#💬｜general-chat

1 messages · Page 89 of 1

primal nexus
#

both

#

i got tons of xl checkpoints too

#

tried meshgraphemer before, it didnt even realize that there were hands there

#

they might be too small idk

#

i post a picture in the other channel

halcyon granite
#

I'm having a ton of issues getting extension like roop, reactor, etc to work because I can't install the visual studio dependencies.

If I dual boot my Win pc with Linux and use A1111, do you think the new install with resolve the problems?

I'm assuming the same extensions run on Linus and a dual boot would mean installing the necessary dependency

brittle oyster
#

I keep getting this error on my A1111, anyone know what it is or how to fix it?
RuntimeError: The size of tensor a (153) must match the size of tensor b (140) at non-singleton dimension 3
It only seems to be happening currently when I use "Custom Hi-Res Fix" on img2img inpainting

pulsar lintel
pale latch
#

Linux has a lot better support for all these various python libraries i think. i haven't used it for over a year though.

#

just my deck

harsh silo
#

My requests have failed for multiple bots. Any insight?

fading prism
#

Is bot no longer available

edgy wigeon
#

bot not working?

verbal delta
#

I am not sure where to put this question, but is there a ai that can take an existing melody and chords and create a variation? I.e. add more instruments, add parts that follow after the melody, add a counter melody?

bronze pebble
#

Has anyone had this error?

ModuleNotFoundError: No module named 'numpy._utils'

placid hatch
#

Do those who use amd gpus have issues with extension compatibility for webgui?

coral dagger
#

Folks please dont shoot the newbie I love Emad Mostaque and write about him often on linkedin can you point me to the page to generate an Imgae I have it all in hand on gpt4 and dalle but yet to do stable thanks peace

sudden ruin
coral dagger
#

Thank you

coral dagger
warm junco
# placid hatch Do those who use amd gpus have issues with extension compatibility for webgui?

There are some rare extensions that could rely on cuda that won't work. (I don't know any other then TensorRT rn)
Im using an AMD GPU and dont have extension problems but for example you can't train models on AMD windows right now.
On Linux it should be possible. If your going mainly for SD. Nvidia is the goto. But if you just want to create some nice Images in whatever resolution and primarily game on it then AMD is perfect too.
On Linux the AMD compatibility is much better right now. (Much faster)
You can also ask me if I should test something on it (extensions etc)

zealous quartz
#

They finally fixed it, i was already going crazy agony

sudden ruin
#

Should be back I think

coral dagger
#

ok

fervent thunder
#

anyone knows a way to install sd/sdxl locally for AMD gpus?

#

an up to date method

warm junco
fervent thunder
#

alr let me see

placid hatch
warm junco
#

Don't know whats cd tuner

placid hatch
placid hatch
warm junco
#

For higher upscale you need to use SD upscale script in img2img

warm junco
placid hatch
warm junco
#

On windows directml you get 4it/s
On Linux with Rocm you get 18-20it/s
Or if you use Shark on Win or Lin you would get 20it/s
But yea shark needs easier usability with model usage and more and better feature support

fervent thunder
#

Radeon RX 550/550 Series

#

What's a RocM?

versed yoke
warm junco
fervent thunder
#

Do I have to install it in the C drive or any drive on my PC?

warm junco
#

You can also install it on C but store the large models on an other drive

full lark
#

Do anyone know why SDXL takes so much time to load a lora in a1111? also, when generating the first image using the lora it can take more than 4 minutes with a 3060 12gb, but after that it takes like 20 seconds 🤔

still glacier
#

could be because you re storing models onto a mechanical hard drive, or because you re running out of ram during when loading the models

warm junco
pale latch
# fervent thunder What's a RocM?

RocM is AMD's implementation of CUDA. It's an API to the drivers, sort of like DirectX or Vulkan are what games use to talk to the GPU.

pale latch
full lark
full lark
pale latch
#

i dont see why that would help at all

warm junco
full lark
#

Oof

warm junco
#

And do you have 20gb free space on C drive?

full lark
#

I have 16gb of ram, and about 45 free space on C drive

#

Maybe I should remove the --medvram command, I used it when I had the 1060 3gb

pale latch
#

i always start to sweat whenever my free space is less than 100gb

#

only remove medvram if your new gpu is over 8gb of vram

full lark
#

Windows always takes much space somehow 😭

full lark
#

I´ll try

pale latch
#

nice upgrade

#

i always try to upgrade to double vram if possible. that one is a double double leap

full lark
#

I just tried loading a lora with all the tabs closed, no lora = 22.8 seconds, lora = 33.3 seconds, changing the lora weight from 1.0 to 0.6 or other value = 55.7 seconds

#

wtf somehow it works and uses less vram than with --midvram

solemn sky
#

i make images in 5 seconds, 2 seconds with an LCM model.

full lark
solemn sky
#

that totally depends on the model

#

personally i don't do much with SDXL, i find the 1.5 models have been refined so much over the years, some of them match and even surpasses the SDXL models.

pale latch
#

i can't get consistent quality with the lcm lora. honestly have no idea what i'm doing with that thing. when i do get quality images out of it, they're sd15 quality , not sdxl

solemn sky
#

no, it splits the process to reduce the vram usage, which slows the process down.... useful if you don't have enough vram but a slowdown if you have.

pale latch
#

seems all these model speed up releases always have huge trade offs

full lark
#

F

pale latch
solemn sky
#

i find 4 "trained" SDXL 1.0 LCM models on civit.

pale latch
#

lcm models, quality is very inconsistent too. seems good if speed is what you want. i'm not sure it benefits quality or prompt comprehension one bit though

solemn sky
#

speed is king

pale latch
#

and honestly, sd 15 lcm and sdxl 15 are very comparable in quality so just use 15

placid hatch
#

medram breaks the generation process into three stages to reduce overall vram usage

pale latch
#

i've got a 4080 so i got pretty fast generations already

pale latch
solemn sky
#

idk why some ppl setup ComfyUI forkflows making 4-8 images that takes minutes to make and then realize they're not even happy with the composition.
meanwhile i can make 40 images and pick a handful with compositions that i actually like and easily fix with an inpaint.

placid hatch
#

Also people essentially just recreating webgui within comfyui

solemn sky
#

i have perfected a custom 1.5 model which i don't even need negative prompts for.

#

and its name is 51f... 51 cuz it's my 51st attempt, and F cuz it's the 6th revision of that attempt (f being the 6th letter in the alphabet).

#

dang... i must have made over 300 models to get here xD

#

trains and merges, lol

mossy tapir
#

Help me please, because I don't see where I can find preprocess images While using train
sadcat

pale latch
pale latch
solemn sky
#

after cleaning my models directory, i still have over 500GB of models.

pale latch
#

reasons like that why i'm not about to start on optimizations that require me to make copies of existing models

solemn sky
#

i was tired of models giving half-assed images with simple prompts, or models requiring the fancy title of being a "prompt-engineer" to make something decent.
now i have something that's perfect for me.

#

so, in my opinion the journey was worth it 😛

autumn obsidian
#

Noob question! Looking to buy a new GPU; but I can't find an exact answer. What does VRAM do? I get that more might be better, but just to confirm. My understanding is that VRAM only allows me to produce higher resolution images. For my own workflow, could I not just stick with 8 gb then upscale? Thanks for your responses!

latent flower
#

more vram = higher res

#

CS1o is a helper now, I remember when he joined

#

long time since I came here

#

😐

#

hows things

#

around

#

where are the researchers

warm junco
wispy lichen
#

hi

indigo zephyr
#

what auto1111 extension have I been seeing that lets you set a lora for lora:strength:BACKGROUND (I've seen :CHARACTER, too)

#

I've always used Regional for that, but I've been making complicated sets of boxes.

pale latch
#

never seen it before but it makes sense

#

maybe its from a website service. Where did you see it before?

indigo zephyr
#

Not sure. Image of a girl with the demon version of herself lurking behind/above her

placid hatch
#

Maybe some kind of scheduling plugin?

indigo zephyr
#

I use SA and Regional, and I can kinda make SA do something close but it dies at the end

junior harness
#

Anyone using amazon aws server ?

olive crescent
#

quick question can i train a model to generate images with the exact object images i feed the model?

pale latch
#

i know theres an extension that renders a background scene seperate from the characters in it. i can't find it though. might've been multi-subject, but i remember it having specific background prompting

placid hatch
warm junco
placid hatch
warm junco
#

For AMD, Directml has the best support for models and Extensions on windows rn.

placid hatch
still glacier
#

They're probably off by now. But I can answer for them, it's bad.

#

If you want proper proper support for all models, all extensions and correct vram usage while using AMD. The best route is still to go Linux + auto1111/sdnext/etc

#

Any ways that I know about to get full performances on AMD + windows involve compatibility issues and other caveat.

pseudo bough
#

Man I finally got a lora to start training in 1111 it's on step like 134 of 2000 just training a model of my own face to start only getting like 11 seconds per 1 iteration but I'll take it

short vault
#

SD launching with wrong GPU (i have two). Can I do something to "select" which GPU should use?

placid hatch
sweet bison
#

Hi, Anyone has a running sample of Stable Video Diffusion?

full lark
warm junco
warm junco
# placid hatch How do you find vram usage?

As aryetis said, on Directml its very bad. Using to high resolution, to many hires steps while upscaling or having other programms open in the background can cause an out of memory error.

placid hatch
warm junco
# placid hatch Do you consider amd worth it given all the tedium?

Depends: If you want to do mainly AI stuff, for youtube, work, job, etc. or your not to techie with PCs, go for Nvidia.
I switched from Nvidia to AMD even knowing the problems AMD had with SD before, as i dont like to pay nvidia prices or support what they did with their gpu releases (low vram amount for high prices, every half a year they release a "super" series so that you feel like you've been betrayed with a worse gpu, doing a crap for community or open standards).
If your primarily want to Game on your PC, or your a programmer/coder, tinkerer, and also want to use AI stuff with maybe some hickups here and there i would go with AMD. Much better prices, more vram, better Software, best linux support.
If you go for AMD everything below 12gb vram of the 6700XT is not that usable imo.
AMD works on getting Rocm (which is used for SD on Linux) to Windows i nthe next months. That could change a lot.

#

For me who uses AI Stuff as hobby, its totally fine even if its slower or require some tinkering on Windows. I could switch to Linux anytime if i want to. Some people also Dualboot so they can use SD on Linux, and game on Windows.

crystal thunder
#

does anyone know why my adetailer eye quality is so much better when i do img2img instead of txt2img? i have the same settings on both

#

i cant figure it out

chilly vector
#

it's not possible to port 1.5 loras to XL right?

crystal thunder
shut frost
#

Hi, are there any quality models you would recommend? I'm mostly working on sd1 but sdxl is also accepted

fervent thunder
#

Hey everyone. I'm looking for someone who speaks Portuguese and is quite good at using Stable Diffusion. Do you know anyone like this? If so, please let me know! Very much appreciated. Thank you guys

clear osprey
#

what the actual fuck
i just discovered that i can litteraly upscale to 16K at a 9:16 ratio without problem (it takes years but it actually works)

tall gorge
#

yeah vram optimizations have come a long way lol

#

its slow af ofc but it WORKS

clear osprey
#

yeah but it's more like i was dumb as shit
before that, when trying to upscale from 4K to 8K, i got out of vram errors because i was upscaling tiles of 3840x2160
but if i actually take a 4K image but upscales it with tiles of 540x960, it works without any problem

bleak matrix
#

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

vale otter
#

Hello

#

I can't seem to be able to use hi-res fix

#

is it fixable with tiling?

#

it's just x2 from 540x960 causes OOM on 8 G VRAM

warm junco
vale otter
#

RX 7600

#

didn't have any options (which seemingly good for performance)

#

I'll try medvram

warm junco
# vale otter RX 7600

you need --use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --upcast-sampling --no-half-vae

vale otter
#

I'm not on windows

fervent thunder
#

hey everyone, Do you know if there is a way to save the prompt text of each image in the filename inn comfyUi ? Actually I found a module that its called "save image with prompt data" it saves the image but not the prompt... is it necessary to do anything else ?

warm junco
#

then yea try --medvram --upcast-sampling

white palm
#

Hang tight, you're still dreaming! 🙂 - - It's been there for two days, how do I get rid of it?

vale otter
mild grove
#

Guys, I'm not finding a Comfy UI workflow for "content aware" similar to Photoshop, not inpaint, but rather to expand the image to the sides, I downloaded the "multi area conditioning" node but I still don't know how to use it, if anyone has it a Jonson file would help a lot

prisma pond
#

how to fix this error guys? ImportError: cannot import name '_TORCH_GREATER_EQUAL_1_12' from 'torchmetrics.utilities.imports'

warm junco
dusk pawn
#

hello! what are considered the best open-source image+prompt-to-video open source models out there right now?

vale otter
#

How can I do Hires. fix manually using only img2img tool?

warm junco
#

In img2img you can use the SD upscale script or the ultimate upscale extension

vale otter
#

I'm just not sure about settings

#

it seems to completely erase detail I added previously in inpaint

#

is it denoising?

#

shoult I repeat prompt or leave it empty?

warm junco
#

And also negative tags

warm junco
swift reef
#

Hello, I'm trying to make the background blurry, but I can't do it more than a little, my intention is to make it moderately blurry, I already eliminated words similar to "blurred" from the negative prompt, but I still don't get the result I want, what could do? thank you

glossy willow
#

focus

#

un focus

rich kestrel
#

nah g

#

team green ftw

#

my body is ready for 5090

neon hawk
#

Hi! I want to use stable diffusion for 2D game development. Currently I using AUTOMATIC1111 with the base model, but its obviously not very efficient. What models and extensions do you guys recommend?

pale latch
#

refined models like juggernaut. loras for purposed styles. xformers should be enabled if you've got nvidia hardware. latent control models are worth looking into. LCM-Lora too

don't expect stable diffusion to do all the lifting. Take it into photoshop and process it out for your game afterwards.

left hatch
#

Hi , is there any really safe for work tutorials ; that does not show picture of inappropriate picture of females ? , I don't know but I have feeling that people just use stable diffusion for the sake of this kind of thing under other excuses , every time I say to myself I am going to master knew thing today about stable diffusion and this technology I end up stopping after I realize that of man it is just another video that shows how to generate image , animation or God knows what of "females" .

#

for example I got intereseted lately about how to make animation using stable diffusion , like "animateDiff" and most time I search a video I found just video of dancing girls .

#

if anyone knows safe place to learn about stable diffusion use cases without being exposed to inappropriate content, it is much appreciated .

dapper minnow
#

Salut

young ivy
#

Gigabyte GeForce RTX 4070 Ti Super 16GB Windforce OC is 1000 dollars

#

damn

#

I remember when 1000 dollars meant a full medium-end computer

#

No I'd rather save up for that KatVR

#

Or if I can figure out a way to build my own platform and only utilize their shoes

oak yoke
tacit field
#

how do i use a LoRA model?

fading crypt
#

Anyone know how to make the streaming structure better? Right now im getting no bullet points or anything until the very end of the stream. (Using gpt4 turbo api)

sullen stirrup
#

Hi folks,

I'm using A1111 with SD 1.5 with custom checkpoints. I have a question: If I a LoRA or a checkpoint gets updated, will the model get updated automatically or do I have to download and update?

rich kestrel
#

and a Ti at that lamo

stray trout
#

Based on the bot channels, I can tell some training is going on with colorful vs matte/dulled down color images.

warm junco
noble mural
#

Hi, does anybody have a practical method of managing lots of similar image results? For example, I have a result folder with over 9000 images but do not want to browse through and cherrypick manually based on sampler used etc.

dusk halo
#

hey one of my firend is unable to join this server , and when i am sending him invite link it sayting that he is block can anyone help me in that ?

sudden ruin
noble mural
sinful sparrow
#

Otherwise if it's personal well....I mean, that can't be helped as a vast majority of humans enjoy making females worryeyes the latest feature will often be used with some pretty lady because that's what a lot of ppl like

ivory sinew
#

The Little Explorer's Magical Journey《小探險家的奇幻旅程》【AI LOOKBOOK】【Children's Picture Book】
This is the first children's picture book I've tried to make. It's actually quite difficult. The difficulty lies in fixing the characters and costumes.
https://youtu.be/uu0v-sY73YU

sudden ruin
topaz parcel
#

is there a way to save both the image, and all of the prompt and settings, in AUTOMATIC1111. ?

sudden ruin
topaz parcel
pale latch
#

alwayshasbeen.gif

fallen axle
#

AMD is poop for SD for the record.

#

Just as an fyi, came from there myself not long ago

pale latch
#

was using a vega64 with rcom for a few months when i began. linux is a better environment for amd. those windows drivers are balls

fallen axle
pale latch
#

went back to windows though when i got nvidia card. wanted to take 11 for a spin.

#

tabbed explorer! sold

fallen axle
pale latch
fallen axle
#

Am I allowed to post SD animations here or do they strictly belong in showcase?

fallen axle
pale latch
#

i post stuff like that in genwithimages often enough. theres a showcase here?

fallen axle
#

I'm super new

#

Umm there is an animations channel

#

Lord knows if anyone goes in there but I'm addicted to posting them.

#

And there ain't no pic posting in this channel which makes me sad

wispy lichen
#

hi

indigo zephyr
#

2024 meme potential aside, I mostly want to make pics of my kid riding a T-Rex

#

Maybe randomly replace people in the family group photo wall with celebs

lofty laurel
#

Has anyone ever dealt with trademark stuff? Our startup is: https://pilot.io/ (we've been using this domain/brand for over 9 years, well before Microsoft co-pilot etc..) - we're talking to a trademark lawyer, and it sounds like they might advise us to change our name/brand, but I can't help but think about "Mike Rowe Soft" and how we were using it first - anyone been in a scenario like this before?

terse plume
#

hi i put a model to run and said "model failed to run", somebody know how to make work?

vale otter
fallen axle
#

and I couldn't use comfyUI

#

Linux the performance is supposedly better, but still eh

lavish lake
indigo zephyr
lavish lake
indigo zephyr
#

Yeah they're just gonna take it

#

With all respect, and experience in this

fallen axle
#

I switched to a 4090 so it's not fair to make a direct comparison at all

lavish lake
#

taking it means going to court, I bet they prefer to do this privately between parties, but i'm no lawyer

indigo zephyr
#

I am and I'm sitting with 2

#

It's just my personal opinion though

lavish lake
#

i know personally of a case (friend) that took a celebrity name URL back in the early days, and they resolved it privately, sold. but those were other days

indigo zephyr
#

I genuinely hope you do/can

#

But don't write any checks against that money until you have it

pallid sable
#

hello, why is my permission removed to generate in bot channel?

#

admins?

frosty pasture
#

has anyone ran into the issue that if you change the adetailer model in the adetailer extension, suddenly it stops working? i start getting red squares over the faces of the people it should be fixing, or black squares etc, just not working

#

and i dont see any error in the cmd.exe

karmic brook
pallid sable
mortal lynx
manic wadi
#

hello, why is my permission removed to generate in bot channel?

sacred mist
#

Hi all, I have a question for everyone: In Stable Diffision in discord you have also banned the generation of pictures "You do not have enough rights to send messages on this channel"?

#

or how do I find the admin?

shut plaza
#

since the stabledreamer is down without an eta on coming back is there any other similar AI you recomend for generating the videos? this one was until now the only one I managed to understand lol

solemn sky
shut plaza
#

In the videos about stable difussion local hosting I saw that they didn't had the option to generate video

#

I'm kinda new to this so sorry if my question is stupid

solemn sky
shut plaza
#

Thanks I'll do that then

#

I thought this was the stable diffusion discord O.o

#

Do I need to download the stable dreamer model or something?

solemn sky
#

Unstalbe is not the same, it's another project NOT related to Stable., was meant for nsfw but now it's pretty much a universal community.

#

no, ppl download models from civitai.

#

it requires some work to get started, first time might seem tedious, it's rly not though.

shut plaza
#

I'll give it a try and see how it goes, thanks for the help

fading prism
#

Are there any other servers that are as good as StableDiffusion for image generation that you can recommend?

brisk imp
#

Hi

unique atlas
onyx ivy
#

为什么在机器人1-10频道中，显示：你没有权限在此频道中发送消息。以前都可以，现在为什么不行了?有没有人跟我一样

fading prism
warm junco
warm junco
left patrol
#

应该是机器人维护吧，公告显示要一段时间呢

viral pelican
#

还不行吗？

fervent thunder
#

👀

#

Hi .... why my stable diffusion not workin?

urban grove
#

why can't I send promt to the bot?

#

Hi

warm junco
real goblet
#

hi

pearl cairn
#

hi

empty shadow
#

E

#

(╯°□°)╯︵ ┻━┻

proper solstice
#

bot not working?

raven tendon
#

idk

#

the same for me i can't write any prompt

#

i'm ban moddos ????

still glacier
#

We can t make it more obvious that the bot is down for now.

white palm
#

there's no deadline at all?

lavish lake
#

No estimated times for having the bots back sorry, this has been posted many times and available at #1047610792226340935

wide juniper
#

Any good anime models that are based on SDXL? I’m still using Anything v5 for 1.5

lavish lake
terse plume
#

there is a very good model to make anime girls?

#

hot womans

#

like nami

terse plume
wide juniper
#

Or are ther no SDXL models, only loras?

terse plume
#

what i mean is that you cant use like tatsumaki Lora on SDXL

fervent thunder
#

Does anybody from here knows where i can get stable diffusion?

#

I wanna use it.

terse plume
fervent thunder
#

Yes, 8gigs

terse plume
fervent thunder
#

DDR4

#

Thanks @terse plume

terse plume
fervent thunder
#

Also thanks for that as I am a python developer!

#

I was also about to ask that!

#

@fervent thunder

shut plaza
#

does someone has an invite for the unstable difussion discord server?

hybrid cedar
#

Prompt: prompt:prompt:Vulcan, star Trek, a female Vulcan, tanned skin, elf-like vulcan ears, arched eyebrows, skinny, slender, femenine body, sensual pink sci-fi dress, in a vulcan white alien mountain, Vulcan robe, fisheye camera, fisheye effect, retro, selfie, summer, transparent dress, sexy dress, 90's, long hair, very long hair, beautiful, exotic, sunkissed
Negative: long dress, covered dress, modest dress, ugly

still glacier
#

that s still not a bot channel :p

#

and bot is still down

bronze pulsar
#

ola

spice rapids
#

Sup

#

I was downloading a repo from nocrypt and this popped up (Misleading:Linux/FRP!MTB). What is it?

fervent thunder
fervent thunder
spice rapids
fervent thunder
#

Oj

#

Ok

#

Can you tell me the repos url

#

Can you please explain me what the repo is about for better understanding?

#

I am helping as a Windows 11 Home user.

spice rapids
#

Oh, I think I've got this thing figured out a little bit. I think it's just another stupid Windows nag

fervent thunder
#

Ohh

#

It happens

spice rapids
#

Thanks for your responsiveness bro

fervent thunder
#

Theres a lot of bugs in windows to watch out

fervent thunder
fervent thunder
#

If you need any further assistance then please convey me!

north pelican
#

hi, What happened, are they getting spare parts? Will this still work?

shy imp
#

hello

#

hope to learn more abt SD here

pine fiber
#

is sd cooking or did they fall off

lavish lake
pine fiber
#

not what I mean

#

they been hella quiet since dall e 3 came out

lavish lake
#

Hmmm… I think they released Stable Video, Sd Turbo Xl and 512, Zero123, a LLM and a Code LM if memory serves well, not quiet at all : )

pine fiber
#

something that isnt a glorified clip diffuser model

#

like actual research 😱

solar venture
#

because I get: you do not have permission to send messages on this channel

sudden ruin
solar venture
sudden ruin
#

Its down, just read the latest message in that channel

fading fox
#

Good afternoon. I haven’t used this server for a long time, but today I discovered that I don’t have permission to write to bot-1 and other channels. I couldn't find a reason for this in the rules. Perhaps it is no longer available for free accounts? If anyone knows the reasons and how to fix it please tell me. Thank you.

solar venture
sudden ruin
languid plover
#

why i don't have permission to sen msg in bots

#

?

vapid hazel
#

Me too, I do not have permission, and I already read above messages.

languid plover
#

any recent announcement ?

sudden ruin
humble frost
#

Hi everyone.

lavish lake
#

Please dont be about the bot is down question 😜

humble frost
#

Where i can ask about SD model, that can understand and transform schematical drawings into interior maps\schemas?

humble frost
lavish lake
#

Models don’t “understand”, they’ve been trained on images and and you can infer new images based on that training

#

There must be LoRAs for technical drawings for sure

humble frost
#

Yeah, i'm understand that. I used that as "trained".

lavish lake
#

Or you can even train yourself one

humble frost
#

That would be a long road, yeah.

#

I just hoped that there is something.

lavish lake
humble frost
#

Not as far as i can see, through stability matrix. Will try to go through hugging face later.

lavish lake
humble frost
#

I want to create encounter maps from simple schemas, as i cannot draw at all. -_-

#

Oh, darn, there is!
Yay.

lavish lake
#

There u go!

humble frost
#

Thanks. For some reasons i though that Stability Matrix uses same search for site, as, well, via Chrome itself.

hearty path
humble frost
#

Second question - is there any other opensource picture generating AI aside of SD?

finite cloak
# pine fiber like actual research 😱

here's image research by one of our researchers released a few days ago https://arxiv.org/abs/2401.11605

arXiv.org
Scalable High-Resolution Pixel-Space Image Synthesis with Hourglass...

We present the Hourglass Diffusion Transformer (HDiT), an image generative model that exhibits linear scaling with pixel count, supporting training at high-resolution (e.g. $1024 \times 1024$) directly in pixel-space. Building on the Transformer architecture, which is known to scale to billions of parameters, it bridges the gap between the effic...

pine fiber
#

ok I stand corrected I apologise for my rude words

terse plume
#

its not the same thing Stable diffusion or others site generate imagens?

#

there is a expert

solemn sky
#

i bet you just need to get your priorities straight, like, look at how many subscriptions you're paying for... netflix, world of warcraft, patreon?
there's a lot of ppl paying for more than they need, there's a lot of free alternatives for many things.

you don't need an Nvidia GPU, just 4GB of vram, the first 4GB GPU came in 2008 which should give you plenty of options on a wide variety of prices.
Unstable Diffusion has regenerating free currency for the AI on their website but i don't use it much so can't say anything about its quality. it's nothing fancy tho, you probably make better images here.

rich kestrel
#

why ppl dont just invest in a basic ass gpu is beyond me

#

they rather struggle with these colabs that keep banning AI tools over and over and over and ppl keep asking "wHy caNt I gEn oNlinE"

#

keep bending over for Google g. Ima be here chilling multitasking while ur begging Google for credits

humble frost
#

Well, tbh, if you want to work with decent models and don't start rendering and then going to a work, because it would take a whole day - you need decent GPU.
Like i had 6GB 1060. Let's say i was not satisfied with speed and results.

vale otter
humble frost
#

Yeah..

spice plank
#

Is there a way to fix the "killed segmentation" error messages?

#

It errors out after generating for like 30 minutes and I've tried a completely fresh install and it still happens.

pine bobcat
#

dumb question, where to put character and training data downloaded from citiviai

rich kestrel
#

nvidia vs AMD is exactly like coke vs pepsi

#

one is clearly superior but people choose to save a few cents by going with the other brand... then you get what you pay for lolz

#

cuda is the true future of AI, and with 50xx series being specialized for AI, my body is absolutely ready.

humble frost
#

Well, yeah, in AI sphere - Nvidia is clearly are superiour.

jaunty yarrow
#

How can I train a character lora? Like the technique comicsmaker.ai uses for example

terse plume
haughty wasp
#

Guys, does anyone know any web app or library or API to convert images to vector for free that gives good results?

jaunty yarrow
finite cloak
finite cloak
#

I don't even know which of the two you're trying to imply is the clearly superior option

jaunty yarrow
#

would like a guide honestly

#

one thing I liked with the site I mentioned is that it got pretty good results replicating the inputs

distant lynx
#

hello!
is there any benefit training SD 1.5 LoRA with 1024px images?

finite cloak
jovial cedar
#

where do you find loras and embeddings? i can't stand amount of nsfw and asians on civitai

vagrant raptor
#

Okay, I did a dumb thing! Enjoy my choices here! I was going through art styles and I got up to Pixel Art. Using this, I found an upscaler I wanted to try out. So yes, I upscaled a picture that is pixelated with the Pixel Art prompt.

proper carbon
#

I would like to know what can I use to convert an image into a Stable Diffusion prompt

real goblet
#

sadcat good

fervent thunder
#

6

warm junco
cobalt fulcrum
#

Is there a discord server dedicated to training loras?

#

Or maybe training models in general

ocean crest
#

Hello everyone, I'm looking for someone who speaks Dutch please

white palm
#

The five stages of accepting the inevitable - denial, anger, bargaining, depression and acceptance

ocean crest
primal nexus
#

can i make a video when i have 1 image as reference, and a sequence of open poses (like 50 or 500 open pose pngs)?

#

i know how to make video2video and img2video and text2video, but i dont really know how to combine one image with a series of open pose images to a video

#

any hints?

lavish lake
arctic coral
#

I put a question in #🤝｜tech-support that I couldn't find a helpful solution for yet. I hope I'm right asking for help there 🙏

lavish lake
arctic coral
opaque sky
#

how can i write a prompt in a way that the main object is a bit further away so there's space for other elements on the image. I want to generate a human face surrounded by patterned circles but the face is always very close up and ignores the circles

primal nexus
lavish lake
primal nexus
lavish lake
#

in my experiments, it works, sometimes it fails too, depending on your image and openposes you can get good results

lavish lake
primal nexus
lavish lake
#

have fun!

primal nexus
# lavish lake have fun!

its interesting, this is uding a "pose guider" and no controlnet with openpose, also there is an "animateanyone" sampler, and no ksampler

#

uhmm my first try failed spectacularly lol its just a garbled mess

lavish lake
primal nexus
#

at least its good in producing completely disturbing bodyhorror movies

#

i think the input image and the video/openpose sequence have to match exactly

oblique star
#

Where to check for docs on pipe() method?
In this code I want to see what are the parameters that I can provide to the pipe(), such as height, width and others?

pipe = AutoPipelineForText2Image.from_pretrained(
    "stabilityai/sdxl-turbo",
    torch_dtype=torch.float16,
    use_safetensors=True,
    variant="fp16"
)

pipe = pipe.to("cuda")

prompt = "a photo of Pikachu fine dining with a view to the Eiffel Tower"
seed = random.randint(0, sys.maxsize)

num_inference_steps = 4

# This pipe(??args)
images = pipe(
    prompt = prompt,
    guidance_scale = 0.0,
    num_inference_steps = num_inference_steps,
    generator = torch.Generator("cuda").manual_seed(seed),
).images
obsidian basin
#

I've been working on a little web app to help cut down on copy-pasting when writing prompts by taking a "card-based" approach to prompt building:

https://app.deckai.io/

I would love it if people could give it a try and provide some feedback 🙂

bleak matrix
#

Good morning, everyone! How are we all today?

polar hamlet
#

Hey, hoping I can get some help with a fresh automatic1111 install on ubuntu 22.04 (virtual machine, gpu passthrough). Automatic1111 installed fine, loads fine. Now I'm trying to setup xformers and local network access. I've added the line: export COMMANDLINE_ARGS="--listen --xformers --reinstall-xformers" to the file webui-user.sh, but it doesn't seem to have any impact. I'm not new to this, I had to bang my head againt the wall a long time to get it working before and I don't want to do that again. GPU is 3060, is recognized by other things in the VM. Any help would be appreciated.

#

Oh, jesus.. I run the webui.sh inside the stable-diffusion-webui directory instead of the one the install instructions say to run that's outside that directory and it's now installing. That's an aweful oversight and I'm sure I'm not the only one running into that mismatch. Maybe someone will read this and fix that. Almost surely not.

fluid flame
#

Boy's animation avatar (╯°□°)╯︵ ┻━┻

#

Boy's animation avatar (╯°□°)╯︵ ┻━┻

latent scarab
#

Hello

#

Why bots are down!

frail dew
#

hello

arctic coral
#

If anyone has a spare second to help in #🤝｜tech-support 🙏 trying to reproduce an image but can't

naive merlin
#

Hey I am new to stable diffusion and I am looking for some help. I am having issue reproducing the style and quality of an image even though I have the exact prompt that was used to generate it (at least according to the creator of the image). I am using Draw Things, and the particular image I am tryng to reproduce is https://civitai.com/images/1742145.

surreal lance
#

is rocm fully working now on windows?

lavish lake
naive merlin
#

I don't know hwo to set clip skip on Draw Things.

#

Nor embeddings

#

It only lets me set seed, steps, cfg scale

naive merlin
lavish lake
naive merlin
#

Sorry what is an embedding?

lavish lake
#

clip skip 3 is a must

#

and the embeddings seem to be easynegative, ng_deepnegative

#

also, I don't think they're using Draw Things, this seems either ComfyUI or Auto1111

naive merlin
#

Alright, maybe that is the issue.

naive merlin
lavish lake
#

no probs, there's a lot of moving parts and you have to match them close in order to get similar results

naive merlin
lavish lake
#

it's on the negative prompt: extra digits, bad eye, EasyNegativeV2, ng_deepnegative_v1_75t, NSFW, nudity

naive merlin
#

Oh i see.

#

Are those embeddings just automatically included in the checkpoint model? I am guessing that is why he had to remove them with negative prompting right?

lavish lake
#

they are textual inversions, another training technique

celest oyster
#

I have a quick question I hope someone can answer. I have very little SD experience because of my crap gpu, I'm looking at a 4070ti with 16GB vram. Question is: is there anything that a 4090 with 24GB vram can do that the 4070ti could not in terms of resolution? I assume if I were to ever max out 16GB vram (if that's possible?) it would start using my regular ram.

lavish lake
celest oyster
#

I assume I might see issues when training neural networks aside from SD

lavish lake
#

resolution is not only limited by vram, also by the models, a 1.5 model can't go too far on it's own, etc... you need to upscale, tile, etc

#

training is where vram gets really taxed, training sdxl checkpoints or LoRAs, depending on how many images you use, will eat a lot of vram, sometimes even maxxing out a 4090

#

if you don't plan on training heavy sets.. a 4070ti 16 GB is fine

#

the new 4070ti super too, it's basically a 4080

celest oyster
#

oh sorry yea I meant 4070ti super

#

okay that's the confidence boost I was after, thanks for the input I appreciate that

lavish lake
#

no probs! that's a really solid card

celest oyster
#

going from a rx570 haha yea

#

I'm living in the stone age here

lavish lake
#

oh you will have fun!

warm junco
eager shore
#

quick simple question I reinstalled comfyui and my yaml file for extra models path did not have .example and the type file will not change over to show YAML. How do I fix this?

lavish lake
#

`#Rename this to extra_model_paths.yaml and ComfyUI will load it

#config for a1111 ui
#all you have to do is change the base_path to where yours is installed
a111:
base_path: path/to/stable-diffusion-webui/

checkpoints: models/Stable-diffusion
configs: models/Stable-diffusion
vae: models/VAE
loras: |
     models/Lora
     models/LyCORIS
upscale_models: |
              models/ESRGAN
              models/RealESRGAN
              models/SwinIR
embeddings: embeddings
hypernetworks: models/hypernetworks
controlnet: models/ControlNet

#config for comfyui
#your base path should be either an existing comfy install or a central folder where you store all of your models, loras, etc.

#comfyui:

base_path: path/to/comfyui/

checkpoints: models/checkpoints/

clip: models/clip/

clip_vision: models/clip_vision/

configs: models/configs/

controlnet: models/controlnet/

embeddings: models/embeddings/

loras: models/loras/

upscale_models: models/upscale_models/

vae: models/vae/`

rich kestrel
#

Otherwise buy a 3090. Preferably pre-owned

#

Vram is everything in the world of AI

lavish lake
# rich kestrel Dont buy 40xx unless its a 4090

not necessarily true, remember that we're moving to fp8 and even fp4. Not everyone has budget for 24GB vram, he's fine with 16 GB for most things beside training heavy sets or heavy animation (lots of frames)

rich kestrel
#

A 4070 is more expensive than a 3090, for less vram, and small performance increase

#

Ideally, buying a cheaper 3090 and saving up for a 5090 is the best plan right now

#

Otherwise , a jump from 4070 to 5090 is less cost effective than 3090 to 5090

celest oyster
#

power consumption is a bit of a concern here, 3090 is closer to breaker tripping 😆

lavish lake
#

with fp8 it will halve... 16Gb will be the new 32 GB

rich kestrel
#

There are always efficiency updates, but there is always also more powerful technology coming out

#

Just wait until SD 3D comes out

#

Then you'll be begging Bill Gates for an extra gig of vram

celest oyster
#

I want to wait for the 5000 series, but I've already been waiting since the 3000 series

lavish lake
#

there will always be incoming a new generation

celest oyster
#

yea it never ends, gotta pull a trigger at some point

#

at least I can get a card now post covid/scalper shit

#

post crypto

lavish lake
#

the 4070ti super you mentioned is a cheaper 4080 with almost the same performance, it's a good bet

celest oyster
#

it's frowned upon for what it's offering at it's price point, but it really is perfectly aimed at me for how old my card is and how badly I need it right now

lavish lake
#

of course there will be 50xx cards, and then 60xx, you just need to balance what you need right now and give yourself extra room to be future proof

celest oyster
#

nervously looks at 80A breaker panel

lavish lake
#

if you can get your hands on a cheap 3090 it's a great card but as said before, not everyone needs 24GB ram

#

I was perfectly fine with a 3080ti 12GB until I got into training and running longer animations, that is where you'll make use of the extra ram

#

and never underestimate the power of good developers nailing down efficiency, when SAI realeased Stable Video, we needed 40GB cards, in about 3-4 days they got it down to 12GB

#

and it's happening on every new model

celest oyster
#

I've come to terms with the fact that I might outgrow 16gb vram, but I think I'm willing to take a little financial hit to upgrade some day.

rich kestrel
#

Also take in account if ur a multitasker

#

I can game while training a model no sweat bc of the extra vram

celest oyster
#

haha that's funny

rich kestrel
#

If power consumption is an issue you can always under volt

celest oyster
#

that's right

#

my rx570 pulls 150w max right now, couple hundred more watts is ok but I can't run the coffee maker if a 3000/4000 series is full load 😧

lavish lake
#

the 4070ti super is also great at consumption, I think less than 300Watts if I remember correctly the launch data

#

less than a 4080 with same perf.

celest oyster
#

I have extension cords running from other rooms into the living room just to sustain this nonsense in some arterial fashion

#

can't upgrade panel or run new circuits cause no certified electrician would touch the wiring without a full $$ upgrade

lavish lake
celest oyster
#

rock and a hard place

#

yea that's a good number for me

lavish lake
#

3090 is around 450W

celest oyster
#

It was funny reading all the 4070tiS reviews cause almost everyone shit all over it, but it's actually a smart move for a lot of ppl.. I think jayztwocents touched on that

lavish lake
#

well it's funny, just the name 🤣 4070 ti super

#

and the vendors will add more shit: extreme OC

celest oyster
#

ti super master extreme oc aorus elite limited edition

lavish lake
#

kinda ridiculous the name scheme... call it a 4075

#

or baby 4080 🤣

celest oyster
#

not 4070TISMEOCAELE ?

#

rolls off the tongue

#

intel cpus suffered a similar naming fate

dark surge
#

Hello

soft moss
#

is there a smarter way to check/see all the a1111 themes? enabling one, and restarting evbeyrthing seems quite the chore

cedar salmon
#

im torn just like everyone else but i was thinking upgrading my 2080 TI to 4080 Super, but ide really like to know timeframes for 50 series....

celest oyster
#

I think all nvidia has said is maybe this year

#

from what I read it's a card they're keeping in their deck until it's a smart time to play it

#

in other words, milking consumers for the maximum amount of money until then

#

but also watching what amd puts on the table

#

it's like a trillion dollar game of crazy eights

quaint saffron
#

Hello everybody! Can someone help me install Automatic1111 and Comfy UI locally on my machine? I would greatly appreciate it!

hallow sable
#

i would not upgrade from a 2080 lmao

quaint saffron
#

Those 2 are supposed to be the best right?

hallow sable
#

how do you not cry when you see the price for a 4080...

#

like 10 years ago you could buy a really good computer with taht

rich kestrel
hallow sable
#

still getting paid the same

quaint saffron
#

Guys, anyone can help me?

rich kestrel
young ivy
#

wonder how the hell I can get OpenVoice to work on Windows

#

seems like it's either pay for their service or use Linux

quaint saffron
rich kestrel
#

or supertortoise

lavish lake
rich kestrel
#

or literally the storm of audio apps that have been released lately

quaint saffron
hallow sable
#

the instruction on git are very simple

#

the only thing most people dont know is to edit the batch file with notepad

young ivy
quaint saffron
young ivy
hallow sable
#

yus

north pelican
#

until when will this not work? or will it no longer work?

young ivy
#

I want something that can replicate voices at high quality, near realism. That can to T2S or V2V

rich kestrel
#

yeah I open a repo and see .ipynb, thats an instant nope and move on

young ivy
#

maybe I'll just pay to use a service instead of fucking around

rich kestrel
#

nah dont feed the machine

young ivy
#

at least then I'll have an interfac etc

rich kestrel
#

applio. mangio-rvc fork. audio-webui

#

all of those are local and work pretty good, although I havent tried applio

young ivy
#

if you have samples that show how good they are at cloning voices

rich kestrel
#

I do

young ivy
#

would take a lot to convince me

rich kestrel
#

want to hear my widowmaker clone voice

young ivy
#

I dont know what widowmaker is

rich kestrel
#

thts a problem then

young ivy
#

just take a celebrity and clone them

rich kestrel
#

if u got the samples

young ivy
#

I don't

#

thats why I'm askingg if any of them provide samples

north pelican
#

hey, until when will this not work? or will it no longer work?

young ivy
#

I'm also in too much pain atm to navigate into each and every github

#

and investigate it

north pelican
#

helpme

rich kestrel
#

ima share you my true-tested audio apps gimme a mo

young ivy
#

so many of the voices that are recorded by randoms and are free aren't made in a professional setting so they have background sounds, heavy room noise, reverbration etc

#

making them horrible to clone

rich kestrel
#

why would you want to clone with someone else's samples

#

the whole point is to train ur own models with ur own samples

#

or u can download premade models

young ivy
#

Because I need placeholder voices, so I need to try and mix two or more voices together and get that into the clone

#

and create a new voice

#

this is why Im saying that the paid versions probably are much better because they allow these things, they also let you use their super computers to generate and train

rich kestrel
#

or just use coqui or elevenlabs for that

#

no point in downloading a trainer repo if u just want generic voices

young ivy
#

I don't want generic voices, I want to clone voices

#

then there's descript

#

there's quite a few and until there's a relatively intuitive installation of an opensource AI Voice app with T2S and S2S, with an interface and all...kinda like how easy it is to get Automatic or comfy etc to work

rich kestrel
#

super tortoise

#

i was looking for the link but am multitasking like a lil bih

#

just gimme amo

#

works out of the box

#

he calls it ai-voice-cloning, I call it super tortoise cuz its a fork of tortoise on steroids

#

the only ish is that it doesn't produce .index files (afaik)

#

but might be just what ur looking for

#

never support cloud services. power to the people

stray kelp
#

Does anyone know how to connect newreality to fooocus

#

Or how to use a LORA in general

lavish lake
#

like auto1111 or comfyUI etc

stray kelp
#

Fooocus

stray kelp
random rampart
#

Best stable diffusion for AMD gpu?

lavish lake
warm junco
# random rampart Best stable diffusion for AMD gpu?

Best performance on Linux:
Automatic1111 on Linux (best usage and fast)
Comfyui on Linux (also good)

On Windows:
Auto1111 Directml (best support, but slow)
Comfyui with directml (also good)
Shark webui (the fastest for AMD cards, but lacks features/usability)

young ivy
random rampart
warm junco
shadow raptor
#

Hey guys, I'm trying to learn fooocus AI, could someone tell me if it is even possible to take an image of a dress and inpaint it into another image of a woman?

lavish lake
random rampart
#

Thanks

eager shore
#

Does getting the lastest Cuda toolkit make sense?

#

12.3

warm junco
eager shore
#

I was thinking that too. thanks

#

I getting my comfy and stable back from overdue hiatus

#

what is your main/fav driver Comfy or sd?

warm junco
#

easy to use, good extension support

potent pilot
#

I have a video clip that I would LOVE to remake using stable diffusion if it can. Everything would stay the same in the video, but switching out the person including the clothes they are wearing. they are doing the same action as the original video, keeping the object in the video that the person is holding, the object that is a pin being clipped, also keeping the sound affect.

prisma hazel
#

mj community is so toxic imagine being ganged on because you won every live prompt battle competition laugh

#

this is why SD is always 100% better

random rampart
#

@warm junco followed all of your instructions. now how do i get it to work with a lora i downloaded?

warm junco
#

Make sure to use a model that is based on the same SD version.
1.5 models will work with 1.5 loras,

random rampart
#

okay

warm junco
#

You can also adjust the lora strength by changing the number behind the lora name in the prompt

pliant ermine
#

with A1111 how do i know what version of SD its using?

#

or what are most people using now days?

lunar wave
#

how to make anime ai images? any good models that could get u the results similar to novel ai?

bleak matrix
stray kelp
#

Im using invoke ai. It says its a standalone, so does that mean I can remove stable diffusion?

#

And then how do I get a SDXL LoRA to work properly with it?

pliant ermine
#

i see your point

snow stone
#

Why cant i use the bots

mental lake
#

Hi, i am trying to install stable diffusion locally, but i keep getting error code "ERROR: Could not find a version that satisfies the requirement torch==2.0.1 (from versions: none)
ERROR: No matching distribution found for torch==2.0.1" anyone got any idea how to fix this?

vestal dew
#

This channel used to be jam packed, lively. Now, mostly dead.

pliant ermine
#

ive been away for some time

#

i do recall it being much busier

vestal dew
#

all the sd channels are like this now - just a few comments each day

hallow sable
#

how many people know about models being used as a vector for malware?

#

Im unsure of sites like civit ai

dim surge
#

Are there any local tools that aid with background removal like remove.bg?

cedar salmon
vestal dew
hallow sable
#

what is the main difference between a model and safe tensor?

versed ore
lunar nebula
#

Hello, I have installed Stable Diffusion yesterday with some checkpoints and loras and played along with it. I have seen that you come generelly fast into NSFW (not that intended). Now I want to show my daughter the tool to create some cool images. But I wonder what I should do to minimize the posibility to get some NSFW content. As far as I understand I can use the negative prompt. But what should I use to get sure that there is no problematic stuff? Also I use some generic checkpoints like Dreamshape and Realism, whihc would you use for generic purposes (thought about some images of teddybears or spaceships etc ...)

lavish lake
# lunar nebula Hello, I have installed Stable Diffusion yesterday with some checkpoints and lor...

This is an excellent question, you can do many things. First having your daughter only use a user/instance that has zero nsfw specific checkpoints and loras. The second would be preparing that instance so it has all that in the negative prompt (nsfw, nude, naked and so on). You can also load embeddings/textual inversions specific for that purpose, like> https://civitai.com/models/99890/civitai-safe-helper

random rampart
#

Something's wrong with my lora. I try putting it in the prompt and it doesn't work.

lunar nebula
lavish lake
warm junco
lunar nebula
warm junco
lunar nebula
#

I use stable Diffusion Wub UI

bleak matrix
#

Good morning, everyone! How are we all today?

crimson knoll
#

你

echo violet
real goblet
pliant ermine
#

There anything special to be done to use sdxl with a1111?

tawdry schooner
#

when will Stable come back?

meager flicker
#

Excuse me, the bot is unavailable, so we can't use stable diffusion on discord anymore , right ?

lavish lake
warm junco
pliant ermine
#

how are most using it?

#

there a recent guide youd recommend?

main haven
#

Hey guys, I joined this server recently because I was curious about the current state of generative AI and Stable Diffusion.

The last time I tried out SD in a serious capacity was... I think 9-12 months ago.

Has anything changed? For example, does it still take a lot of VRAM to run models? I'm on an old NVidia card with only 6g of RAM so I could only go medium VRAM without having to close a bunch of applications... hoping some improvements have occurred in terms of memory usage.

I also saw SDXL on CivitAI. What's the situation with that, and can pre-existing LoRAs/etc work with it?

warm junco
#

if you have a low vram gpu you need some additonals cmd args in the webui-user.bat
ask in #🤝｜tech-support for any help on that

pliant ermine
#

Oh I thought you meant like, use comfy ui

warm junco
#

sdxl works in auto1111 and comfyui

sinful forum
#

jnjjjfcc

woven owl
#

how can I create videos with automatic1111?

warm junco
quick scarab
#

Why are all the bots down?

lavish lake
fervent thunder
#

hey guys its been forever since ive done SD is automatic 1111 still the one to use?

lavish lake
fervent thunder
lavish lake
fervent thunder
#

Big thanks

fervent thunder
#

hi

#

Didnt touch ai image creaton for 6 months

#

and i see from stability matrix that there are a lot of interfaces

#

can someone give me a quick rundown what is the best way to use it now

warm junco
#

It will be extremely slow

#

Auto1111 would be the best for your GPU

#

There are the performance settings covered too

fervent thunder
#

oh thank u sm

#

apparently i have to like redownload auto1111 cause it like failed to download torch or smthn

warm junco
fervent thunder
#

i have 3.10.6

#

i think thats good

broken terrace
#

Does anyone have a link to that very handy database with i/s per gpu please ?

velvet fossil
#

does anybody know anything about Magnific AI?

#

I know they use Stable Diffusion

#

but what has changed?

upper rose
#

what s theb est message to create a constant AI model with a similar face and body in different situations?

#

best method*

lavish lake
velvet fossil
dry tapir
#

Hello everyone!

I am not Sure where to write this question, but I will just ask it here:

Regardless of which UI this would work. Is there a way, if you have a picture of a character as a 3D Modell, and I sorta import it into SD and just want to turn the picture into a digital drawing.
Is this possible? Just changing the style of the picture without major changes in clothes and face?

I've tried various approaches with controlnet and so on, but I think I am missing something...

I am currently using the automatic1111 UI, so it would be neat if it is possible in this. But if there is a easier approach in for example the Comfy UI then I would like to hear that too

Thank you in advance

Best regards

warm junco
#

Best would be to use the IP-Adapter model for that

#

Maybe with canny and openpose too

dry tapir
#

So I would have three controlnet with each of them using these models? Should I have a specific setting for each model or just the general ones?

#

But okey I will try it out thank you!

warm junco
#

If you want the exact same form then use canny.
Same for the openpose for the pose

#

IP-Adapter is for getting the style

#

IP-Adapter face is for the face

#

I would start with IP-Adapter + openpose

dry tapir
#

Okey! Would you like me to keep you updated if you are interested?

warm junco
dry tapir
#

Okey! Thank you very much! ☺️✨

warm junco
#

Np, if I have time tomorrow I will try 3d to digital art. You can also send me an example if I should try something specific

dry tapir
#

That would be terrific! Sure I would gladly send you my example picture tomorrow! If this would work I guess many people could take advantage from that 🤣

shadow raptor
#

Sup boys, random question, is there AI to create sounds?

warm junco
dry tapir
#

You can try out Stable Audio. Its limited but still pretty good

shadow raptor
#

Ok, another random question. I have fooocus, but struggling to understand on how to restore/improve quality of my old pictures, that possible?

shadow raptor
dry tapir
#

No only online afaik

lavish lake
ripe vine
#

Excuse me , can i know why i can't use bot to do anything.....

lavish lake
earnest shadow
#

that dub….. so….

dim surge
#

I would like to start creating some animation, and have already installed plugins with controlnet for krita. Was wondering part from animate diff, if there are other tools to create pc-based Ai videos without having to use paid for cloud services

glossy temple
#

Is there anyway that for example a flat tshirt input is getting gernerated on a person?

#

I also trained a lora on a product but sadly the written stuff isn't good at all. So the product was kind of the same but the writing wasnt good. Is there a way to make it better?

supple kindle
#

Guys, How do i cut off unnecessary light on character? It keeps generating some shine, sunlights on character. I'm tryna create webtoon sort of thing. I add special (light, atmosphere, shadow etc) effects later defends on whats going on that cut. Even i write negative prompt it seems not working.

pastel willow
#

hello everyone

#

I have a question, which is better to use? Fooocus or AUTOMATIC1111/stable-diffusion-webui?

#

I am very new to this

fervent thunder
pastel willow
#

which one would you recommend ? I am fine with it being complex and hard to learn

warm junco
dim surge
#

What are the requirements for video creation

warm junco
dim surge
#

What about using control net pose and generating tons of frames

warm junco
#

Img2img also features batch process so you can easily run through a folder of images

#

Controlnet supports that too

tidal palm
#

Why can't I chat in bot?

warm junco
lunar nebula
#

Hello, I have a question. Can I read out the parameters I used for the picture later? I used the Stable Diffusion Web GUI and as I understand it saves the information in the PNG-Metadata...

warm junco
lunar nebula
#

Ah, thank you. Haven't seen that tab.

dim surge
crystal thunder
#

does anyone know why my loras appearance changes so much after inserting a break?

#

i would like to keep the appearance but also have the benefit of the break system for color bleed

fervent thunder
#

Stable diffusion and also llms in general are so gpu hungry, literally need 3k build just to get them running. 5k+ for a solid performance and trying advanced features

lunar nebula
#

Hello, another question, is there some kind of commenting sign (to deactivate a prompt without deleting it)?

warm junco
signal mica
#

Morning. I'm trying to improve my understanding of how everything works. Is this a fair definition for sampler and steps?

Stable Diffusion starts a generation by creating a fully random noise image (using the seed - which itself can be randomised). The model then figures the amount of noise in the image before removing it. This is repeated many times (steps) until eventually an image free from noise is created. The process used to calculate the noise and produce the next iteration is described by the sampling method.

lavish lake
dire trellis
brittle heart
fervent thunder
#

H

#

Their bot is so complex

#

A single bot does alk the stuff

#

All*

thorny trellis
#

Hi everyone! Could someone please help me understand the pricing for fine-tuning an SDXL model for face features via stability ai API? I'm also interested in the cost of accessing the fine-tuned model after training. Thanks in advance!

signal mica
lavish lake
lavish lake
woeful quail
#

getting rid of mental aphantasia with good nutrition

crude hemlock
#

maybe silly question... can I use sd1.5 trained Loras on an img to txt with base prompt of SDXL?

lavish lake
crude hemlock
hallow sable
#

people have multiple versions of SD? how does that work? do you just make another directory?

jade mesa
#

Anyone have any experience with animateanyone?

outer wing
#

I need some assistance with a generation. I feel like it's ready difficult to achieve this. I'm trying to create a image of a man sleeping on the balcony of a home in the snow outside. With his window open, and someone else sleeping inside his home. But nothing I put comes close to this. Any suggestions?

prisma pulsar
nimble shard
#

what kind of AI products, assuming anything is allowed to be used, that would help a political candidate in winning an election the most? i mean very specific use case, for example deepfakes for what specific use case, in technical sense

you might be able to cover this so that people are more aware for safety purpose?

celest oyster
cedar salmon
#

ya figures, they have that new production to offload..

rotund orchid
#

why i cant create image ?

shut frost
#

Most of the models are trained with female characters. Do you know any good models that are doing great on male characters? Preferably anime style

cosmic gorge
#

I dont know the range numbers of what is low or high in advance sttings on all settins, what numbers are low or high prompt strength, generation steps or seeds what numbers sould i put in order to what results? Alsi, what is the difference in models SDXL V1.0 ou stable diffusion v1. 6? What all of this means?

shut frost
# cosmic gorge I dont know the range numbers of what is low or high in advance sttings on all s...

Disclaimer: I'm also a beginner so there may be some false information

SDXL is for higher resolution and details, It takes longer to generate images with SDXL
SD 1.X is the regular version most people use. It is for daily usage and there are so much more models and LorA's for this version

There's also sd 2.0 but don't bother, literally everyone had a problem with this version, and close to no people use it.

#

You can't custom your seed according tou your wants, Seed is like a navigation point of how the denoising process will result. So start with "random seed" option, generate a couple of the SAME PROMPT with random seeds, and if there's a specific image you like, you can use that seed and alter the prompt to get more similar results to that one.

cosmic gorge
#

Thanks

#

What about the other 2 things?

slate dust
#

There is no SD 1.6. You might be thinking of WebUI which has had a version 1.6. There is SD 1.5 and SDXL (plus turbo editions). Just use 1.5 models if you dont have a powerful machine.

shut blaze
#

Guys, is there any anime fine tuned SDXL turbo model?

cosmic gorge
#

Hi i could nor buy credits, apears We are unable to authenticate your payment method. Please choose a different payment method and try again.

cosmic gorge
hollow gazelle
#

Hey , is anyone using VMs on google to train and generate?

rocky holly
#

dream A woven labels of Galaxy

worthy tree
#

are people in SD working on any humanoid bot project?

#

just curious

#

something on the lines of tesla's optimus?

dim surge
#

hi, is there an equivalent of canva/invideo on local pc?
as all these services are paid and stack up to quite a hefty expense

scarlet moss
# cosmic gorge What is pronpt strenght?

You can prioritise certain words in your prompt to make them more prominent. You do this by adding brackets for example with: cute fluffy cat
You can make 'fluffy' more prominent by typing: cute (fluffy) cat

Each set of brackets multiplies the emphasis by 1.1 so ((fluffy)) would become 1.21 strength. You can also just type in the number like in the following: cute (fluffy:1.5) cat

It might depend on the model, but whenever I tried to go over 2 the picture started devolving into a non-sensical mess of noise, so I tend to stick with 1.1 to 1.4.

median rune
#

hai

#

cat

polar urchin
#

Admin please?

signal mica
dark dragon
#

Hey there, I really don't get through this.
in the free stability AI membership it states that its not for commercial usage. Does this mean that SD is no longer free for commercial use?

lavish lake
manic grove
#

bot always down ?

lavish lake
ripe gazelle
#

Hello

dark dragon
kind jewel
#

Hello

lavish lake
bleak matrix
#

💐Good morning!🦄 How is everyone this beautiful day?

harsh silo
#

Where does n00b go to learn about stable diffusion?

signal mica
lavish lake
#

youtube too: olivio sarikas, sebastian kamph and a lot more!

signal mica
#

Question about adding quick settings and whether they need to be adjusted per check point.

For example: I was copying another image prompt which used SD VAE and Clip Skip. This article about Clip Skip suggests its only really relevant on 1.x models. So my question is should I disable it Clip Skip if I am experimenting with say picX or another checkpoint and does the same rule apply to SD VAE?

lavish lake
#

the vae will affect the look and colorspace, so having the correct one always helps

#

about clip skip, if the models have been trained like that, you need to use it correctly as stated on the model/lora

#

most models/lora have that info (at civit.ai, etc...), they also state if the VAE is already baked, so you don't have to choose one

pastel lynx
#

Hey guys! Anyone knows how to better avoid the tiling issue when using the Ultimate SD Upscale node ?
Sometimes it works by decreasing the amount of noise in the sampler but when i do several upscales to increase the definitions i usually get that tiling issue with colors and bugs between each tiles..

dim forge
#

Hi guys. How I can use this bot?

lavish lake
potent dust
#

Can anyone point me to some workflows/tutorials for getting a person's body type into stable diffusion? My faces look pretty good with ReActor but for the bodies, I'm just winging it trying to adjust prompt weights on different body parts until I get something close enough to trick my brain.

polar urchin
#

@lavish lake Hi, I need to talk to you privately if it's possible! Thank you!

lavish lake
neon radish
#

Yo

fervent thunder
#

yo!

zealous hamlet
#

I'm jonesing to generate some videos ,like when will the bot be up????

potent dust
#

When you train a face from a set of photos, do you use reActor at the end? Do you merge your trained face with an existing checkpoint to create a combined custom checkpoint? Can you recommend a tutorial on training on a face?

lusty plank
#

anyone here who captions pics for lora training, do you caption the emotions as well? Like if the person is smiling, angry, etc, should you caption the emotion? I go back and forth on what's better.

supple hatch
#

Is there an Auto1111 specific discord server?

hidden hollow
#

Hello everyone!
I have rich experiences with sfw, nsfw image generation platform development.
My experienced image generation platforms are DALL-E, Stable Diffusion, Midjourney, Randomseed.
Also custom chatbot development using OpenAI is my major.
If someone needed to develop something like this, I am ready to work for it.
Best Regards.
Thank-you.

warm junco
supple hatch
warm junco
supple hatch
#

Whats an appropriate channel for the controlnet face-id model?

warm junco
supple hatch
#

Thanks

warm junco
#

np 🙂 see you there

worldly sail
#

I got into SD about six months ago and had a lot of success with lora on my 3070. I have been out of the loop since and will have a 4090 soon. What direction should I be going?

  • SDXL?
  • a1111 or comfy?
  • LoRa or something else?
  • Kohya for LoRa?
    My plan was to just go back to LoRa with Kohya and SDXL. Play with Comfy. Is A1111 a waste of time with SDXL?
clear oyster
#

Is sd 2.2 going to be coming soon?

warm junco
forest trout
finite cloak
#

first version of what became XL was called sdv2.2 until it was understood as a whole new beast

worldly sail
#

Thanks, both of you. Yeah I plan on playing with both. It sounded like SDXL was not supported on A1111 at some time but ideally I could stick with what familiar while I get back into LoRa training.

clear oyster
#

Also does xl need better GPUs then standard?

finite cloak
#

nvidia 20xx, 30xx, 40xx desktop versions all work fine

#

laptop versions are more mixed

clear oyster
#

I'll try and see if xl works too

finite cloak
#

note that you need a good or well configured ui, like comfy or Swarm

clear oyster
#

I see, checking on that

novel spear
#

.

clear oyster
#

BTW, how does stability AI profit from all of this if everything is free and open source?

novel spear
#

is there special channel for developers to ask dev questions in?

finite cloak
clear oyster
#

Wow sdxl turbo is only one step wow

novel spear
lavish lake
lament plover
#

hmmm, how do I get permission to use the bots ?

lavish lake
pliant ermine
#

im my old install of a1111, it wrote a log.csv for images...i cant seem to find the option to enable that now. anyone know where it is?

clear oyster
#

i dont know what I did wrong, I followed the steps in the stable diffusion art guide but my sdxl turbo images are like they need more steps
cant upload pictures here

#

they are all grainy like when you set a normal model to a single step

#

can anyone help

#

i literally turned it on and off and its fixed lol

#

what is the word to make the images look like real photos?

slate dust
#

"A photo of" and add some photographic prompts like "35mm f2.8" "Kodak Gold" "Canon 5D"

#

I also like to neg out "painting/drawing/cartoon/airbrushed"

austere valley
#

hi … i'm new here and search for info's for the app draw things … any hints are welcome

warm junco
austere valley
#

that's what i know currently … want to know the general usage of this app, the meanings of the parameters, what happens when I use this or that … means the general usage of the app …

summer lodge
#

q

hallow sable
#

werty

wooden spear
#

A

austere valley
vague vigil
#

Hello world! 🙂 Do you think it would be possible with AI to generate a drawing that looks like a panel of Little Nemo in Sumblerland, drawn by Winsor McCay? 🤔 (Or even a page.)

warm junco
austere valley
warm junco
austere valley
trail socket
#

How soon will the bot be back? before Saturday?

near pulsar
#

hello. I start getting the message "you dont have permission ..... " in the the server rooms of bot-6 and up. WHat is that? why?

honest lynx
#

#diffuse-together diffusers, thank you so incredibly much for helping me get a start here! Just won a monthly “special stuff and ai” on International Underground Music Video Festival and will be in the annual!

floral umbra
#

time to transfer 400GB of SD 1.5 models to my server. 2TB nvme is full, so i can wait 40 sec for spinning rust for 1.5, then 7 sec for SDXL from the local nvme lol

oak patrol
#

Does anyone have experience generating assets with SD? Looking for tips and resources

hallow sable
#

why do you have so many models?

vapid tangle
clear oyster
vapid tangle
clear oyster
#

what gpu

#

@vapid tangle

vapid tangle
#

but i also got just 8gb ram

tiny carbon
#

Why don't I have the authority to send messages in stable diffusion

trim magnet
# vapid tangle but i also got just 8gb ram

you cant run it with just 8gb of ram,its slow af on my pc because i only have 16gb of ram,you can do like me of increasing pagefile but because its using the pagefile as ram it will be even slower

sonic anvil
#

Hi guys, is there any leaderboard for text to image models? I prefer the kinds like chatarena which is ranked by human comparisons.

wicked gyro
#

does anyone know of a way to insert a comment into a stable diffusion prompt that will not be rendered or effect token count etc

old river
#

why why why

placid void
#

cmooon its nearly been 2 days

tired shadow
#

？

neon hawk
#

Why is this ai a 100 times slower if I change the width and the height?

stone wedge
#

?

sudden ruin
lavish lake
#

the best way to have more resolution is upscaling

bleak matrix
#

Good morning, everyone! How are we all today?

tiny kestrel
#

hello

signal mica
#

Good afternoon. I'm looking at the JuggernautXL model and wondering what is meant by "HiRes: 4xNMKD-Siax_200k with 15 Steps and 0.3 Denoise". I'm guessing its something to do with high resolution but thats as far as I get. Can someone explain it to a simple soul?

lavish lake
#

the hi.res fix is using an upscaler, those are models trained on big images, that 4x_Siax mans that it's best for 4X scaling

#

if you give it 1024 you can get 4096px out

gentle saffron
#

Hey everybody, enjoy your creativity day

signal mica
clear oyster
clear oyster
#

ofc fine is subjective lol

shut umbra
#

Hello everyone, I just started with stable diffusion and I am currently facing some small problems. I am using fooocus and I am so far satisfied with how it works. Now I want to change the size of individual body parts and I can't get it right with inpaint. Does anyone know how to enlarge or shrink the breast size, for example?

thorny trellis
#

I’m looking to fine-tune SDXL with** Lora** method for my new project. I’m wondering if I should go with TPU or GPU for this task. Can anyone tell me which one would give me better performance and be more cost-efficient?

hard orbit
#

Hey totally off topic of stable diffusion but I've paid 3 different companies for ai headshots and they are bad. One had my head like way to big, one of them turned me asian and the 3rd just looked cartoonish. Does anyone know a good viable solution for this.

shell hazel
hard orbit
shell hazel
hard orbit
shell hazel
#

Sounds like you want AI to do what photoshop does

hard orbit
# shell hazel Sounds like you want AI to do what photoshop does

pretty much. I just liked the idea of spending 35 bucks and gett 40 shots... It would take me hours to edit that much in photoshop... plus while im a graphic designer Im not well versed on faces and that type of thing. I create logos and marketing graphics for compnaies.

shell hazel
lavish lake
shell hazel
smoky igloo
#

hey guys, I'm looking for a better open-source image generator,model currently I'm using PixArt-alpha, and sometimes it generates low-quality images

woven forum
#

Yes I was not impressed with it.

lucid marsh
#

hi guys

#

can someone help me figure out how to utilize more VRAM in fooocus

#

I find the documentation lacking and I can't get my 7900XTX over 1GB util

#

?

cedar salmon
rich kestrel
#

If you're looking for a photorealism option anyways

sick venture
#

Hello, is it normal I have generating speed about 2 it/sec on SDXL 1024x1024 with RTX 3060Ti? Or is there some problem?

warm junco
sick venture