time between a 3090 and a 4090?

[lilac reef]

57% faster in inference.

in the real world about 35% because not all optimizations have been done yet.

[lilac reef]

New vs New you bet. Used sight unseen get the 4090 or play Russian Roulette these days.

it will use the last 16% of the die that is currently vacant for about 1800-2k USD MSRP

With how bad AMD, and Nvidia, is having at unloading cards it may never be made.

I said MSRP, not real world

Well, jacking us all off with their hideous pricing the consumers are telling them to take a hike

btw, the MSRP (good luck getting one close to it) for a 4090 is 1599 USD

Let me say what I heard Jensen mention that the 5090 will be 2-2.9x faster than a 4090. Now, consider the pricing rumor puts the card around 2500-3k MSRP.

This is getting beyond ludicrous.

bet me, and pay up when they do

24GB

they refuse to budge on vram due to their business lineup being 32gb+

Now they are resorting to gimmicks with compression so they don't need to give more vram

will come with the 5k series cards

Yeah, I am super sad

Game companies have yet to even say they will embrace this texture compression nonsense.

the show at the end of the month comes the official announcement.

On a 4090 this compression will take 1.5ms to do vs the non compressed of 0.47ms

1.5ms is hideously long

on a 4090 too, wtf?

so, if the 5090 is 2x faster that would make the decompression almost as fast as currently non compressed textures take.

again, get nothing more, some less, and pay more. YAY

hardest hit will be the competitive esports people.

I just want a 24GB card for SD

FPS

latency is your enemy

I can't they are used and I don't buy used site unseen

tell that to the esports people.

Where I live you get nothing used. They either used them till dead (any and everything) or they throw them in the trash (sometimes find good shit at the curb).

Used push lawnmower half rusted 200-250. Well, 300-350 new. Yeah, righto to them.

I am saving for a 4090 then when I get everything I am done for 12 years. Thanks Nvidia because of their hideous pricing my every 3 cycles (6 years) is now 6 cycles (12 years).

yes

nvidia and amd both used the crypto for big gains and refuse to let that gravy train go

Here we are in the states 1 bank failure more to a hard hard recession and the world is teetering on a recession to a depression and they are asking stupid prices

this isn't 2020 this is 2023 time to let it go

My thought has been, since Dec, that come the end of summer prices will drop some. We just need the 4060 to come out first. Instead of lower prices, though, they stop production for a month. We shall see what happens.

[meager nest]

Do yall know how to get less blurry images on models? I feel like on certain models it feels like its slowly deepfrying the images basically

yet I see people with 75 steps and crystal clear images, I get deepfried/blurry ones at 35

Its like each step adds to the deep fry

[broken smelt]

Do you have any LORAs or lots of high weight values like 1.9, or mabey like 20 1.5s? I've noticed that this tends to happen if LORAs are over trained and also when there's just too much weighting going on in the prompts. You could also try removing bokeh if you have it.

[meager nest]

Several loras yeah.

Also using embedding.. embeddings dont play nice withem Im noticing 😦

[fervent thunder]

hi

what does dreambot archive do

[surreal lance]

Guys I installed ROCm 5.5 on ubuntu but after restart i'm stuck in login loop. When i type my password and i press enter my screen goes black and i get to the login page again

Happened to someone else?

[unreal crown]

If someone has your png file image can they see what prompt you use or no?

[broken smelt]

if you have settings for the information embedded then yes. you can disable this in the settings. you can double check it by uploading the pic into the PNG info tab

[unreal crown]

What category would it be in

"save text information about generation parameters as chunks to png file"👈?

[broken smelt]

probably that yea, dont have it commited to memory. double check it afterwards in png info

[unreal crown]

Is there anyway to remove the embed from old images?

[broken smelt]

Yeah you can either just load them up into krita or photoshop and export them without any metadata or just run them thru SD again with 0 denoise and the setting turned off

[copper crystal]

ubuntu doesn't support the latest libs and packages. you probably broke something when forcing it to take a new driver library version like rocm. you can get it on there but then you're in unsupported territory and should know what you're doing.

if you're going to live on the edge and always install the newest stuff though, just use an arch distro. Makes it so much easier.

[sudden drift]

Has Peter Gabriel announced and shown the winning video for his competition yet?

[sudden ruin]

The Contest got extended to the 26th of May

[rugged mirage]

I'm getting really crap results with 2_1 checkpoint (which seems to be the latest official one) using the stable webui colab. Anyone have recommendations for recent general checkpoints they like?

[abstract quarry]

for 2.1 I use artius

[rugged mirage]

based on search I'm guessing this one? https://civitai.com/models/27739/artius-v21

[broken gull]

Where can I find more news on that social media platform project that emad was talking about a while back in chat?

[abstract quarry]

yes. There are also other good models, I haven't tried all. But artius improves results on a lot of different styles

[rugged mirage]

nice, thanks! Reading the discussions about it, it does seem pretty promising

[delicate tiger]

anyone knows what the going price for a rtx 3090 is?

[brazen atlas]

so everyday when i wake up i am hopping that i can get my vr gf through stable diffusoionajdn OPEN AI WHEN WOULD WE BE THERE

so if we combine 3090 and virt a mate and open ai we might see the light at the end of the tunnel

yeah

[velvet pivot]

will SDXL be open sourced?

[wise ibex]

i cant selfrole

[peak portal]

what gpu

[broken smelt]

Cheapest I see on PCPartPicker is 1400, most are going for like 1600ish up to 1900 tho

No reason not to buy a 4090 at those prices tho, they're around the same price

But ya ideally you buy super cheap and buy used. Just pray the previous owner didn't mess it up lol

[delicate tiger]

ideally new? i'm always worried that a used one might have been used for mining

linus said it's ok? i thought mining will burn the shit out of the card

[neat condor]

swinRLx4 doesn't work with latest auto1111 torch2 anymore peeps?

[quaint tapir]

Is there a Colab notebook for GPT4?

[gloomy cedar]

even if a gpu is used for mining its capabilities arent really going to be lowered

theres a whole youtube video on how it doesnt do much to the card

[craggy geyser]

How can I generate images of existing anime characters? Like, if I want to generate a picture of Shogun Raiden from Genshin Impact, the AI gives me a dude with a samurai armour

[delicate tiger]

you gotta train your current model on images of raiden

[delicate tiger]

mind sending a link?

[gloomy cedar]

unfortunately no because i dont even remember what the video was called 💀

[undone marsh]

Someone with Automatic1111 installed that can prompt some images for me? Can't in Midjourney ( not allowed). Here the prompt type (tweak as needed) "Outside, Elon musk crucified with a big space rocket in the background, he surrounded by robots looking at him and hitting him, high quality, high resolution --version 5 --stylize 1000 --ar 3:2" Thank you

[craggy geyser]

Lol

[quaint tapir]

The modifiers would be different

[undone marsh]

@craggy geyser 😅 Could you do that for me?

[undone marsh]

Do you mean my prompt?

[quaint tapir]

Yeah the modifiers at the end will be different for automatic1111, I don’t think it uses MJ style ones but I could be wrong

[undone marsh]

Ok, no problem. Just send the prompt as a guide. But if to make it work has to be different I'm good with it. Just the intention to break internet 😅

Sent*

[knotty turtle]

Is there a way to convert a checkpoint model to LoRa directly in A1111?

[undone marsh]

Of course tagging the artist

[toxic oak]

What do you think is the best approach for Visual Novel type of content? (Like the one used as tutorial for ComfyUI)
Generate background and character together, changing their poses and facial expression through inpaint, area and controlnet?
The problem is, this will slightly alter the background.
I was thinking if it's better to generate background separately and then generate characters, and then manually edit them in? (not an ideal approach because of poor background interaction - like with a foot buried in snow, etc.)

[gray rain]

can anyone explains whats the difference between 2d and interpolation? or just throw me a link ill read up myself

[dreamy temple]

is there anysupport channel here

or is there anyone can help me setting up mine , i don't get error but still it doesn't work

[gloomy cedar]

what setup are you doing

[waxen pasture]

#🤝｜tech-support

[dreamy temple]

ty

[surreal lance]

7900 xtx

[surreal lance]

I needed to update to the new driver because i can not use my gpu with rocm 5.4.3

[copper crystal]

I understand that and it's a good reason for wanting to update. However, living on the cutting edge with ubuntu is unsupported. If you're going to do that, use arch instead. It's a rolling release distro instead of locked in versions that ubuntu provides. It's designed to update to the edge a lot better and the community support around that is so much more extensive.

At the end of the day, choose the distro you like the most, since linux is free and open. You can modify anyone you want, anyway you want. It helps to know what you're doing though. ARCH linux kind of forces you to learn while ubuntu tries to be more user friendly and more stable.

i'm kind of doing what i hate too. You ask for linux support and people are like "just use an entirely different distro!" which, i understand is a steep task. With the way SD is and what's needed from it though, Arch is just such a superior ecosystem than the debian/ubuntu side of things.

[undone marsh]

Hi! Is there anyone that is using A1111? I was trying to prompt some images in Midjourney but because of the content type I'm not allowed... Here the PROMPT: Elon Musk crucified surrounded by robots hitting him, space rocket in the backround, high quality, cinematic, high resolution. Anyone would like to generate these images? 😆 Thank you

[lilac reef]

Except they had two that were junk and one was utter rubbish and he threw it into the bin. rattled and shit.

the other was so degraded it could not be fixed even if we wanted to.

[fervent thunder]

The Chad has lost everything

[peak portal]

it's going to be tricky as heck

i had a struggle on Ubuntu 22.10 and 23.04 and 22.04 and i was on Vega 56

[fervent thunder]

@peak portal think I'm going to grab a m40 over a 3060 12gb. You think that is wise to run on a standalone server where times don't matter

[peak portal]

i think you're free to make that choice 😛

whatever you do just don't use a vega56

https://en.wikipedia.org/wiki/CUDA#GPUs_supported

looks like the m40 only goes up to CUDA 5.2... if you get one for free maybe it's worth it

[fervent thunder]

So prob better to get 3060?

More future grade

[bright needle]

hello! 🙂

I'm new here, I'm just curious if anyone actually tried stable diffusion with AMD GPUs? I know there are unofficial versions, but has anyone tried it out?

Hope this is the right place to ask

[sudden ruin]

There are a few people in #1011228477954998273 that made it work, if you search for it im sure youll find what youre looking for

[bright needle]

thanks!

[hot minnow]

is there an official stable diffusion instagram account?

[tepid island]

bruh, 3060 is like £350 and 6700xt is £350 with 6700 being £290, nvidia is not the choice for that, and dont take me as a hater coz i use a 3090 and will b getting a 4090 when prices drop, but amd r winning rn for that £200-£600 price point bro

[sudden ruin]

Yep there is, should be easily findable

[hot minnow]

@tall liondiffusion ?

[sudden ruin]

in one word yes

[hot minnow]

yes in one word

[broken smelt]

Honestly if you have a 3090, is there any reason to upgrade? Besides unlimited moneys of course lol. A 3090 should be good for at least a few years from now.

[sudden ruin]

Power efficiency is what comes to my mind

[broken smelt]

Hmm then you'd have to do some maths of how much power it costs to run a 3090 for x amount of hours vs a straight up buying a 4090 and doing the same thing
And then rationalizing it for a few years
Which honestly isn't that bad of a math question but still much more brainpower than thinking w/e and moving forward with the 3090

[rugged mirage]

it says 4-9c/hr for runniung 4090 at max at 450w so at 10 c * 20hrs a day * 30 days it's 60 bucks a month, so a bit of efficency up or down is only a little bit of a difference monthly and that's with near nonstop use at more W than most use

so I think you can mostly just ignore that part unless it's very close

[broken smelt]

Right so it makes sense if you do (4090 price) - months you want to run SD in a row * 60 with those numbers

Who wants to do math tho, man that's a lot of work

[rugged mirage]

yeh but the efficency difference is more like 10% I think optimistically ? so 6 bucks a month difference, and for 99% of people who wouldnt use it at max for so many hours all the time it'd be more like 1 buck a month difference from the efficency

[trail wing]

anyone using stable-diffusion-webui-depthmap-script ? I used to load .ply into blender but now with .obj the colors look messed up

[sudden ruin]

Keep in mind that power prices widely differ for each country

[tepid island]

well mine costs a fortune, UK is 34 pence a unit, so thats like 16 cents per hour, double what u pay, its cheaper to buy a 4090 in the cloud for 300 euros a month then it would b to run my gpu all day long

[crystal palm]

Hola!
Alguien en español sabrá como puedo generar imágenes en la web de dream studio, partiendo de una foto?

[sick onyx]

how to change the ip of the gui for kohya ss?

[sick onyx]

when training a lora, can i leave certain numbers out or does it need to be every position filled?

[rough canopy]

Don’t leave numbers away if you’re developing the MLM; they can have an impact of what you’re trying to teach to the regression.

[sick onyx]

whats MLM?

[rough canopy]

Machine Learning Model

You use LORA algorithm to do so as far as my knowledge.

[sick onyx]

iam trying to test what i realy need to get the details in so iam having seperate datasets that i want to combine

[rough canopy]

They need to have the same structure and format.

[sick onyx]

so basicly dataset1 basic shapes, numbers 1-40 and dataset2 is very high res close up numbers 50-60

when combining the datasets, do i need every position of the numbers of 1-40 filled?

or can i start the training with image 1-33 and images 50-54?

[rough canopy]

train each dataset, dateset1 can predict from 1 to 40 and the second dataset can predict from 50-60 yet knowing that if you combine them you can have different outputs with both or same models.

[sick onyx]

yeah iam doing plenty of testruns

trying to see what works and what doesnt

[rough canopy]

If you’re doing separated datasets, you can classify them to generalize the datasets, if you have an image with the number 23 for example you could pass it through the first datasets between 1 to 40 and the second dataset that is 50 to 60 combined this way the outputs gives you what you need.

[spice plank]

Is there a way to see what prompts the ai can't interpret? I'm trying things like "video game armor" armor from Skyrim, tera, etc and I keep just getting plate armor. Using rev animated btw

[finite wing]

Alguém br

[fervent thunder]

any nsfw here?

[fervent thunder]

this server is for sfw

OH HELL NO

that's not to say SD can't generate nsfw, you just don't post it here.

[dim ice]

Yoo best checkpoint for anime?

[abstract quarry]

I don't know about the lora ui you are using, but you should shuffle the data and train on everything at the same time

[sick onyx]

doing that

its getting better and better lol still i need the help on how to get the details right

going from 40 repeats to 100 helped immensly, not sure what iam missing at this point :X

should i train at 768x768?

[hallow nymph]

anyone use draw things on mac? what are the best settings for photos that are similar to MJ?

[sick onyx]

how to save every single sampling step in automatic1111?

[alpine bear]

abyss orange mix 3

[fervent thunder]

meinamix works great for me

[fervent thunder]

you mean midjourney? can you describe what style or kind of images you want?

[magic jungle]

Yes what is the tags for creating something more similar to midjorney?

[boreal depot]

I just make 3k with dropshipping now

[astral pivot]

does anyone know how deforum works

need some help

[tired garnet]

hello. i struggle a lot with deformed body parts using sd xl and api. I put a lot of negative prompts but still Im getting a lot of bad arttifacts.

e.g

prompt = f"RAW photo, a close up portrait photo of 26 y.o woman sitting in a classroom, long haircut, pale skin, slim body, background is a blackboard, (high detailed skin:1.2), 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3"

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

its getting me some bad eyes, legs, hands etc

what I want to get is a stock photo style of a woman sitting in a classroom. seem not complicated, right?:)

[fervent thunder]

use negative embeddings, like badhandv4

[crimson crown]

Hello, I'm using the stability.ai to generate the image.
But I need to create the multiple images at same time.
Like this:

https://prnt.sc/xyy1ZsgV72Wh

as this, all cup has same generated image

[shy lagoon]

is there a good prompt to recude the overlay of objects in a picture , for example flowers often overlay eachother, im not a native english speaker soI cant come up with a good description for that ?

[robust jackal]

Hi

im new to stable diffusion. Basically i cant generate images. it will stuck in queue. how to fix this?

[warm junco]

Provide a Screenshot of the cmd Window in #🤝｜tech-support

[robust jackal]

fixed it thanks.

[sleek oxide]

gm

I just got Stable Diffusion installed last night and ran a few tests. Pretty cool! Wish I wasn't on an Intel GPU though. Does anyone have any tips on optimization with Intel GPUs?

[peak helm]

Hello

who have logo database?

i need logo for company names PowerCon

Can anybody do it for me?

[fervent thunder]

#💬｜general-chat message
How can i fix this?

[fervent thunder]

which one? A380, A750, or A770

No

just using Input directory and Output directory

is not the same

https://youtu.be/NP_aDmzdWRk
I am following this video

at about minute 10

[silent lily]

Yo can anyone here help me with training an embedding ive asked in finetune but not many people check there so id be glad if someone could help. I can give details in the dms

[woven totem]

Hello people! i'd like to use Automatic 1111 WEBUI on alternative Free Cloud GPU, cause free google colab is not more available for it.

does anyone have recommendation s?

or any free cloud GPU that uses Google Drive as repository

[frozen crypt]

Hi, anyone could help make a ruined jungle notice

A prompt, pls

[sleek oxide]

oh, nothing that fancy. I guess my all my graphics are handled by my i5 CPU on this laptop. Do you know of any optimizations for my situation?

[fervent thunder]

so you mean integrated graphics

sorry cant help you, honestly the best option is just to run a cloud hosted instance

[sleek oxide]

Cool, thanks. I figured as much but worth a shot.

[regal snow]

Anyone know a lot about sd-dynamic-prompts? I'm trying to make a wildcard where each line has a few comma separated phrases, but it doesn't seem to work.
I don't want to have possible 50 combinations in my wildcard - I only want a few possible outputs in my wildcard, each containing a few prompt chunks

for sake of example lets say I want my wildcard to have either of the following selections... but including commas seems to break it
blue house, white shutters, brown roof, grey door
white house, blue shutters, green roof

[plucky shoal]

[green roof|brown roof| yellow roof|red roof]
Shouldnt it be separated this way?

[regal snow]

Yes, but let's say there's a bunch of combinations that I don't want

eh, I just have a feeling that it's limited in this way

[signal dome]

i am late to announcement what does emad meant with ?
1§x

[karmic brook]

1§x

[silent lily]

anyone here trained embeddings before ?

[signal dome]

mm what

fine even chatgpt dont know then no one knows

[wild steppe]

if you know you know

[floral otter]

Hiya, I'm running Automatic1111 on my laptop with a GXT1650 graphics card on Win 11. Any performance tweaks I can make so I get the best experience?

I currently just run with the --lowvram tag

[silent lily]

is that a no?

[wild steppe]

sure plenty have! you can ask in #🔧｜finetune or check #1047197565365538826 and sort by the embedding tag

[silent lily]

i have and noone answered its been like 2 hours

[wild steppe]

try #🍥｜anime lots of fine-tuners are active there

[silent lily]

alright will do thx

[warm junco]

Sure, use --medvram --xformers

And also use the multidiffusion tiled vae extension for higher res

[fierce kraken]

Does anyone know if I Can install A111 on powermac intel?

[eternal totem]

I just installed A1111 to try it out and the results look fairly messed up, how can I improve the image quality: https://gyazo.com/bf7f675645f7954c732c6e9bb85d1631 Prompt used: "woman looking camera while the sun is shining her in the face, hyper realistic"

[fervent thunder]

can SD generate poses of a character based on a front view only? Mean i made the character, front pose, and im wondering if the 8 points turnaround can be done by Ai using the Pose Editor

[ornate dawn]

wow, finally!

[autumn onyx]

How to add LoRA ?

[shadow wraith]

vram stays occupied after nan check???? unloading the checkpoint then reloading seems to reset it back to normal (faster than resetting a1 anyway)

[forest lily]

Discord Server

Hope that helps!

[thorn pilot]

Is there a setting in A1111, so when you send to img2img from txt2img, it sends the final prompt and not the dynamic prompt?

[toxic fjord]

hello everyone I have a problem with deforum, I can't get more than 90 fps in the prompt the error is:
Error: 'Extra data: line 1 column 4 (char 3)'. Check your prompts with a JSON validator, please. The full error message is in your terminal/ cli.

I have no idea how to validate this JSON. If someone does please help

[peak portal]

using ControlNet Pose

i think that's what you were trying (Pose Editor?)

[silent lily]

does anyone know how to train an anime character with TI embeddings ? because ive treid everything and every time it looks nothing like the source images

ive asked everywhere and noone can figure out how to train anime characters with TI embeddings aparantally

[sly eagle]

Hello, where can I ask what pc is better for automatic1111 in quality and budget price?

[earnest escarp]

can anyone help with this?

[sly eagle]

I think you have maybe over trained your embedding, in finetune people can explain better

[silent lily]

wdym overtrained every 50 steps i get an image and none of them over the 3000 steps i did looked anything like my source image

[abstract quarry]

that sounds strange. Usually TI works quite well even if you just have a single image

you probably have to give more details about your training image(s), captions and settings

[spice plank]

Is waifu diffusion a checkpoint or something else entirely?

[fast sage]

What's the difference beteween a lora and a checkpoint?

[spice plank]

From my understanding a lora is used to sway a checkpoint in a specific direction, eg if you want a specific character or outfit a lora is better than hoping the checkpoint will understand

[warm junco]

Yes its a checkpoint

[thick trout]

when I try to run txt2img the data and sampling bar is stuck at 0% any idea why?

[plucky shoal]

1st image takes bit longer, what is your PC specs? @thick trout

[thick trout]

i5-12400F rtx3060

python scripts/txt2img.py --prompt "a professional photograph of an astronaut riding a horse" --ckpt <path/to/768model.ckpt/> --config configs/stable-diffusion/v2-inference-v.yaml --H 768 --W 768  

just using this command here

CUDA is at 0 for some reason

[plucky shoal]

i dont know much about using command line. Sorry. But seems o.k. to me. Probably something is missing. It seems to be work for CS1o ....

[thick trout]

are you using the webui?

[rugged mirage]

most people are, I'd set that up unless you specifically want to run it directly in the terminal so you can hook it to your own scripts

[thick trout]

ah alright sounds good, ill try the web ui first

[warm junco]

Yea let me know if you need a tutorial link

And for installation questions feel free to ask in #🤝｜tech-support

[thick trout]

already got an error haha

[verbal fable]

Hello where do i paste the "Copy Generation Data" from civitai website in SD?

[lethal spindle]

hi guys. do you know why stable diffusion was trained with square images? i was searching for that information but could not find anything beside "gpu works faster with power of 2"

[floral otter]

Thanks, that really does help

[warm junco]

Np, i didnt noticed you used lowvram before but dont use lowvram and medvram together. Only --medvram

You can also add --autolaunch so your webui opens in Browser after running the bat

[warped path]

couldn't find a channel to ask, is there a place for install help?

[plucky shoal]

#🤝｜tech-support

[abstract quarry]

guess because they were lazy...
you usually train large batches of images and when you do so, all images in a batch have to have the same size. If you want to train on different resolutions you have to write smarter training algorithms that split data in buckets of same resolution and then make batches from them.

still, training on square images has so many disadvantages that I don't understand why they haven't come up with a better training routine

[broken gull]

Train a checkpoint on 16:9 and 21:9 aspect ratios please

[ripe ridge]

hey, where do I add prompts to create art

[broken smelt]

Could be wrong but I'm p sure the SD models are all actual research models. I'd imagine it's much easier to work with the data that's all homogenous (512x512 in this case). Im fairly certain they're slowly upping that number as time goes on, 2.1 is 768x768 and I'm not even sure about sdxl

A fun fact that clicked after reading it is that they probably batch processed the cropping of 512x512 the millions of images that they trained 1.5 on, leading to why we have faces and legs cropped out all the time when we generate images.

[rustic nova]

Hello everyone. How are you feeling? How's life? I have a little problem.Is there anyone who understands CUDA PyTorch2.0 kernel installation. I previously downloaded CUDA 11.7, yesterday I downloaded CUDA 11.8 . Do I need to remove CUDA 11.7 from the folder where CUDA 11.8 is now? Because I am getting some strange changes when generating. Yesterday I completely reinstalled the SD as written on the AUTOMATIC1111 website. After that, in SD Browser I got the following message: python: 3.10.6 - torch: 2.0.0 + cu118 - xformers. Which means, sort of, that everything is fine and everything is installed correctly. I first installed with "--opt-sdp-attention". I also tried with "--opt-sdp-no-mem-attention", but the results were not very good. I have a 4080 card.Now I have xformers, by the rules of "facebookresearch/xformers" on GitHub.
This is what I get when I generate 1 image.

[long grove]

hey

[abstract quarry]

as far as I know, pytorch comes with it's own CUDA driver. So it doesn't matter.

[abstract quarry]

yeah, but it's not about the overall resolution. Of course it's much more expensive going for 1024x2024 pixel than 512x512 pixel. But they could have allowed for different aspect ratios (e.g. 768x512 and 512x768). Of course that's all not that easy in practice as you want your gpus run at full throughput, so each batch should run at similar time and memory requirement. But I think with a few engineering tricks it is possible to implement bucket based training.

[scarlet crest]

hey

[floral dawn]

cunts

@karrssteun

was the goat

[rustic nova]

I had 2 folders with "CUDA": "11.7" and "11.8" in the same CUDA folder. It's not easy. I wrote the path to 11.7 three months ago. I have now rewritten the path. Removed the 11.7 folder and left 11.8. And in the path I changed 7 to 8. I don't know if that's right or not? It seems to be working now. But I don't really like it. It's really bad at rendering faces.

[broken smelt]

If your webui shows that you're using the latter one then there's nothing to worry about. The older one not being used at all is why I simply deleted it when I updated my webui.

Older version of cuda is just taking up space and will likely never be used again

[rustic nova]

The problem is that it doesn't draw all the steps. I set it to 100 and the SD does 40 and that's it. I set it to 40 and the SD does 38. This is not always the case, but sometimes the image can be seen as incomplete. That's why I asked, can I remove 11.7 altogether? But I don't know much about it, maybe it won't work at all. I will have to reinstall everything again. But it doesn't give any errors when opening the SD. Could it be a pytorch 2.0 thing?

[broken smelt]

Unsure, I have the preview turned off since doing so actually increases generation times.

But yeah I would suggest a quick reinstall if you feel like that's the issue, should only take a hot minute.

[rustic nova]

That's probably what I'll do. I'll remove all CUDA 11.7 - remove it completely from paths using "sysdm.cpl" command and see what happens. If not, I'll clean out Windows completely and then reinstall SD. Thanks for the help.Have a good evening.

[low prawn]

hi need help

Im addicted to the genration button

doesnt matter the prompt

doesnt matter the seed

as long as i can generate im chilled

ive stopped gaming

ive stopped it all

its just about 1 button now

any experience in this?

[fervent thunder]

everyone gets bored with stuff sometimes. Or if they don't, then I haven't met one yet but that might be because they never go outside because of the issue :P

[low prawn]

hmm, interesting, im close to that, wonder what might seperate them and myself

im thinking gpu capability o0

must be

do they also pray that gitpull seeks no error

i do..

[fervent thunder]

How would I do img2img of a portrait where it just enhances everything?

[low prawn]

maybe #1011634831467221033

[copper crystal]

the sdk you've got installed is what developers use to "build" pytorch with cuda support. i don't think you need them installed just to use pytorch2

regardless, i've got 11.8 and 12 installed side by side

[copper crystal]

this might be in image to image. see, originally, stable diffusion was never meant to do image to image. it's sort of hacked in there. heres how. So, it looks at your step counts. say, 100. Then it looks at your denoising. say, 0.7. so it to make the system work, img2img will take 70% of the steps worth of noise, and add it to the image. Then it pics up from there. Which for 100 steps at 0.7 denoise would be only 70 steps. If you put a total of 50 steps, img2img at 0.7 would be start as if 15 steps of nooise has been removed so far, leaving 35 steps left.

Another possibility is the sampler you use. one of them decides how many steps it needs and just does that. i am unsure what that is, might be ddim?

[half latch]

“Connection Errored Out” does anyone have a solution for this? Thank you 🙏

[tulip yoke]

throw your pc out of your window

[peak portal]

the scheduler "converges" at a specific number of steps

ah, right, scheduler = sampler for some people

in fact most schedulers other than Euler A tend to converge at some point, some of them, earlier than others

[broken smelt]

I actually recall a setting in a1111 regarding the img2img step amounts due to this

[copper crystal]

heh i been saying scheduler and someone corrected me the other day so i changed. funny now that its come back to me this way. i like sched better i think.

[copper crystal]

yeah. it forces all the steps in that scenario. adds the noise but starts from step 0

[peak portal]

in the diffusers codebase they're referred to as schedulers throughout the whole source tree. i heard sampler only from people who use Auto1111

it really confused the bejeezus out of me.

also i wonder if that behaviour you refer to is specific to that tool because i'm not sure that's exactly how it works for my pipeline

it doesn't help that there's like 3 or 4 different img2img pipelines. there's Lambda Labs diffusers finetuned model, holy shit that thing sucks ahahahaha

(3 or 4 popular ones, i mean. there's probably 12-15 total)

[copper crystal]

yeah i'm having trouble navigating to the meat of the tech lately. a lot of the jargon is up in the air seemingly

think i'll go back to the basics and learn from there. try to understand the diffusers code

[broken smelt]

Ngl it's important to remember that a lot of SD is used by hobbyists. I'd personally just stick with whatever actual researchers use in terms of terminology and jargon lol

[chrome sundial]

Is this server offers art creation or just for stable diffusion community as I can't give prompt in showdown or any other channel

[wooden widget]

Anybody willing to support a guy and subscribe to my free AI newsletter where I share prompts, tools, news in a quick and hopefully humorous snippet?

https://www.theindifferentspectator.com/

[abstract quarry]

it's open source, just download it and create art on your own pc. Much more fun than using stupid discord bots

[surreal notch]

Are there any models for things like TTRPG battle maps? All the models I've run across seem to focus on characters and environments, but nothing like game token icons or battle maps.

[rapid yarrow]

Doe’s anyone know how to create images like these found here? https://www.instagram.com/feli.airt/

Or like this https://www.instagram.com/p/Cr_o8UFoU3k/ ?

[elfin pivot]

Any know any tutorials on how to make textual inversions ?

[copper crystal]

https://www.reddit.com/r/StableDiffusion/comments/zxkukk/detailed_guide_on_training_embeddings_on_a/ i like this guide for embeddings

[fading parrot]

not SD related, but is there anyone dabble with chatgpt that know something about prompts to jailbreak gpt ?

[hallow surge]

how to generate photos

[fervent thunder]

here I believe https://beta.dreamstudio.ai/generate

Can I host stable defusion on my computer ?

[rugged mirage]

yes

[fervent thunder]

Do you have links for the docs about that please ?

[rugged mirage]

first hit in google for 'stablediffusion install'

[grizzled flax]

Hello guys, i've started to use stable diffusion like 3 month ago, and managed to create this music video, i'm a musician from Kiyv, so hope you give me your opinion about this work, here , or better on youtube😌

https://youtu.be/yijYVFSGMQ4

oh, sorry , wrong chat probably

[bleak matrix]

Good morning, everyone!

How are we all today?

[bleak matrix]

#1072220168534642768 followed by #1080946152318443610

[fervent thunder]

Thanks

[tulip yoke]

Lets dance the amogus

[fluid zealot]

Hello everyone! did anyone experiment with gif2gif? if so have you figured out how to have the same character to appear in every frame? in terms of looks and outfit, like the only thing changing is the animation between frames, thank you!

[bleak matrix]

Np!

[bleak matrix]

You might want to try #🎥｜animation

We also have the animation contest going on right now, so you might find ppl who have used it in #1097876660101066854

[fluid zealot]

very kind of you! thanks i will

[wind hamlet]

I made this text-to-youtube-video(pseudo) kinda thing using chatGPT and dreamlike stable diffusion model

https://youtu.be/TKSJlrxPGp8

I would like suggestions for models and video animations

The dreamlike thing kinda does okayish work but I've been stuck in January/February in terms of model knowledge

Are there any other models I should use lol

[hallow plaza]

What has the largest impact on vRam usage? Is it the number of images, learning rate, batch size or epochs?

[rugged mirage]

image size and batch size

learning rate, number of images and epochs shouldn't matter for ram usage just how long it takes

[hallow plaza]

Does batch size impact how long it takes enough to justify a slower card with more memory?

[rugged mirage]

depends what you are doing really but a slower card with more memory can do everytrhing that a faster card can do just slower

a fast card with less memory is faster on everything but there'll be things it cant do at all

so it depends if you'd want to do those thigns or not

[hallow plaza]

In this use case which one would finish sooner? Larger batch size, slower card, Smaller batch size, faster card.

[rugged mirage]

not sure and depends on the case but probably larger batch size when a large batch size makes sense

[rustic nova]

Thanks, I worked it out yesterday and everything seems to be working fine now:)

[broken smelt]

has anyone played with the noise multiplier setting? i wonder how results will change with controlnet tile keeping things relatively cohesive

[waxen moss]

Is there a way I can search a model to see if a token is included in the training data? I see people refer to "using this word isn't useful, it's not in the training data" and when I'm troubleshooting prompts, I want to be able to find that out

[fervent thunder]

would anyone help me to understand how to pose a character (if possible) using Image2Image and Control Net? I made a character, front pose and placed in Image2Image. Took a pose template and placed in the Control Net with Open Pose model enabled and None in Preprocessor. Setting denoiser from 0 to 0.05 keep my character but wont generate the poses of the template. Setting Denoiser to 1 generate the poses but nothing to do with my character. Half way values generate weird artifacts. I can send screenshots or private messages will work to if needed. Thank you in advance 🙂

[broken smelt]

denoise of 0.05 in img2img wont really change your image much, try playing with the slider until you find something you like

[fervent thunder]

when i increase that value the generation has nothing to do with my character

im trying to transfer the poses on my character....im using an image i draw, im not generating the character

@broken smelt if u read the rest of what i wrote, I tried all the values of the denoiser....what I explain in there is that only up to a value of 0.05 keep the look of my character

[slate dust]

SD doens't know what your character is unless you create a model or embedding of it

[fervent thunder]

so is not possible to generate poses of a character based on a existing image...correct?

like change the poses without changing the look

[waxen moss]

0.05 for denoiser isn't going to completely change their pose. It's a pretty drastic change to the image. The best way to do what you want is to photobash it in my opinion. Cut out the head of your character, and paste it on someone doing the pose, then run that through image2image and controlnet again, this time turning on openpose full and probably depth, and use a low denoising until you get the desired result

[slate dust]

you could change the pose but it won't necessarily keep the original design

[fervent thunder]

@waxen moss at that point is faster create a 8 points turnaround by hands, that would be faster...i was trying to short cut

thank you for clarifing tho

[main salmon]

do anyone knows how deforum have been built?

and how can i make extensions relying on sd?

things like temporalkit and discodiffusion

anyone have anyidea?

[ionic cove]

so what is this discord for exactly?

[rugged mirage]

just read the name of the discord to find out what's it about

[lucid heart]

YOUR WHAT?

BAHAHAHSGAGA YOUR FUCKING WHAT

I love you already

[ionic cove]

bruh i got lost in the channels

and had to find chat again XD

[lucid heart]

Literally, like bro, who designed this place

It’s overwhelming

[rugged mirage]

I just muted some channels and collapsed some sections when I joined

[lucid heart]

I feel like you could really clean it up, so many of these feel unnecessary

I can see why my man’s don’t want this server back

It’s a mess

It’s also kinda sad looking how there’s so many people and so little activity…

Dang

Ight, Imma dip. This place is so dry I joined while having a coughing fit

[rugged mirage]

I didn't mean it as 'I mind it' really, I meant it as 'you can easily customize it to only see what's relevant to you and then it's manageable'

[spark patrol]

OH

THIS IS THE CHAT

WRONG ONE

HAHAHA

OOPSIESS

literally am so lost here omg

[bright wren]

can someone explain this so i understand it? the red one and green one are things i can update, right?

ok i cant share images.......

nice

well...... guess cant share my problem

[rugged mirage]

you can in the channels that are about asking for help like #🤝｜tech-support not in the first place you click on randomly

[bright wren]

gocha, saaanks. thought general would be a good start

[slender talon]

do you guys think that people will create an ai that takes millions of samples off the internet and uses those samples to create a specific genre of music

[stark swallow]

from what I understand, music is much harder for ai to generate than images/text. (good music that is)

[slate dust]

It's definitely already being made

[rapid yarrow]

What is a prompt I can use to make a person stand far in the distance?

Thank you!

[split scaffold]

fart diffusion

[spare spade]

how do I showcase my art

[hallow plaza]

Is there a tool that is better than blip captioning? Something that can generate highly detailed text?

[alpine bear]

WD1.4 tagger extension is pretty good, though I've only ever used it for anime... and it was trained on anime. So your mileage may vary.

But for anime, it is way better than blip.

[hallow plaza]

Thank you for that. I sometimes do anime, but for now I was looking for real people.

[abstract quarry]

that's called OCR and is basically solved many years ago. Don't know a specific model but there are probably plenty of them if you google ocr

[rugged mirage]

solved is pushing it, maybe nowadays but if 'many years ago' is more than 5 then there's definitely a lot of images and fonts even the best techniques tripped up on

[obtuse peak]

took me my whole day to get the program back up km$

[abstract quarry]

sure, you likely find images of text today that cannot be recognized. But I remember SVM based methods 10 years ago that could easily extract texts from images without even using convolutional neural networks

I just say you probably don't even need a super fancy model for that except your images are extremely challenging

[tiny spindle]

hi, new here!

what is the age limit in this server? (apart from the basic 13 y.o. discord already does with tos)

[broken smelt]

Idk if there is one lol, the server is pretty PG-13 imo since it doesn't allow nsfw

[tiny spindle]

ok thanks

[peak portal]

yikes

why?

[glass elbow]

Hello there!

how to createges using GANs ? its not helpful using midjourny or kinda boring as i haven'tthe paid version. can this server help me?

[tiny spindle]

because i might be below 18 and the rules didn't say anything about age limit so i thought it should be ok to ask

[peak portal]

it's a don't ask, don't tell policy on discord, hoss

[tiny spindle]

oh, well sorry

[broken smelt]

I'd say it's only half of that, no one will care if you're under 18 here lol

[fervent thunder]

Just don't say you are under 18

Problem fixed

[broken smelt]

That's implying there's problems if you're under 18 tho, there are none

[fervent thunder]

Vice article next week "how I infiltrated a stable diffusion discord"

[tulip yoke]

Guys

Did you read rule 32 already

[solar swan]

When did this server just turn into ads for raffles and contests and whatnot

[charred mesa]

These announcements are interesting but uhhh

SDXL download when 👍

[karmic brook]

the first part of the announcement is literally an integration launch haha

[karmic brook]

maybe soon-er than you think

[charred mesa]

thanks

I can't wait to realise that I might not be able to run it on my hardware

[karmic brook]

perhaps there will be some workarounds for this common fear

[forest vale]

api endpoints isnt an 'SDK'

[soft basalt]

how do I download Stable Animation

[fervent thunder]

SDXL hype

[next swan]

React... 🤣😏

[woven token]

hey, can somebody help me
I trained my model on my photos and it took quite a long time, but unfortunately i'm stupid and I forgot my original prompt

is there any way to check it?

in files maybe

[wild heart]

How does one get the dreambot archive role?

[karmic brook]

All the links in the announcement post will help you, along with #1098959457964277831

[karmic brook]

you don't until our #👥｜roles channel is reopened! Which will likely be back on early next wk Monday

[soft basalt]

oh the link in announcement was dead for me for some reason

[wild heart]

ah ok, thanks

[karmic brook]

updated and should be good now

[past mantle]

here used to be a chat you could get the pictures from

where is it

[strong lion]

hey where the bot

i want crearte images

pls

where the bot

@karmic brook bot

where the bot

help

staffs

[karmic brook]

please don't spam and ping
the bot is currently not available server-wide

[strong lion]

how i can acess stale difussion?

[karmic brook]

#1072220168534642768 and #1080946152318443610

[strong lion]

no

i dont want read

[karmic brook]

Then I cannot help you

[strong lion]

wft

you aer staff

[karmic brook]

and you are rude. If you'd like to get started, everything you need are placed in those channels to get you up and running. You have to read a bit in order to set up and make an effort

[strong lion]

sorry i was joking i am reading 🙂

[strong lion]

i cant acess with google colab?

staffs help pls

i want buy

Wow, what a lack of respect for anyone who wants to buy the service. I liked it so much, but until now, you haven't helped

[strong lion]

srry pls help me 😦

@karmic brook

[untold chasm]

Gzino - please be nice man.

[strong lion]

help me

[strong lion]

when I access the stable diffusion in the collab I can enter it but when it generates an image "error" appears

[untold chasm]

I would ask in #🤝｜tech-support

[wild steppe]

ask in #🤝｜tech-support and please be respectful

#✍🏼｜rules-and-tos

[wispy jasper]

How's stable animation better than existing tools already available?

[untold chasm]

If you are nice to people here - you will find many are incredibly talented people who will happily help you.

[strong lion]

thank you 🙂

[untold chasm]

Good luck!

[strong lion]

thank you too but, did I disrespect someone??

[wild steppe]

you keep pinging staff and refuse to read through any links we provide you

[wooden quest]

I may be late to the game but when is Stable Animation launching?

[strong lion]

sry I was just kidding so it was sent I read it but I'm having trouble accessing it I sent the message as you requested 😦

[untold chasm]

Think there is a event in like 30 minutes which will reveal that info - could be wrong!

[wooden quest]

Good to know, thank you.

[bleak matrix]

Hello everyone. I hope you are all having a wonderful day. How are you all?

[signal dome]

stable animation or deforum diffusion ?

[wild steppe]

Stable Animation stage is starting shortly, head over to #1029055412764422214 to drop any questions and mods will pin all relevant questions in chat!

https://discord.gg/cyP4Vkse?event=1105889719168020570

Aaaand we're LIVE! 🔴

[signal dome]

i doubt it will support rtx 2060 6gb .not hoping

[wooden quest]

Great, thank you for asking. How about yourself?

[cinder herald]

hey, could anyone help me with making a telegram bot with stable diffusion? (WILL PAY)

[steep star]

How to install stable installation?

[naive kraken]

Can someone send medarc discord link? Link in the channel is dead.

[rocky burrow]

have they mentioned / estimated when Stable XL will be released so we can run it locally?

[wild steppe]

https://discord.gg/Ng2uReuD

[wild steppe]

it's in testing and continually being improved on, so no ETA atm but as soon as it's ready ^^

[rocky burrow]

Thank you. I'm really excited about this one. hopefully soon! 🤞

[wild steppe]

the team is working hard on it! i can attest to that haha

[fervent thunder]

where do i prompt?ç

can someone help me?

[surreal perch]

@fervent thunder

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

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

[bleak matrix]

Pretty good, all things considered! Healing, which is good!

[rose nest]

hello im new to this stable foundation thing

[autumn onyx]

What is the Twitch Channel they were streaming on?

[timid knoll]

https://twitter.com/EMostaque/status/1656748050612645902 Ok I'm so hyped on that answer... Maybe in a couple of years we'll have AI cards in our pc like we had 3D graphic cards

[karmic brook]

https://www.twitch.tv/stabilityai

[autumn onyx]

This wasn't where they were streaming today. the one that was on discord. How to find the replay?

[karmic brook]

It isn't a Twitch channel, but a YT channel where our stages get uploaded to.

It will be uploaded there when it's ready.

https://www.youtube.com/@stablefoundationai

[autumn onyx]

Oh Ok thanks

[torn wadi]

can i view the video? i missed the whole show..

[clever spire]

still i dunno how to install it can anybody help me

i have zero knowledge with programming languages but i just wanna animate

[torn wadi]

how do i download the sdk to my stable diffusion on my local machine?

[torn wadi]

did you find the replay?

[odd sparrow]

simple question: will an embedding trained using textual inversion on sd 1.5 work with any other sd 1.5 based models, for example the base model of Deliberate https://civitai.com/models/4823/deliberate is 1.5, so should/would my embedding work with Deliberate?

[karmic brook]

@torn wadi #💬｜general-chat message

[torn wadi]

can somebody please send me the replay video of the new Stable SDK animation that was today ?

[torn wadi]

but thats not the replay video of the new SDK that was just showed..

[karmic brook]

In my message I say it will be uploaded to this channel when it is ready.

[rugged mirage]

my speaker is making a quiet hissing noise when i click generate image, even when I switch it to headphones (it still happens from speaker)? Is it normal? Am I dying?

[torn wadi]

oh, ok.. lol my fault..

[real zodiac]

ground loop/bad shielding/psu/gpu working too hard, it's pretty common

[rugged mirage]

but how does the gpu working hard is making the speaker that is far from it make a sound

[real zodiac]

the gpu or more likely psu causes interference with the mobo

[rugged mirage]

huh

[wary bone]

I won raffle 😁

[real zodiac]

the psu affects the onboard audio basically

and it manifests as some sort of whine or hiss on the speakers

solution: go wireless (no really, it helps)

[rugged mirage]

weird, i wouldnt expect it to affect it so much every time to always make a hiss on a different component

wireless speakers? yeh definitely not

[real zodiac]

wireless headphones

or usb speakers should work too

alternatively some sort of filter

[rugged mirage]

even headphones I dont like, harder to ensure quality

and batteries are annoying

[real zodiac]

well wireless headphones often come with their own hiss

it's less annoying than the wired hiss from my experience at least tho

[west sapphire]

That announcement video had commands to install the new animation model. Is that working or was just a teaser?

[rugged mirage]

nah, no LED or at least I disable as much of it as I can. My amp is a bit wonky though so it probably takes it very little to act up

[willow coyote]

always adds hiss on purpose

[hallow plaza]

Are there any good Blip2 git repos that work well on runpod (with a GUI)

[copper crystal]

it's always the transformer that does it. where the ac turns into dc. led's themselves are relatively noiseless for radio reasons.

shitty ac adapters with poor shielding are problematic

i have crystal clear wireless headphones. they're not even quality they're just expensive gamer headphones. no signal, no line. They have internal processing that just cuts the noise

ANY headphones plugged into a hot line should have some 'hiss' to it. It's called line noise. it's always been a thing

[copper crystal]

yeh this one is cause the gpu is putting out hard coil noise from it's capacitors, and his speakers are probably plugged into the front of his case. so the little wire that runs from his mobo up to his front panel is acting like an antenna and ccatching a lot of that noise coming off the line.

in the past, i would get it if i was scrolling while using hardware acceleration in my browser. i've since gone to an all digital audio chain, until the dac in the headphones

the actual speaker wire may catch the gpu noise even if the case doesn't provide sufficient shielding

[spare kayak]

I'm sure this has been asked 9000 times but I got this error when trying to install the program.

Error code: 1
stdout: <empty>
stderr: Traceback (most recent call last):
File "<string>", line 1, in <module>
AssertionError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

Is there anyone that might be able to walk me through this?

[broken smelt]

hopefully someone in #🤝｜tech-support can, might get a better response there

[spare kayak]

Thank you so much 🙂

[copper crystal]

any unshielded cable will act as an antenna. it's just physics

[oblique kelp]

Hello, I'm new here, and I try to run stable diffusion without internet and this happend, Installing requirements for Web UI
Traceback (most recent call last):
File "E:\ai\stable-diffusion-webui\launch.py", line 355, in <module>
prepare_environment()
File "E:\ai\stable-diffusion-webui\launch.py", line 299, in prepare_environment
run_pip(f"install -r "{requirements_file}"", "requirements for Web UI")
File "E:\ai\stable-diffusion-webui\launch.py", line 129, in run_pip
return run(f'"{python}" -m pip {args} --prefer-binary{index_url_line}', desc=f"Installing {desc}", errdesc=f"Couldn't install {desc}")
File "E:\ai\stable-diffusion-webui\launch.py", line 97, in run
raise RuntimeError(message)
RuntimeError: Couldn't install requirements for Web UI.
Command: "E:\ai\stable-diffusion-webui\venv\Scripts\python.exe" -m pip install -r "requirements_versions.txt" --prefer-binary
Error code: 1
stdout: Requirement already satisfied: blendmodes==2022 in e:\ai\stable-diffusion-webui\venv\lib\site-packages (from -r requirements_versions.txt (line 1)) (2022)

stderr: WARNING: Retrying (Retry(total=4, connect=None, read=None, redirect=None, status=None)) after connection broken by 'NewConnectionError('<pip._vendor.urllib3.connection.HTTPSConnection object at 0x000001DF31693FD0>: Failed to establish a new connection: [Errno 11001] getaddrinfo failed')': /simple/transformers/

how do I fix this problem?

problem solved, sorry for the attention 😅

[plucky shoal]

How? Skip something? @oblique kelp

[oblique kelp]

ummm? The context skip in here what is?

[plucky shoal]

I mean how you solved your issue @oblique kelp

[oblique kelp]

Since I often to do analyze, I know how to solved it, So there is a word that say, to create public link set share=true in "launch"

[plucky shoal]

o.k.

[oblique kelp]

so Just search it and.the founded the setting, which is say, skip pip installation = false, I just change the word false with true

[plucky shoal]

Thank you!

[hearty sleet]

Hi everyone, I'm new here, where can I ask for help installing stable diffusion ? isn't this server designed for this?

[plucky shoal]

#🤝｜tech-support

[hearty sleet]

👌

[obtuse peak]

another day alive a new day to fix stable diffusion

[steep star]

Is Stable Animation going to be open source?

[indigo sapphire]

Hlo

Can anyone help me

[broken smelt]

This was asked during the Q&A and they started talking about how they helped out with creating deforum so we can view that as the open source version instead.

[silver pebble]

yo

"i get An exception occurred while trying to process the frame: list index out of range" when trying to do the video2video thing

what needs to be changed

[west halo]

does somebody know if there is a way for me by using stablediffusion to upload a photo of a couch (with background removed) and simulate that exact couch on a living room for example??

I'm kinda of a newbie but I believe Stable is not so good at taking exact references like exact words, letter, objects and recreate them into different scenarios

[thorn pilot]

Have you tried using an inpainting model to fill in the blank area around the couch?

[west halo]

yeah that's a good suggestion thanks, I believe it's not perfect tho as I was looking at maybe generating the couch in a different perspective instead of relying in the same static position of the couch

yeah but probably is the best what can be done with the current version of Stable

[cinder herald]

paying someone who can make a telegram bot with stable diffusion

[paper cloak]

is anyone here familiar with deforum?

[lunar rapids]

how to access the animation model guys ?

[indigo sapphire]

How can we use stable animation

[blazing zealot]

I just I found a version of stable diffusion that works on your phone or iPad but only for Apple devices

[stuck meteor]

can anybody help me out ?

[copper crystal]

https://dontasktoask.com/ maybe this will help

[main salmon]

do anyone know how websites like kaiber made?

how do they use sd deforum in their own website

also is there a diff between sd animation and deofrum on a1111 ui?

[main salmon]

https://platform.stability.ai/docs/features/animation?_gl=1*gri7k2*_ga*MTQzNTg3NDUxMy4xNjgyODg2NDE3*_ga_W4CMY55YQZ*MTY4Mzg5ODU2Mi42LjEuMTY4Mzg5ODcyNi4wLjAuMA..

they provide here some explanation to how use and install it with the pricing also

[copper crystal]

it's not an open release. requires a subscription to their saas unfortunately.

not really a community tool. is more for very rich hobbyists or people looking at business strategies.

[fervent thunder]

trying to help a friend get SD

anyone know how much min system RAM you need ? (not vram)

she's got 8GB

[peak portal]

how much VRAM does she have?

the system ram requirement changes depending on what GPU you've got but just rest easy knowing 8gb likely isn't enough. that computer needs to be about 16gb. preferably 32gb, especially if she has 8GB or less VRAM

the model layers end up offloading into CPU / system memory when they don't fit entirely in GPU space

if you have no GPU and intend on doing CPU rendering, then you need that 32gb of system memory because you'll be loading the model in FP32 mode, 32bit, which uses 2x the memory to hold onto compared to 16bit fp16 models that can run on the GPU

[broken smelt]

dont some of the bois in #🏞｜general-with-images work with 8gb

seen some good stuff in there

ive also heard good stuff about tiled vae from multidiffusion, requiring less vram usage for gens

[stray arch]

Hi

[mossy widget]

Im relatively new to this image generation landscape -- the models on civit.ai are not accessible via an API right? And they must be downloaded, run locally (or in google colab), and prompted in a specific way to get the level of quality outputs that are shown in their galleries on the site? Is that correct?

[versed turret]

Heyyy, When I try to install an extension this is what I get

it just states error

How do I install deforum

[smoky cypress]

the pictures in #🏅｜pantheon are so good, how can I get such pictures? Do I have to download the model on my laptop?

[karmic brook]

They're from our SDXL Bot that is running to select roles within the server! it'll soon be opening up to more and more users here on our server, which will be announced in #📣｜announcements when the time comes .
This was the last update on it -> #📣｜announcements message

[rapid atlas]

I was into AI art generation late last year then got super busy with life and came back and it's like I've been cryofrozen for a thousand years I swear

[short marten]

hi

i wanna know if we can use 2 innits for a single video on deforum stable diffusion

the google colab one

cuz like the results from prompts on deforum arnt the best

[unkempt zephyr]

can anyone explain what the
"filewords
instance token
class token"
mean?
and how are they different from training prompt instance prompt?

[fervent thunder]

What does xformers do?

[fervent thunder]

faster speed and reduced memory consumption

Mhhh, well im having a problem installing it and #🤝｜tech-support is not really responding to my queries, what do i do?

[fervent thunder]

How are you trying to install it?

Every possible way and i get a huge error code in cmd

I just get a huge error message no matter what i do

How exactly you're trying to install it?

I installed all dependencies. FIxed torch, torchvision, everything

What do you mean by that

I want to know what exactly you're trying to do...

have you tried just adding --xformers to arguments ?

Yes

I followed this guide here https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Xformers

But to no avail

hm, idk

Yeah idk, ill bother #🤝｜tech-support more later

Your current error indicates it can't find folder

But...you shouldn't need to build xformers at all, it should automatically be installed if it's in arguments

Ok 1 sec, let me try it again

Hmm, i cant send an image here

Like, every time i launch SD it says torch 1.13.1 is installed. While i have torch 2.0.0 installed

And xformers is in arguments, it just ignores it

[rapid atlas]

If I want two pictures that are similar of an object, one is shiny and new and the other is old and broken, how do I make it look similar? Is in painting my best option? Otherwise there's just too much variation, I'll end up with a shiny TV remote and a broken Xbox controller, for example

[keen lark]

is it normal to be limited to 768x512 resolution with 8 GB of VRAM?

i'm struggling to push it up to 768x768

on AUTOMATIC1111

[charred mesa]

I have 6GB and I use higher

are you using xformers or medvram or not?

xformers can lower VRAM requirements a lot while having no performance penalty (in fact, it might even improve it a little compared to medvram or lowvram)

[tribal bison]

hello, where can i find the stable diffusion 1.5 model?

[rapid atlas]

Would that be on hugging face

[calm steeple]

Huggingface

[dull hare]

I have a question , what is the model name that can make something like digimons ?

need to test some arts but i am not sure which model can do this

[spice plank]

I've got a couple loras that I keep getting images in low res, the author gets decent images and with other loras it seems fine, what settings am I missing?

I get the image I describe but it just looks like it went through a 16 bit filter, I have tried every variation of low-res, low quality etc in the negative prompts

But I've only been using SD for a week now so I thought maybe I'm missing something.

[copper crystal]

that might be a broken VAE but i have not enough experience to say for sure. 16bit would mean it'd look like super nintendo graphics, or kings quest VGA mode

that seems pretty extreme and like somethings very broken with your configuration

[abstract quarry]

maybe it's overfitted. Try using lower lora weight

[molten monolith]

What type of models are Stable Diffusion and ControlNet respectively? They are definitely neural networks but are they GANs or CNNs?

[copper crystal]

they're diffusion models, i think. https://arxiv.org/abs/2112.10752

[spice plank]

I think it was up to high, I was trying 0.6 but 0.4 gives clear images. Weird.

[reef scarab]

Hi i have question not sure if any of you guys know but i am still going to ask. Is their a way to see how a lora model was trained after like upload it somewhere and it will tell u all the settings?

[dreamy wren]

Can anyone please help me here? Mac Book Pro M1 Max 64Gb

Launching launch.py...
################################################################
Python 3.10.11 (main, Apr 7 2023, 07:24:53) [Clang 14.0.0 (clang-1400.0.29.202)]
Commit hash: 5ab7f213bec2f816f9c5644becb32eb72c8ffb89
Installing requirements

Launching Web UI with arguments: --skip-torch-cuda-test --upcast-sampling --no-half-vae --use-cpu interrogate
Warning: caught exception 'Torch not compiled with CUDA enabled', memory monitor disabled
No module 'xformers'. Proceeding without it.
Checkpoint divineelegancemix_V5.safetensors [0a03c339c7] not found; loading fallback f222.ckpt [9e2c6ceff3]
Loading weights [9e2c6ceff3] from /Users/spiffy/stable-diffusion-webui/models/Stable-diffusion/f222.ckpt
Creating model from config: /Users/spiffy/stable-diffusion-webui/configs/v1-inference.yaml
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
Applying cross attention optimization (InvokeAI).
Textual inversion embeddings loaded(0):
Model loaded in 7.1s (load weights from disk: 2.0s, create model: 1.5s, apply weights to model: 1.2s, apply half(): 0.9s, move model to device: 1.4s).
Running on local URL: http://127.0.0.1:7860

When I click "Generate"in Safari browser,
Something went wrong Connection errored out." Python quit unexpectedly.

To create a public link, set share=True in launch().
Startup time: 18.8s (import torch: 2.6s, import gradio: 2.8s, import ldm: 0.6s, other imports: 1.7s, setup codeformer: 0.1s, load scripts: 1.8s, load SD checkpoint: 7.6s, create ui: 1.4s, gradio launch: 0.2s).
loc("varianceEps"("(mpsFileLoc): /AppleInternal/Library/BuildRoots/97f6331a-ba75-11ed-a4bc-863efbbaf80d/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShadersGraph/mpsgraph/MetalPerformanceShadersGraph/Core/Files/MPSGraphUtilities.mm":228:0)): error: input types 'tensor<1x77x1xf16>' and 'tensor<1xf32>' are not broadcast compatible
LLVM ERROR: Failed to infer result type(s).
zsh: abort ./webui.sh
spiffy@spiffy stable-diffusion-webui %

[signal dome]

its literally running

go to local url

[dreamy wren]

Yes I can,

[dreamy wren]

When I type something and click generate, Python crashes
To create a public link, set share=True in launch().
Startup time: 18.8s (import torch: 2.6s, import gradio: 2.8s, import ldm: 0.6s, other imports: 1.7s, setup codeformer: 0.1s, load scripts: 1.8s, load SD checkpoint: 7.6s, create ui: 1.4s, gradio launch: 0.2s).
loc("varianceEps"("(mpsFileLoc): /AppleInternal/Library/BuildRoots/97f6331a-ba75-11ed-a4bc-863efbbaf80d/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShadersGraph/mpsgraph/MetalPerformanceShadersGraph/Core/Files/MPSGraphUtilities.mm":228:0)): error: input types 'tensor<1x77x1xf16>' and 'tensor<1xf32>' are not broadcast compatible
LLVM ERROR: Failed to infer result type(s).
zsh: abort ./webui.sh

[signal dome]

sorry m not smart enough to understand the coding

i just run like a casual user

mine works fine with vlad or automatic

[zealous bear]

How to generate these images? Can someone explain how to use this?

[abstract quarry]

diffusion model. But it's using primarily visual transformers and also CNNs (unet)

[molten monolith]

What about controlnet?

[sand karma]

but it looks bad

[tidal lichen]

Hello!

Please pardon any stupidity, I had a baby and disappeared for a while.

What's the best way to train custom images on the 1.5 model these days?

Back in October it was ShivamShrirao's colab...

Is there anything that runs local on a 12 GB card?

❤️

[tidal lichen]

and...

What is the latest, official, public release of SD?

Is it this?
https://huggingface.co/stabilityai/stable-diffusion-2-1/tree/main

I can NOT seem to get ANYTHING good out of any of these, I'm a little bewildered...

Is there a site that has a bunch of example prompts I can copy/paste to experiment with? So far everything I've found is producing drastically different results on my machine. Wondering if I have some missing dependencies, or ... ?
(Using a new install of AUTO1111)

[broken smelt]

SDXL isnt out yet is it

if you wanna see examples of 2.x pics, theres a channel dedicated to that #1045349359044280360

1.5 is more popular from what i read though, simply because more and more people are finetuning that

civit.ai has tons of models if you wanna see what you can do with it

[tidal lichen]

Yeah, I'm just thinking I must be the world's biggest idiot 😆 I've re-re-re-downloaded-just-in-case the latest "official" releases but yeah ... no matter how much I tweak and tweak, I just can't get anything that doesn't look like garbola.

[warm junco]

With 12gb vram you can install a sd webui local

[tidal lichen]

Yep! I have that. Just getting terrible results with it 😅

[warm junco]

then show your settings an results xD

probably its just a simple thing your missing

[muted skiff]

Which channel should I use to discuss the new local models based on Llama?

[tidal lichen]

Well, here ... totally new/fresh install of AUTO1111, freshly-downloaded models, no modifications, etc ... prompt "a beautiful city" and nothing else:

ah, I can't post images here 😅

[muted skiff]

pos tit in general-with-images and paste the link to the message here

[tidal lichen]

Thanks!

Here:
#🏞｜general-with-images message

[abstract quarry]

hi, you should download one of the popular models from civitai

there is a huge difference between a finetuned model like Dreamshaper, artius, protogen and so on and the 1.5 or 2.1 basic model

[fallen river]

guys, I have updated Automatic 1111 and all the addons, and now the LoRA don't work anymore!!! what's going on???

did they change the way LoRA are prompted? None of my LoRA works now

[warm junco]

update your extensions, then restart the whole webui

[fallen river]

I did it, but LoRA now don't work

I have restarted everything

pls help

[regal snow]

ControlNet just added a new “Reference” preprocessor which helps make new images that are similar to the input image(dog sitting > dog running)

[marsh moth]

Hello everyone and admin. I tried many AI images generations tool before and I still could not find a simple feature that suits my needs. Please help me. I am creating 20 images and more for my comic character. Whenever I type new prompts for different outfit, styles of that character, all the tools I used before will create new character face and style for me. No, I don't want that. I want the tool understand that I need new outfits for only one character

Please help me is there any way to do that?

Thank you for your help

[broken smelt]

Check your extensions, is LORA checked?
I noticed that when I updated my webui to the latest one, all my extensions were disabled and LORA was one of them. Might've just been me idk.

[unkempt hatch]

any models that special in faces? I've tried the facial expression models, but most of these are biased to beautiful women, and I'm looking for diverse faces

[knotty blade]

why dreambooth need webhook discord ?

what is it and what it do ?

[split scaffold]

what would be the best way to make stable diffusion as realistic as possible

[broken smelt]

Probly with a model that's specific to producing realistic work
I'd look at civitai to visually see which models suit your tastes the best
There's also tricks you can use like noise offset, a popular flashlight lora, analog lora, etc

Some of the regulars of the discord post pretty often in #🏞｜general-with-images with some crazy good realistic photos, could always ask in there or #📝｜prompting-help how to get better results

[craggy sonnet]

guys where can I ask for help with an error on Google Colab?

[broken smelt]

Probly #🤝｜tech-support

[craggy sonnet]

thanks

[stone wraith]

Hey, gm!

[abstract quarry]

you have to train SD on your character. If you have 20 images that should be enough. Look for tutorials about lora or Dreambooth

[tidal lichen]

Hey folks! I'm trying to find a model with the hash of 4bdfc29c, how can I do so?
(Already tried Googling; also "hash not available on Civitai")

[real zodiac]

that's just base sd 2.1 afaik

hashing function changed in webui so it might not match

[tidal lichen]

Yeah I'm trying to reproduce results of PNGs I've downloaded here (learning new/unlearning old) and ... yeah, not successful so far. The model name is different but similar? And ... urr.
Thanks 😉

I'm guessing there must be extensions and whatnot that I don't have.
I don't know anything, I just finally started using 2.1 (was on 1.5 forever)

[broken smelt]

You can simply add
Git pull
To your webuser-UI. Bat

Then remove it afterwards

[crisp sundial]

https://civitai.com/models/65127/animal-rizzdom

check out

[tidal lichen]

I thought that was for updating it before launching ... or does it also download updates as needed while generating?

[tidal lichen]

The hippo nipples are next-level, pigs should have 6-8 too 😆

[crisp sundial]

milf

[unkempt leaf]

I have a hard time to figure out Where to get a overview over sd versions? Automatik 1111, SD 2.1, sdxl.. What the Heck is the difference? 😉

[near sluice]

Hi folks, I'm revisiting my SD setup and think there may be better methods. I want to be able to use control net and generate high quality. Ideally I would love to do video.

What are great resources for considerations and and tutorials for setting up Stable Diffusion locally? What is your dream setup?

I can get tech support if needed and potentially even build up a machine. I want to be sure I'm understanding the best options and how to think about it.

Thank you for your suggestions!

[broken gull]

anyone that got any ideas in tricking chatgpt in giving me a valid json file to use in deforum with prompts and everything ? I have managed to get him generating json files that are valid but they don't work in deforum for some reason

(if you answer please reply with a ping)

[lofty scarab]

I have a question for you guys
I'm using vlad
when I launch it tries to update but I get this
error: Your local changes to the following files would be overwritten by merge:
webui.bat
Please commit your changes or stash them before you merge.
Aborting
Updating 93b0de7e..819d105d
What do I need to do to resolve this?

[real zodiac]

backup your bat file if you made any changes to it and git checkout .

[normal coral]

hmmm is there a good resource to get Lycoris to run?

[rugged bough]

holding myself not to buy 3090

hoping models will get more optimized soon

[surreal lance]

2023-05-13 23:10:43.226009: I tensorflow/compiler/xla/stream_executor/rocm/rocm_gpu_executor.cc:843] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2023-05-13 23:10:43.226028: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1990] Ignoring visible gpu device (device: 0, name: Radeon RX 7900 XTX, pci bus id: 0000:0c:00.0) with AMDGPU version : gfx1100. The supported AMDGPU versions are gfx1030, gfx900, gfx906, gfx908, gfx90a.
Num GPUs Available: 0

Do anyone know why i have this error? is the version of tensorflow that is not compatible?

[smoky bloom]

hi all

[rancid relic]

que pelota gerson

bajala con el pecho pele

muy bien definido

is there any AI image generator (outside of Midjourney) that's good at generating decent faces?

[warm junco]

Stable Diffusion 🙂

[spice plank]

Where do I put control net poses?

[warm junco]

extensions/sd-webui-controlnet/models

[spice plank]

Ew, that will make that folder cluttered. Thanks tho

[rancid relic]

you sure?

k I'm gonna trust you

[warm junco]

yea im pretty sure 🙂 you can look into multiple channels here to see some high quality stuff

you can also check civitai.com, there you can find community made models you can use with a local Stable Diffusion install

[pallid flame]

why deforum doesnt work on stable diffusion , its tab is missing , does anyone have similiar problem ?

[blazing pasture]

Where is the best place to post successful prompt generations in this discord?

[plucky shoal]

@blazing pasture prompt as text or image? If anime in anime, or in general with images

[sweet depot]

Hey guys, I just released a darkest dungeon Lora. Would mean a lot to me if some of you tried it out 🙂

https://civitai.com/models/65324

[lime flicker]

can someone help me? i want to uninstall xformers, but i dont know the folder that it is in

[warm junco]

Search with Explorer for xformers

[low prawn]

Who ever came up with reference only preprocessor is a god amongst men 😎

[outer wing]

Question - I'm using img2img and a simple "Resize by" with nothing in the prompts. Clicking "Generate" results in "Waiting..." then nothing. Is this a bug or am I missing something? Not changing any other parameters other than Scale to 2.5.

[plucky shoal]

scale image is better in extrass
have you not some error in console?

[outer wing]

Hmm, not really seeing an error message in console.

Going to try again from scratch.

Will show you the whole output.

[plucky shoal]

i got issues with inpaint as well without errors. Just suggesting extrass for resize as there is very much upscalers

[outer wing]

Ok, going to try "Extras" for rescaling.

Hmm, Got this when only using "Extras" for rescaling.... "Something went wrong Connection errored out."

Not sure why.

[plucky shoal]

what is size of your original image?

[outer wing]

Like 1.2k x 2048.

[plucky shoal]

oh no wonder there is error. By default it resize by factor 4

[outer wing]

Yeah , I just saw that.

It's doing something now, I'll try to reduce it once this attempt is over.

By the way, thanks for reaching back to me on this.

[plucky shoal]

np, glad if i can help. But i doubt i am helpful 🙂

[worn adder]

Hey guys, I'm a product manager looking for work. If anyone has any leads it'd be much appreciated! 🙂

[zealous bear]

Go subscribe! Help family member grow!

Relaxing lofi beats ~ Relaxing forest music ~ Music for studying ~ Lofi music to study 🌲🌊🎶
https://youtu.be/eVMhmhNBGdw

[hallow plaza]

I know that I asked this before, but I haven't found an answer. Does anyone know of a blip2 project that works in the cloud?

[mighty niche]

how to fix "collecting torch" frozen code?

[sharp depot]

How do I add custom upscalers to webui? Where do I find these?

[red torrent]

someone here using amd polaris gpu on linux?

gcn 4.0

cant get webui to run

[spice plank]

I think there is a fix for that on the wiki, pretty sure I've seen that posted elsewhere

I'll see if I can find it

[red torrent]

gfx803 is last supported on rocm 4.5

rocm still somehow works and rocminfo detects the gpu

but torch does not work

stderr: "hipErrorNoBinaryForGpu: Unable to find code object for all current devices!"

HSA_OVERRIDE_GFX_VERSION=10.3.0 = segfault

And pytorch from here does not install on ubuntu 22.04: https://github.com/xuhuisheng/rocm-gfx803

[abstract pivot]

Emad on BBC 1 (UK) this morning! I see you !

[warm bane]

So is LORA training basically just a model? Doesn't this mean that you should be able to do this online? Or does it take too much computing power for it to be worth it?

[magic citrus]

Not all of my posts make me equally proud but I think this is a special one and worth a read. I tried to overtrain text encoder and Unet separately and compare the results. A few very interesting things to observe there. I'd love to hear your interpretations + everything I did is downloadable so if interested, you could do a bit more testing
https://followfoxai.substack.com/p/overtrained-text-encoder-vs-overtrained

[abstract quarry]

always enjoy reading your blog and like the systematic way you evaluate training on open datasets

[magic citrus]

Thank you for kind words, keeps me motivated 💪💪

[trail stump]

hello

i cannot use stable diffusion since i tried to switch to the beta channel. i started receiving endless cuda errors and cant find a method that would resolve the issue

and no one is really helping

how can i fully purge stable diffusion from my system?

[fervent thunder]

can i use created images from Stable Diffusion GRisk GUI 0.1 - on my instagram account without any limitation ??

do i own that image? created from Stable Diffusion GRisk GUI 0.1 A.I

[next jolt]

Hello, I want to learn Stable Diffusion, can you recommend me a tutorial?

[nova dew]

How do I find the options available on the extras parameter for a text-to-image request?

[limber mango]

Does anybody know if there's any extension to automatically choose an appropriate height and width when you use img2img

[peak portal]

@karmic brook this bot is here only to spread political propaganda

[peak portal]

it should actually use the input image details, no?

                    new_image = pipe.img2img(
                        prompt=positive_prompt,
                        num_images_per_prompt=4,
                        image=image,
                        strength=user_config["strength"],
                        num_inference_steps=int(float(steps)),
                        guidance_scale=guidance_scale,
                        generator=generator,
                    ).images

the Img2Img pipeline method doesn't consume height/width options. it seems that if anything is forcing a particular size, it's some custom bit inside sd-webui that forces it to a specific one.

[limber mango]

It stays the default 512x512 always and I have to manually change it (while using automatic1111 webui btw)

[wheat hazel]

anyone has an issue with vladmantic where the generate button is not available even after just starting the ui ?

[warped bough]

(( what that means?

[plucky shoal]

((something)) ? @warped bough

[warped bough]

yes

[plucky shoal]

more attention on that word, group of word inside

[warped bough]

what about - ?

[plucky shoal]

can be written (something:1.5) as well >1 bigger attention <1 lower attention
This is just connection of two worlds robot-duck.
Probably something negative, not sure

[peak portal]

there's actually different methods used for prompt embeds inside sd-webui

one of them works with (this:x.y) format and another uses this++ format--

make sure you're using the one you think you are

[copper crystal]

do you mean emphasis? embeddings are called by typing the filename without .pt

++ is a relic of the past. People would use that before clip models were really understood. It's just trying to affect the clip model with ++ tokens. (token:1.1) is the technical implementation that actually implements a multiplier on those tokens in the math itself. However that math works that is.

(token:1.1) is automatic1111's implementation i think. also uses the (token) method which extends like ((t)) or (((t))) or deemphasising like [t] or [[t]] but eh i dont like the squares for deemph. i use them for prompt editing

[hushed python]

Does stable diffusion keep a database or history off-computer or in the cloud that keeps track of every creation or is it anonymous

[broken smelt]

It's completely off the grid if you run it locally

Disclaimer is that xformers might bleed old results into your generations, or so I've read.

But generally, noone is getting any info from you when you run SD, at least from the popular GUIs out there.

5+ years ago tho, you ran the risk of getting pickled. It's a good thing I chose to not play with the tools available on the internet then because safetensors weren't a thing back then lol

[bleak matrix]

Good afternoon, everyone! How are we all this lovely day?

[hallow plaza]

I have a collection of faces of different characters I'm trying to train on. Would it make sense to say add woman and man to the description. For example should I saw Ryu man and Chun li woman? Will that give better results in general?

[copper crystal]

everyone still runs the same risk that that attack allows. What it would've been is a giant blob of malicious binary data at the back of the cpkl format. later on another attack process would look specifically for it and use that. So the concern was that the file format could be used to get a lot of people to download a payload. Usually a game crack would then sniff out that cpkl with the payload it wants, and executes it. only then being picked up by resident virus scanners, but its too late! the code is in the house!

that same attack can be done with any .png. for years discord didn't scrub any images or videos and any media you got thorugh it could have payload.

[peak portal]

@bleak matrix @plucky mauve is political spam bot

[bleak matrix]

Please file a ticket for #1010934719455707218 or report messages through apps or by using ⚠️

So we can see them

And take a look

[peak portal]

ty, done

[bleak matrix]

There's a lot to learn when it comes to training, like the words/images you should choosd. You can find more info in #🔧｜finetune as well as #1080946152318443610

[raven gate]

anyone have any tips for Lora training but the source images have the same background?

I want just the girl, flexible for every background in my prompt

[bleak matrix]

You mean like the background is the same, but the girl changes in each picture?

[raven gate]

I meant the background is the same in all my source/training photos. The girl is the same.

So when I prompt using my Lora, the girl appears, but the background is always the same

even if I put like "Beach background"

[fleet kayak]

can anyone give me good spes for me to run sd on

[bleak matrix]

Yeah, because if you put the same background for your training, it is going to overtrain on that. That said, you can get around it, potentially, by doing things like adding sand and water as negatives, putting more positive prompts in, etc

[fleet kayak]

im bout to deploy a server with gcp but idk what specs to use

[raven gate]

I see

I was just wondering what I could do to the source images if I wanted to retrain

maybe making it transparent but idk if that works

[compact perch]

How do you generate multiple images, with different prompts? I'm using wildcards, but I just get the same prompt for every image in one generation

[bleak matrix]

The best way to train is to have you person in a wide variety of images and places.

Then you can use in-painting/masks, etc. You can use a solid color, too, and then use that color as a negative.

Unfortunately, it may take a bit to override something like gray, whatever. Once you get a good seed, though, you can use it on any of the images with relatively the same prompt and kind of use it as a process to mesh it together.

There's a lot of very neat stuff that's come out very recently, and with controlnet, or things like Blender, you can have more control over the process of your image as well.

[bleak matrix]

I suggest heading on over to #📝｜prompting-help

[hallow plaza]

Thanks

[bleak matrix]

Np!

[bleak matrix]

You might find more information in #1011228477954998273 Have you ever used SD before?

[frozen light]

Where do I go to start creating images?

[wild steppe]

read through #1072220168534642768

[surreal perch]

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

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

[compact perch]

@bleak matrix Thank you

[bleak matrix]

Np!

[peak portal]

@bleak matrix do you know of any published articles/experimental data on fine-tuning SD 2.1? looking for examples of different learning rates vs dataset size, labelling strategies etc

[broken smelt]

Appreciate the rundown and yeah it aligns with what I know of with today's current environment. I've always told people when they ask about security with SD/popular models that if the risks were real then at least a few thousands of us all have compromised hardware and we just don't know it lol. I still am wary to this day about malicious code on exif data since that's what I've always read about happening the most.

[fleet kayak]

i have but mainly on colab before it was bannable

is 624gb of ram enough

nevermind its only 72

[copper crystal]

yeah payloads are always something to be on the spy for. nothing to really worry about if you stay up to date and don't run sketchy software

[stuck grail]

hello im trying to get a model and i cant find the custom models channel

[warm junco]

Hey here it is:
#1047197565365538826

[stuck grail]

i am trying to find an 18+ model that the site im on says its in the server

[warm junco]

Your on the official SD Server, here isnt any 18+ content allowed

So you might want to visit civitai.com for specific models

[stuck grail]

ur right my bad but i did download a couple of them

[reef drift]

hey, can anyone help me? I wanted to make one of these effects but I don't know where I can get it

[steep star]

Are all these faces the same?
https://media.discordapp.net/attachments/1011634831467221033/1107543942569918544/grid-0001.png?width=924&height=924

[plush rune]

Hey there, I'm trying to try about stable animation and have installed via python on mac using " pip install stability-sdk" but I can't figure out what to do next or where to put API key or how to access? any tips or good tutorials?

[normal plover]

Just goto civitai

[stuck grail]

The models I'm looking for is not in there I checked it's in a different discord server

[normal plover]

The name of the model you're looking for?

[warm junco]

There is an other discord for this Kind of stuff with a similar Name.

[devout hull]

Hello

Can I still use Stable Diffusion for commercial use or stuff changed in the last few months??

[balmy cliff]

if ur talking about unstable diffusion, that place is dead af and is even stricter than here

[young ivy]

how is it stricter?

[devout stone]

I don't know if anyone else has any better ways or already do this but I created an app on my Mac to open Stable Diffusion on it's own so I don't have to keep typing everything in all the time. VERY easy to with automator on Mac. If anyone want's the script I can send it.

[stuck grail]

@normal plover @warm junco I am in the server that has the model u joined yesterday sorry for not being clear on that yesterday

[brazen atlas]

hi guys anyway i can chage the face without chaning the entire pics? to aheago

[hasty bane]

guys, I post some art on tt, and there's a guy who keeps reposting it as if he was the one who made it. There is nothing that can be done since there art does not have copyright, right?

[bleak matrix]

Good morning, everyone! How are we all this lovely day?

[wispy rose]

What's tt?

Tiktok?

[hasty bane]

twitter

[wispy rose]

If tiktok has a reporting function, probably that's a good first step

Same with twitter i guess, use their reporting tools if they have any

It may just be a bot that's reposting

[hasty bane]

he posts it on facebook, I don't think I'm going to do anything, I'm just going to start putting logo in the middle of the images

[narrow hamlet]

How can I test the Stable animation sdk?

[bleak matrix]

You can view the Colab/info in #1098959457964277831

[tough fog]

hi hola, anyone can help about how to install run this ?

[wispy rose]

#1080946152318443610

[bleak matrix]

Check #1072220168534642768

[tough fog]

theres nothing about installation

[wispy rose]

yeah there is, under resources

[wicked badge]

I would like to get rid of some of the roles that ping me about challenges, but it appears that whatever roles I have do not grant me access to the role selection any more. Did permissions change so I need to do something before I can adjust my roles? I get a big fat "No Access" on the link in the start channel that is supposed to lead me there.

[wintry stream]

Hey Dunkeroni! We had to temporarily private the channel & it will be coming back soon! in the meantime, you can right-click on yourself and remove any roles you'd like!

If that doesn't work, ping me and let me know what roles you want removed and I'll take care of it for you 🤝

[wicked badge]

Okay, thanks.

[honest spruce]

Hi, I'm new to SD after playing with Midjourney. I was wondering; If I downloaded SD onto a machine, could I then train it solely on my own image library? I'm a stock photographer and have roughly 15K (at least images) The goal would be to be to sell the images on so only training it on my images would be essential.

[tough fog]

funny, ok thz

[fervent thunder]

GPTE is the best AI resource I've found. You can search over 2,000+ AI tools https://gpte.ai

[glass linden]

Run StableDiffusion on Google Cloud Platform Using Pulumi, https://github.com/calufa/StableDiffusion-Kubernetes-Pulumi-GCP/tree/main/infra 😄

[abstract quarry]

train from scratch is not possible. You would need billions of images for that

you can finetune on your images and SD allows for commercial use

[honest spruce]

Thanks for the reply. This was a hunch. Unfortunately my clients wouldn’t want images unless they’re completely sure they’re not going to have any copyright issues. Looks like firefly might be the way to go.

[floral umbra]

Oof, got no access to the channels the announcements speaks of

And can't wait till SDXL is out for automatic1111!

Protip, don't ask to ask. Just state the actual question lol

[deft tiger]

So uh, i cant access the roles channel to choose my own roles

[rapid atlas]

"Just go to bot 1 - no access" "this links to a channel you don't have access to"

[amber river]

So I'm sure this has been answered somewhere but- is sdxl going to be an open source model / if it is open sourced, will it be a less capable version because of the fear of nsfw? (since- I know public backlash can be somewhat devastating, and the benefits to having a private sota model are substantial)

[wild steppe]

Can you press CTRL + R to restart discord and check again please?