💬｜general-chat

[languid helm]

is there anything i can do better in this idea ?

any suggestions would be much appreciated :')

[limpid knoll]

hey everyone
is the discord bot completely free to use?

[mystic cape]

i just got comfyui, anyone got good tips videos or links they reccomend?

[crimson willow]

yes

[limpid knoll]

nice ;D

my computer is not strong enough to run automatic1111 reasonably

[pastel lynx]

Did someone could make work Deforum after install DW pose ControlNet ?

[brittle dune]

Hi, frens

[fervent thunder]

Hi I made a command line program to request SD to generate picture based on prompt and seed

if you are interested https://github.com/sorainnosia/sd-req

in windows : sd-req.exe norepeat "rock in a river" 1 seed_start 1 seed_end 100 will generate 100 images using seed 1 to seed 100 for prompt "rock in a river"

need to enable API in SD by setting SET COMMANDLINE_ARGS=--api in webui-user.bat

[fervent thunder]

is AI capable to understading vectors to show up the direction of some motion that you want to generate?

[marble knot]

Do someone know what Prequel use for their IA generation? Im sure they not using the same neuronal learning as SD or Midjourney uses and i would like to have some clues if someone could help
If you don't know Prequel its the App people use in social medias to look like an anime or a cartoon character even in video (with consistence and no glitch)
The closest thing i have in mind is "Style Transfer" or something along those lines but still have no clue where to look at.
Im willing to pay to use something like that, i just hate Prequel because its on Android only and wish i could have the same thing for Windows or Linux.

[bleak drift]

So how come even if I put stuff like Photorealistic and 4k in my prompts, and include negative prompts like blurry, jpeg artifacts, etc.

I just get really low quality images?

Is it because the scale I selected isn’t big enough for the imported image (I’m inpainting)

[bleak matrix]

If you need help with prompting, I suggest going to #📝｜prompting-help

And post your full prompt 🙂 along with your settings for your image

[clever palm]

hi guys, do you know what checkpoint and lora should I use to get this 2d illustration effect?https://trznstudios.com/cdn/shop/products/21_e36d8827-a968-489f-b3b9-4a6467f6f3d7_2048x2048.jpg
https://trznstudios.com/cdn/shop/products/6_50ba4bbb-2566-4bb5-81ce-525850d87ada_2048x2048.jpg

[pale badge]

Is it possible to get prompts back after I submit one and get a server freeze message? Or is it just gone at that point?

[warm bane]

Hey guys, which room do I use for help? I can only see help with prompting

[tall gorge]

#🤝｜tech-support

[warm bane]

oh idk why it didnt show up, thanks 😄

[glass citrus]

can someone please tell me what model is being used to generate images here on discord? is it sdxl 1.0 or 0.9 or something else?

[gray wolf]

what are yalls go to negative prompts?

[glass citrus]

Here's mine: artifacts, smudges, object stretching, objects mixing, distorted, blurred, low detail, low res, low quality, extra fingers, extra heads, 6 fingers, 2 thumbs, extra hands, extra heads, body stretching, body distortion, low facial details, facial blurriness, artifacts on face, artifacts on body, ugly, poor image quality, bad looking, out of frame, text, number, blurriness, half car in frame, low detailed vehicles, artifacts in vehicle, no crisp details in vehicles, AI artifacts, not symmetrical, improper design, half vehicle visible, bad hand, lowres, monochrome, grayscale, jpeg artifacts, signature, watermark, username, blurry, artist name, text, error, ugly, tiling, disfigured, deformed, low quality, pixelated, blurry, grains, grainy, text, watermark, signature, out of frame, disproportioned, bad proportions, gross proportions, bad anatomy, body out of frame, duplicate, cropped, cut off, draft, extra hands, extra arms, extra legs, extra fingers, extra limbs, long neck, mutation, mutilated , mutated hands, poorly drawn face, poorly drawn feet, poorly drawn hands, missing hands, missing arms, missing legs, missing fingers, fused fingers, unnatural pose, out of frame, low resolution, morbid, lank background, boring background, render, unreal engine

you might find a thing or two repeating. also for sdxl 1.0, only a few words really works well: ugly, disgusting, artifacts, watermark

[covert needle]

tried it and its working so far thanks king

[gray wolf]

Thanks dude

[cobalt crag]

What is easy diffusion?

is it faster than Comfyui?

[spare fable]

A simplified front end for Stable Diffusion.
And no, not generally.

[cobalt crag]

faster than everything?

[spare fable]

No. ComfyUI tends to be the fastest.

[cobalt crag]

Alright

well rn for me

I am intending to switch from ComfyUI to Automatic1111

Automatic111 is very hard to install

especially xformers

[cobalt crag]

Do you know how to make Automatic1111 detect checkpoints from ComfyUI? models folders?

[spare fable]

I do it the other way around. Move all your checkpoints to where A1111 stores them and edit the extra_model_paths.yaml file in the ComfyUI directory to point to them.

[cobalt crag]

Oh that's cool!

[cobalt crag]

It doesn't work :|

I typed "D:/sd.webui/webui/models/Stable-diffusion"

and it doesn't detect even tho I already move my checkpoint to there

[spare fable]

If you're in Windows, you'd use backslash, not forward slash. For instance, mine is:
base_path: C:\users\soulc\stable-diffusion-webui\

[cobalt crag]

nope still doesn't work

D:\sd.webui\webui\ btw

sorry

[spare fable]

Is the "models" folder in the D:\sd.webui\webui\ path?

[cobalt crag]

oh you mean this?

[lyric estuary]

Noice. These models can take up to hundred gigs easily. Let alone having 2 web UIs. Thanks for the tip 🙏

[cobalt crag]

It's still doesn't work

#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: D:\sd.webui\webui\models\Stable-diffusion\

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

#other_ui:

base_path: path/to/ui

checkpoints: models/checkpoints

gligen: models/gligen

custom_nodes: path/custom_nodes

or do I need to locate base_path to webui-user.bat?

[spare fable]

You have D:\sd.webui\webui\models\Stable-diffusion\ as base_path, but it should be D:\sd.webui\webui\

configs should be models/Stable-diffusion
vae should be models/Stable-diffusion

[cobalt crag]

I already set it like you said

It still doesn't work

[lyric estuary]

So looks like basically this can be applied to lora and control net models as well then 🤔

[spare fable]

Yes

[spare fable]

I can assure you that if the paths are correct, it works. Did you restart ComfyUI after the changes?

[cobalt crag]

Yes I do

[atomic forge]

hello, can someone help me

[lyric estuary]

Probably easiest is just browse to the folder using explorer and copy the path from the address bar to prevent typos

[spare fable]

Did you rename the example file so it's ending in yaml?

[cobalt crag]

nope

[lyric estuary]

Sup

[atomic forge]

i have installed the mov2mov extension

[cobalt crag]

ah it added extra search

now it's still won't appear

[atomic forge]

but it doens appear to me

[cobalt crag]

then I need to change path

[atomic forge]

i already retarted ui

[lyric estuary]

Need hard restart probably, close the command line window as well and when it's starting up check for any errors

Don't have experience with mov2mov tho AI video generation still sucks

[atomic forge]

i´ve tried everything

i ot the git link and copied to install

got*

i closed google colab

openned

and nothing

[lyric estuary]

Ah dunno about google collab

I ran SD locally

[atomic forge]

i wish i could hahah but my computer doesnt support

[cobalt crag]

My cmd it said

Adding extra search path checkpoints D:/models/Stable-diffusion
Adding extra search path configs D:/models/Stable-diffusion

[lyric estuary]

If you close the browser the session will be terminated and I don't think if you revisit the site it'll remember the settings

[cobalt crag]

That are wrong directory :(

[spare fable]

Paste your current yaml.

[cobalt crag]

#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: D:\sd.webui\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

#other_ui:

base_path: path/to/ui

checkpoints: models/checkpoints

gligen: models/gligen

custom_nodes: path/custom_nodes

[atomic forge]

it remembers, bc it´s working with my drive

[lyric estuary]

Got Nvidia card?

[spare fable]

Take out the / at the beginning of checkpoints, configs, & vae.

[atomic forge]

maybe there is another type of extension

[cobalt crag]

It worked!

Tysm

[atomic forge]

same kind

[atomic forge]

my gpu is 4

[lyric estuary]

Nah I don't think it's saved. Was using collab to train lora

[lyric estuary]

4 what?

[atomic forge]

so, what do i need to do?

gb

[cobalt crag]

Adding extra search path checkpoints D:\sd.webui\webui\models/Stable-diffusion
Adding extra search path configs D:\sd.webui\webui\models/Stable-diffusion
Adding extra search path vae D:\sd.webui\webui\models/VAE
Adding extra search path loras D:\sd.webui\webui\models/Lora
Adding extra search path loras D:\sd.webui\webui\models/LyCORIS
Adding extra search path upscale_models D:\sd.webui\webui\models/ESRGAN
Adding extra search path upscale_models D:\sd.webui\webui\models/RealESRGAN
Adding extra search path upscale_models D:\sd.webui\webui\models/SwinIR
Adding extra search path embeddings D:\sd.webui\webui\embeddings
Adding extra search path hypernetworks D:\sd.webui\webui\models/hypernetworks
Adding extra search path controlnet D:\sd.webui\webui\models/ControlNet

All correct?

tysm

[lyric estuary]

Cry

[lyric estuary]

Still some slash there btw Windows only recognize backslash

[spare fable]

That'll work fine at that point once he had the correct base path and took out the leading slashes in front of the other lines.

[cobalt crag]

Anyway

I must find a way to install Xformer 20

[lyric estuary]

Ah good to know. Thank you ser

[cobalt crag]

for Automatic1111

I already have 17 but I have no guidance and I'm stuck

myeh nvm

Imma uninstall A111

It's very hard to install and make it run fast and smooth like ComfyUI

Idk why

[fervent thunder]

how many images can I request per hour or day? I got timeout ed

[lavish lake]

on the bots here? there is no limit, timeouts usually happen on nsfw stuff, keep it clean!

[fervent thunder]

I was requesting roads with trees, mountains, that kind of stuff. Suddenly I received a timeout

[lavish lake]

can you share the prompt to see what might have triggered=

[fervent thunder]

unless it accidentally generated some nfsw.

[lavish lake]

there's a blur filter that usually grabs those

[fervent thunder]

wet road with curves made of cobblestone, with a valley landscape, petals scattered on the road, night time, cherry blossoms at one side of the road. Behind the cherry blossoms there should be an old village with lights. atmospheric haze. Chinese light poles illuminate the road. full moon with clouds.

Negate: glare and aslphalt.

[lavish lake]

seems OK, weird

seems that one of the prompts had nsfw

it was always the same?

[fervent thunder]

I attempted to uncrop some square images in dropclip. I noticed one or two rare occasions where it filled the space with random anime or halloween stuff.

images from that same prompt

[lavish lake]

please let me try and figure it out

[fervent thunder]

all images were more or less the same theme

[lavish lake]

you sure it didn't have "ass" or "butt"?

those sure trigger timeouts

[fervent thunder]

the negative prompts had "excess glare, excess water" and/or "asphalt", unless "as...phalt" is mistaken by as...?

[lavish lake]

unless it was maybe a typo "ass" instead of "as"

if in doubt always refer to our https://discord.com/channels/1002292111942635562/1040299965647437905

we can't have nsfw content, even accidental gets deleted and probably timed out

[fervent thunder]

the last thing I did before the timeout was click "redo"

[lavish lake]

If you feel that the t.o. is not fair, you can always appeal/open a support ticket, we'd be gladly there to help or assist

thank you and keep it up!

[mint lily]

Can someone explain the balance between the use of tokens, as individual words, and phrasing?

If I write a description like, "a person standing on a hill holding a jar filled with cat soul"

vs. "person,hill,jar,cat,soul" .. how does SD handle grammar parsing?

[lavish lake]

a prompt first gets "tokenized", so each individual element of the prompt is a token, for instance the "a" in your prompt is a letter but it's also a token, so each time an "a" happens it will use the same token value, then that get's to the encoder and rest of the chain

[pale latch]

the model has a level of natural language processing to itt too. it understands language not just as tokens, but how they relate to each other and what they mean in different contexts

grammar can have an effect but not like it is with LLM

[lavish lake]

"person, hill," will generate something different that "a person standing on a hill"

it also depends on the model you use, with sdxl you can use more natural language

sd 1.4/1.5 tends to be more strictly tokenized, that's why you do stuff like (masterpiece), portrait, in the style of

[pale latch]

the base model's pretraining is a big component to the language processing. stable diffusion models use previously trained models to bootstrap them, if i got it figured right. clip vit-L, name sorta looks like that for stable diffusion 1.x. i think. it's openclip and vit-l in sdxl

[lavish lake]

G and L yes

[pale latch]

clip is a whole nother thing that gets trained with image paired with captions

[lavish lake]

those are the text encoders

the chain is prompt > tokenizer > encoders > scheduler

each model varies

but it's those same basic blocks

sdxl has two encoders.. clip G i believe is the one that is more natural language and clip L is kinda like 1.5 etc

[pale latch]

yeah. base training though, the sdxl model is created with these models that were trained much earlier in the chain. tehy're the pretraining

[hushed quarry]

and the fact the unet has much larger transformers helps too ;D gets more context out of the text embeddings

[solar condor]

comfy ui is faster ??

[mint lily]

@lavish lake @pale latch from my research it looks like, because of CLIP, it's tokenized and positional info is added in, then it's passed through a stack of multi-head attention "layers". So, yeah, it produces some about of relationship between tokens due to the positions being passed in, but I think it depends on the capacity -> the ability to actually understand grammar

(where multi-head attention means the learned matrices multiplied by the input end up giving more or less weight to certain parts, based on its learning of the words/relationships, and those are concatenated (ie. different heads "pay attention" different things and then they're added together (end-to-end sort of.. == concatenation))

and all of that was based on the optimization of the clip <--> image relationships during training

[lavish lake]

maybe @hushed quarry can explain this better than me but basically a token is a number... something like "hill" becomes "485789" and those numbers are the ones that the encoders use, yes it's language but then it becomes numbers and go into really big matrices or arrays called tensors, which do the complicated stuff... multi dimensions, latents, noise and diffusion

[mint lily]

yeah, the tokenization is apparently basically just words.. embedding optimizes that for neural net efficiency

(tokenization == breaking it up into the known units and converting those to individual unique numbers)

chatgpt apparently uses some learned tokenization where various character sequences end up being the tokens

[lavish lake]

LLMs are like another completely different beast... above my paygrade

I think the core is the same but on another scale

7-15-30 billion

[sweet orbit]

Anyone happen to know if there's a Tile ControlNet/ControlLora being trained for SDXL?

[rocky star]

anyone here that can help me with this problem?

[hushed quarry]

pretty much nailed it with this. Its a dual relationship between both of them. The embeddings are important to the depth of info they describe and ofc that being placed in the positional encodings for the transformers in the unet to make some relationship from with the actual images its training on

larger/better unets have the potential to glean much more info from dense encodings

[cobalt crag]

It is for me

It take only 1 minute for me to generate any image with 20-30 steps

Plus, easier to install

with built-in xformers 0.0.20

[cobalt crag]

It take 30 seconds for me to generate 512x512 with 20 steps

[true mauve]

alright what you want help with?

[cobalt crag]

Are there Token Merging in Comfyui? How to get it?

[deft epoch]

Does the server have a rule against giveaway?

I couldn't find any in #✍🏼｜rules-and-tos

[lyric estuary]

I need one million dollors

[true mauve]

........................

[true mauve]

you want to host a giveaway in this server?

[deft epoch]

I want to give a 3-month-worth PC Gamepass code for free

don't ask me why tho, I just want to.

Code is right here with me.

[true mauve]

If its not mentioned then I think its ok but I am not sure

[lyric estuary]

Can I have cash instead 👉👈

[deft epoch]

I know this crazy place where you can get your cash

it's called job.

[trim magnet]

ill trade u that for my knuckle sandwich

[true mauve]

anyways why you want to give away ur gamepass?

[trim magnet]

👊

[deft epoch]

Yes.

To someone who would use it.

[lyric estuary]

Elaborate about this... "Job" 🤔

[true mauve]

mexican nose candy seller

i could use a free gamepass 🙃

[deft epoch]

I need a mod to clarify me this first so I don't end up getting banned.

[lyric estuary]

Do it in dm?

[true mauve]

try pining some of them

[deft epoch]

I've seen people getting banned from others servers for dming members.

They considered it a loophole in the rules.

[trim magnet]

im gonna have to call the police

[deft epoch]

@thick wharf Sorry to ping you but could you please help me? It's about the rules.

[fervent thunder]

How can I run stable diffusion on cli mode?

I want to use cmd rather than gui

Found this but idk if it has same functions as gui one/ https://github.com/CompVis/stable-diffusion

[broken cave]

what's your objective? you can make api calls against comfyui and automatic

[true mauve]

------->https://modal.com/docs/guide/ex/stable_diffusion_cli

[cyan nebula]

i keep getting error code 1 on prompt .. anyone knows how to fix it

[cobalt crag]

Are there any resolution or aspect ratio for SD 1.5?

[plucky shoal]

dont go bellow 512 and above 768, you can try above but higher chance of duplicates.

[cobalt crag]

What about SD 2.0?

[cobalt crag]

512x640 will probably do

[hasty hornet]

if we're talking about base 1.5 model - it's trained on 512x512, so best results generated with same resolution. (go higher with high res fix)
Some community models can go higher, read their description before generating, they generally have some tips.

Training on higher resolution helps, but even those models might lead to duplicates and some proportion weirdness, just (alot) less likely.
So...if you want to go higher res , probably best way to just stick with highres fix, even tho it's slower.

[cobalt crag]

oke

Are there any workflow that allow me to automatically upscale generated image?

for ComfyUI

[fervent thunder]

Yes they exist

[cobalt crag]

:0 how to do it, could you please tell me?

[fervent thunder]

You can find examples at the comfyui github

[wet mural]

started with SDXL today so very new. if I want to generate quickly and in decent quality is it good to use 1024x1024 with upscaler>none? or 512x512 and keep hires upscaler on?

[warm junco]

As sdxl is trained on 1024x1024 keep that resolution or 768x1024 for portrait for example

Then you can also upscale or use the refiner

[wet mural]

ty, a single image is taking 14 minutes on a 3070 I guess the gpu is just that slow

[warm junco]

Ah 8gb vram. Then you need to edit your webui-user.bat and at the line COMMANDLINE_ARGS= you add:
--xformers --medvram-sdxl --no-half-vae

Then it should be like 1:30 per image

[wet mural]

sweet i'll try that thx

got tired of midjourney's bullshit filters so looking for next best alternative I think sdxl prolly is it rn

[warm junco]

Stable-diffusion in general is the best thing as you have the ultimate freedom of creating

Many community extensions, more models and Lora's to choose from.

Sdxl is a comparable model to MJ yes

[wet mural]

yo that edit in the bat file cut the time down to seconds

now it is becoming so yea

[warm junco]

Perfect 🙂

[wet mural]

they should include those things in the bat file by default. or maybe the automatic thing should add that by default.

[cobalt crag]

Can we use 2 Loras at the same time?

1 checkpoint + 2 Loras

in ComfyUI

[stoic seal]

what is the best prompt, positive and negative, to make sure that the generated AI image would NEVER be naked in ANY CASE whatsoever? can any one please help?

[wet mural]

I would assume "clothing, fashion, etc" in prompt and "nude" in negative although I don't think you even need it not too sure just installed a couple hours ago lol

[raven stratus]

"NSFW, Nudity, Naked, nude, topless, nipples" should be a good combination for a SFW negative prompt, possibly with a high weight. It highly depends on the models and tools though.

[fervent thunder]

didnt know, ill check the api

[cosmic summit]

Hello. Please tell me which models are suitable for drawing furry art.

[deft epoch]

I agree on not going below 512 but

I got good results when using 512x912

models like Aniverse even encourage you to do so.

Bigger heights are good for when you want full body portraits.

[cobalt crag]

I wonder...

how to train Lora 🤔

[trim magnet]

money

[distant flint]

I'm trying to do inpainting to generate background behind an image, but I'm really struggling to get the result I'm looking for. I've already done my best very roughly removing the background by hand, and I'd like the model to use this as the base for inpainting, for it looks like inpainting always start with a middle gray (for example inpainting with lower than a 1 denoise always looks terrible).
Is there a way to use the full image and mask for generation without removing the masked part at the beginning of generation?

[cobalt crag]

How to stop ComfyUI from randomizing seed?

[distant flint]

control_after_generate to fixed in your KSampler

[cobalt crag]

oki

tysm

[distant flint]

np 🙂

[terse girder]

☕

[cobalt crag]

And when I clicked Queue Prompt

it won't generate

[distant flint]

ComfyUI skips generation when it's already done it, you sure you haven't generated with the same seed and settings already?

[cobalt crag]

Or I need to change my prompts in order to generate the same image again?

[leaden ferry]

I'm trying to create a children's book for my kid - with using SD for the illustrations.
Anyone done something similar? I.e. multiple images in a illustration-like output, with the characters being taken from photos? Any guide out there?

I've tried following multiple YT videos, but most of the tutorials are using different models and my output sucks

[distant flint]

if you're using the same seed, prompt and settings, yes, otherwise you'd get the exact same result 🙂

[cobalt crag]

alright

[wet mural]

not sure if anyone would know but if you know of a good realistic model for horror, creepy, bloody stuff I'd love to know! otherwise I'll play around with sdxl base which is already pretty solid from my test

[distant flint]

I'd recommend looking up a simple img2img tutorial using whatever you're comfortable with (ComfyUI/A1111), and using a model checkpoint made specifically for children illustration.
You don't need a specific model but it might make it way easier for you to have a consistent artsyle (and will make prompting quite a bit easier)

[leaden ferry]

Thanks. Yeah, I guess I just have to keep on keeping on till I get consistent results.

Would you recommend ComfyUI vs A1111? (I have experience with the latter only)

[distant flint]

ComfyUI's node graph system can be a little daunting, but it really helps developing intuitions on how it works. Once you got your workflow down I'm sure both are good options, ComfyUI definitely gives you more flexibility though 🙂

(and there's plenty of tutorial out there for it)

[cobalt crag]

What's the difference between Xformers 20 and 21?

I just updated my ComfyUI, wishing to generate some image faster

[cobalt crag]

Do I need to study about this all alone??

[terse girder]

👍

[cobalt crag]

How to uninstall xformers for ComfyUI?
I accidentally update it to 21

[languid helm]

whats the best way to add textboxes in the images generated by stable diffusion

[knotty turtle]

Do we know if Kappa_Neuro's SDXL Loras on Civitai are legit? I just started downloading a bunch of his stuff and then discovered some comments that made it sound sketchy

[lyric estuary]

Just start creating bro youll learn most of the things along the way

[undone garden]

What's the best way to resize a portrait photo for training? 512x200 or 1310x512?

[queen pier]

👍

[empty delta]

Hey folks, is there any notebook with SD that has similar results for style transfer to styledrop by google?

[stray glade]

Does anyone have a way ro get shark fins on a character

[fallow pine]

Where I can find comfyui specific generations from users that comes with the custom nodes? To practice..

[broken cave]

what is your objective?

[broken cave]

what do you mean comes with the custom nodes?

[broken cave]

do you mean for creating a LORA, or for a fine tuning? use the example here https://github.com/huggingface/diffusers/tree/main/examples/dreambooth, observe it creates random crops for you

unless you specify "center crop"

[fallow pine]

I have seen an YouTube example where dropping an image into the comfyui interface will show up the process and nodes used in generation of that image... That way i mean

[spice plank]

Thoughts on DPM++ 3M?

[gilded anvil]

Hello. Does anybody now how can I use denoising strength with stable diffusion inpainting model?
Apparently there used to be such parameter but its gone now in recent version

[crisp dagger]

hi dudes, can i use stable on android?

[fervent thunder]

How much would it cost to train a model like the ones minevision has? Of pixel arts of guis, minecraft characters and renders?

#🏞｜general-with-images message Like this

[pale latch]

you could run it on a host PC and then connect to that with android and control it from there

qualcomm has a soc going around trade shows that can do rudimentary stable diffusion with custom models

[crisp dagger]

that sounds pretty awesome. remote computer. what.program do you use for it?

[pale latch]

lots of ways to get started. it's all open source scripts and community projects.

[lyric estuary]

wut

PNGs and their metadata don't contain THAT much of information

[lyric estuary]

one hundred billion dollars

[fallow pine]

you can regenerate your own workflow though.. by dropping the image into the interface

and i have collected a few samples from a site and those show up everything when i dropped into the interface

png can hold all data i guess but not the jpeg

[broken cave]

no

[broken cave]

i wouldn't overthink it. every image sharing website removes the metadata from pngs where those workflows are stored. either the PNG from the github repo for the custom nodes exists and has the workflow, or it doesn't

[fallow pine]

even the a1111 generated png will show up in the comfyui workflow..

[lyric estuary]

Make sense 🤔

The image browser in a11 still have all the information

[fallow pine]

i guess that's why i can't generate those from civitai examples.. they maybe shared the way so it can't be regenerated in the comfyui

[fallow pine]

you can even see all the details using a site which shows image metadata

[lyric estuary]

And not as much infos as I thought they were

Yeah just text, would only eat up few kbs

[fallow pine]

it's your choice what you want to share in your outputs.. if you allow it then everything will be with the picture

somebody knows if comfyui can work without a vae or not? like in a1111 we can turn off vae but on comfyui it looks like it's a must

[pale latch]

comfy has node inputs and loaders for vaes

[barren geode]

hey guys for my vid2vid transformation i am faced with this error message when pressing queue prompt any ideas how to fix would be greatly appreciated:

here is the error message I am getting:

Error occurred when executing ImageScale: 'NoneType' object has no attribute 'movedim' File "/Users/myname/ComfyUI/execution.py", line 151, in recursive_execute output_data, output_ui = get_output_data(obj, input_data_all) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/Users/myname/ComfyUI/execution.py", line 81, in get_output_data return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/Users/myname/ComfyUI/execution.py", line 74, in map_node_over_list results.append(getattr(obj, func)(**slice_dict(input_data_all, i))) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/Users/myname/ComfyUI/nodes.py", line 1416, in upscale samples = image.movedim(-1,1) ^^^^^^^^^^^^^

[pale latch]

make sure its all updated

[barren geode]

it is all updated i have checked that

[pale latch]

https://github.com/Kahsolt/stable-diffusion-webui-vid2vid this one? hasn't been updated since february. that might be the issue

you might not have imagescale installed iether. that seems like a partial error message tbh

oh mb you're using comfy. i dont know

[compact nimbus]

@boreal quest nice job from banning me from mj

[barren geode]

thank you for trying anyway sir i appreciate it

[cobalt crag]

Are there any prompts that allow my model to randomly generate hair color?

Is it... hair, random color?

(((random hair color)))?

[fallow pine]

if you were Automatic1111 1.52 user and upgraded to 1.6 did you notice that output results are always changing? it's not coming the same as before.. is it happening to you?

[fervent thunder]

what if i turn my ai arts into nft

[rain basin]

hey sd fam been away for a minute, lost some of my links on old pc, where's the place to check if SD recognises a keyword, famous person or artist?

some clip library or interrogator or something

[spice plank]

I would also like to know that.

[dense flint]

yo! suppose I already have an animation and I want to change its style in a way that might reinterpret the input as something else.. what kind of tool can I use for that? ideally with persistant data across frames, i don't want to process each frame individually and have flicker

[rain basin]

k i found one https://laion-aesthetic.datasette.io/laion-aesthetic-6pls/images i think there a couple others

https://rom1504.github.io/clip-retrieval/?back=https%3A%2F%2Fknn5.laion.ai&index=laion5B&useMclip=false and this

https://rom1504.github.io/clip-retrieval/?back=https%3A%2F%2Fknn.laion.ai&index=laion5B-H-14&useMclip=false don't know which is correct index tho

[cobalt crag]

Are there any way to change where Comfyui images are saved?

change directory to save images in different folders

[lyric estuary]

This is kinda my problem too currently. Was thinking will "and", "or" or "if" conditioning will actually works in prompts.

But so far in my test "random hairstyle" seems to be working. However, no short hair on females, or even buzz cut. Most of the time it shows bangs, ponytails, etc but all long hair. No curlies, wavy, afro, nothing that wild.

Anyway might want to look at this, makes more sense and it's working :

https://github.com/adieyal/sd-dynamic-prompts

Was stuck on how to generate the background/environment and make the AI randomly set the environment to both indoor and outdoor. That extension works for it {indoor|outdoor} albeit generic, empty ones.

Also that's for A1111, thus I'm going back to A1111 for the moment

[cedar kelp]

Hello sir, how to begin with this study?

[spice plank]

Google a tutorial

[cedar kelp]

any guild from stable foundation official?

[spice plank]

I dont know what that means. Are you an AI?

[trim magnet]

u can join my CoC guild we gaming

[spice plank]

As an AI language model..

[vestal dew]

Says will train a model....on the page has two bad images of two random ladies...not much of a showing of any quality work

[true mauve]

clash of clans in 2023?

[undone garden]

Do you think sdxl will ever reach midjourney in terms of quality after further refinement by the community?

[true mauve]

Yes

Remember "Midjourney is not just a model its a whole pipeline"

[floral kestrel]

Hello there, I will ask a really specific question here so its okay if you dont have the answer. Does the Amd gpu performance on SD is still best on Linux currently ?

[warm junco]

Yes

That and the performance with Shark on Windows

Hopefully AMD gets rocm windows support near end of the year

[floral kestrel]

will it change something about performance ? or will just allow us to switch to windows ?

I eard something about cuda on amd , is this something we could expect to use soon ?

[warm junco]

It will be a huge performance boost and better vram usage.
On Linux SD (auto1111) for amd uses rocm so expect the same performance if it works on windows

[floral kestrel]

yeah I mean yeah sure huge boost on windows since on windows now its crap , im working on linux epecially for that

[warm junco]

Yep, if the support is there you can switch back xD

[floral kestrel]

yeah well more disk space then haha

[fervent thunder]

Hi! Does Stability.Ai have a sales team we could reach out to? Tried sending multiple emails, but got no response. Would like to batch order a larger ammount of credits at a time.

[bleak matrix]

You can try asking in #1025467151206854736 if you like!

[willow mountain]

Dear Stable Diffusion experts. I would be glad to have an answer to the question. So. Let's say I have an image of a bag on a transparent background. Can I make it so that Stable Diffussion would integrate it, for example, into a Fashion model and the bag would be in the girl’s hands? The same thing, for example, with the dress that will be worn by Fashion Model from Stable diffusion? If possible. What can you recommend?

[spice plank]

When hitting skip, it used to save the incomplete image, it doesn't do this anymore after updating, why?

[fiery fox]

Hi guys I'm kinda new to AI and Stable Diffusion. But my pictures changes everytime (after every update i think), when i recreate the exact same photo with the seed and all the promp stuff. Same Seed, different picture. But just small details, like a bit different eye color, a bit different ring etc..
Can I just copy and paste my stable-diffusion-webui folder and never press manually on update? for after detailer and stuff, and I will alwasy have the same version?
Cause I will probably update to 1.6, and i don't have interest in even more changes to the pictures now.. If you know what I'm trying to say. Thanks

[fervent thunder]

this discord server sucks for support

[fiery fox]

seems like it.. are there other ones? That's the only one which I'm in 😅

[true mauve]

Try creating a Lora

[fervent thunder]

idk really know of any other one you have to figure most stuff out yourself

[fallow pine]

Anyone has tutorial on wildcards?

[fervent thunder]

api documentation sucks

[fiery fox]

oh ok.. maybe we become friends and you teach me everything?😂😂

[warm junco]

Hey, there are many channels for support. Best would be checking #📝｜prompting-help or #🤝｜tech-support

Depending on your questions

For any other stuff there is also #1010934719455707218

[warm junco]

These little changes are made trough restarting the webui as pytorch cuda uses different processors at startup. Also xformers can cause this.

And yes you can keep your 1.5.2 and install a 1.6.0 in a different folder. You can also use cmd flags to get your models from the other install so that you dont need to copy them

[fallow pine]

is there anything wrong about running automatic1111 without xformers? i'm not sure about it's benefits or whatever but as i've heard it can impact on output so i want to stay away from it

[warm junco]

Xformers makes the images non-deterministic meaning you can't generate the exact 100% same image as someone who used it. It doesn't affect the Quality.
But the difference is so minimal, like a hairstrain or other little details like a chain texture etc.
The performance boost it gives makes it a no brainer to use.
Most people use it.
Other settings like tokenmerge can cause the same effect even restarting the webui cause this by default.

[fallow pine]

doesn't different automatic version come with different venv folder and files? will different versions be able to work with one venv or need separately downloaded?

[warm junco]

Its using always the same venv

If you delete the venv folder it will automatically recreated

[fallow pine]

can i keep different versions webui-user.bat file in the same place or have to keep them separated?

[warm junco]

You would need to install 1.6.0 in an separate folder.

But it makes no sense staying on old versions with bugs if the new version is stable enough

[fallow pine]

i think people are still using old versions because of exact regenerations or to work with old creations

every version comes with some output changes i think

[warm junco]

No, if the code for generating is untouched.

Outputs will always change. Like I said even restarting the webui can cause this. Using different graphic cards affect that too. But not every new version will make groundbreaking changes in the output

[fallow pine]

That's true

what's the code for generating anyways?

[warm junco]

Its the whole stable-diffusion process

You can find it on github from stability ai

Automatic1111 uses it and made a usable Ui for it and included a whole lot of features and changes.

[wheat hazel]

this (foo, bar:1.2, stuff:1.3) is correct syntax in a1111 ?

[lean laurel]

Hey guys, what's the channel that people are using the bot to generate images? Or it's not available anymore?

[warm junco]

Hey, there are #1100170312106127410 - #1101178553900478464

[wheat hazel]

stupid non lalr grammar...

[lean laurel]

Thank you!!

[somber granite]

Anyone off to CogX next week in London?

[magic sluice]

how can i save my loRAS i made using rodpod to my local machine?

[cobalt crag]

Do you guys have any cool upscaler for anime artwork?

[pure sable]

Did you ever fix this?

All my text values are also there twice lol

[warm junco]

Fatal-Anime-5000

Or 4xAnimeSharp

Or Remacri

[cobalt crag]

You mean RealESRGAN?

[warm junco]

No, you can find many more upscalers here:
https://upscale.wiki/wiki/Model_Database

Put the .pth file in the models/esrgan folder to install them

[quasi carbon]

Hi guys
How to use images?

[plucky shoal]

i know it some time, they moved to https://openmodeldb.info/
But anybody going to upscale wiki will learn it 😄

[thin abyss]

How many times do I gotta vote to get rid of this 2 minute thing? I've never had it pop up before until like 10 minutes ago

[crimson belfry]

yeah i have never seen that before

I have been using this for awhile, Is there a reason why we are limited? 2 mins per gen is kind of crazy

[thin abyss]

I have too since last year actually and this hasn't happened before.

[marble portal]

is there a limit on how many images we can generate?

like credits on dalle 2

[jaunty fractal]

Anybody have any tips for good checkpoints, loras or workflows for generating pixel art?

[median arch]

hello, I have simple question
let's say I have 2 scenarios:

1. image generates with Hires. fix + 4x-UltraSharp
2. image generates then it's moves to Extras and upscaled with 4x-UltraSharp
   is output any different?

if no, then what is the reason to use Hires. fix?

I rather generate 20 images with low resolution and choose 1 to upscale but what if for some reason Hires. fix makes my images any better, for example: what if I use Hires. fix to upscale it by 1.3 and then upscale it again in Extras by the same sampler?

but let's say Hires. fix is just an option to upscale automatically... then I have another question: what if I upscale image by 1.1 many times instead of one time by 2?

__
I found that Hires. fix has Denoising strength so maybe it's purpose is also to make some "changes" to the originally generated image

[night igloo]

Hey, my stability ai account info is not loading up, I am logged in but it loads infinitely ... anyone else with this problem?

[neat cypress]

new very comprehensive tutorial for GPU poor > https://youtu.be/dpM02YMj8FY

[pale latch]

looks like the new "tracking information" conspiratard theorists are full force today. They've discovered "invisiblewatermark.py" library

i might encourage the idea that ip address and mac address are being embedded, just for the luls

same people want meta data on every image too

[median arch]

how to enable restore faces in 1.6?

what is the reason they removed it from main menu?! it was essential

[fervent thunder]

so after a whole day I've found that the AI sees mountains in the background and tries very hard to paint a new image with similar mountains. The bot gives more weight to the starting image than to the words.

[solar condor]

yoyo i have a small question ! what can i install if i have mac m2 pro to make it faster ˆ

[warm junco]

Upscaling with Highres fix is the best way to upscale images because the Highres fix can enhance/improve the image quality, depending on the denois level. The extras tab upscaler can't do this

Use extras if your image is already highres

[warm junco]

Its in settings now. It got removed as its not really helpful as Highres fix and the adetailer extension are better or even inpainting

[median arch]

but what if I want to generate like 150 images and pick only 1

with Hires. fix it would take much much longer

[warm junco]

Yea you would need to generate without using Highres fix and pic one then load it back and regenerate it using Highres fix

[fervent thunder]

great. Found a way to have some control over the image. Erase what you want by hand. Then input the image. This provides better control than stating what you don't want with the prompt.

[solar condor]

heyy any of you tried dreambooth ? does it works on mac ?

[worthy crane]

hey guys need help

[narrow epoch]

Hi guys, I just deployed SDXL refiner model on Google cloud vertex ai, I know it is a text-2-image model but can I pass in “image” as input with prompt (with the KWARGS) in order to achieve good results like what I got from dream studio?

Looking forward to any possible help here🙏🏻

[spice plank]

Honestly adetailer should be a default addition

[undone garden]

Better than controlnet tile + ultimate SD?

[warm junco]

Thats also a good technique. But it would use it after I generated the base image with Highres fix

[lilac rune]

https://github.com/SpenserCai/sd-webui-discord

I have developed an open-source discord robot based on stable differentiation web ui that supports most of the SD web ui's functions and extensions. For ease of use, almost all options are supported for selection

[fervent thunder]

i perfer google colab web ui

[potent epoch]

Has anybody made a scrap bot that takes all the keywords from all the images inside civitai to make a searchable database of keywords, I find it hard to come up with prompts to make what I imagine, and looking at other photos is not what I'm looking for because each different model can produce from the same prompt different result so the image is nothing to be based on, a searchable keywords database would make more sense, has anybody done that already?

if someone redirects me to the appropriate place I will thank very much

[cobalt crag]

What is VAE btw?

[vestal dew]

invisib;le watermarks...not good

[hushed quarry]

Stable diffusion is a latent diffusion model. The term latent here describes the compressed form the model generates images in. It does not generate pixels on an image but an image latent (compressed image). The VAE is a model train on compressing and decompressing images into this specialized format. So what happens is SD generates this compressed latent of an image. Then you use the VAE to decompress it into a full image at the end

[cobalt crag]

oh oke

Are there any VAE that is cool for 2D Anime style?

[trim magnet]

vae cant change style of a picture,only a model can

[cobalt crag]

Oh oke

Can VAE be used with SD 2.0 checkpoint?

[jagged tide]

hello, is there a salon that helps?

[copper orchid]

me taking the word Zombie out of an example negative prompt and putting it into the regular prompt

https://tenor.com/view/spoon-big-spoon-funny-mad-evil-stare-gif-17762991

[heady spear]

hello guys.....

[stable pawn]

I try to make a Watermelon pattern, like the shell/skin of a watermelon, but whatever I try I get watermelon slice and seeds with the red inside, I have a hard time with language so I looking for words and suggestions that can help me get a pattern like that, or any other tips and tricks to make game like patterns.

[fervent thunder]

CUDA error: an illegal memory access was encountered\nCUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.\nFor debugging consider passing CUDA_LAUNCH_BLOCKING=1.\nCompile with TORCH_USE_CUDA_DSA to enable device-side assertions.\n"}

can someone tell me what the issue is

[fast flax]

Hi newbie here

@ancient sail here

[lilac rune]

Someone has already tested that SD webui discord supports oogle collab web ui

[obsidian terrace]

i had made the stupidest mistake
So, i was using that TI embedding in comfy, and I was like... wtf am i getting only noise as a result, 100% of time?. Well, it turns out (embedding:80085) doesnt affect 80085. it just increases the weight of the word "embedding" 80k+ times. Turns out, you can't have numerical names only.

and no, it was actually in the negative prompt

[upbeat temple]

hii

[fading osprey]

guys how'd you like start learning about stable diffusion i want to learn SD-11111 becuz it makes really nice models but im a beginner lol should i just refer da #1080946152318443610

[shy shard]

I've created a model for someone with their likeness but they don't have a good pc. Is there a good way i can setup their model file with an online generator?

[fervent thunder]

I have the impression that ppl in artstation hate AI. I made an account and published AI generated images. I received likes, then hours later the person removes the like.

[brazen quarry]

What's the resource with the bunch of fine tunings?

Maybe Im usingthe wrong term

https://stablediffusionweb.com/#demo oh wow, I see this service includes "styles", but I know that's not the right term

Embeddings? https://civitai.com/

[crude marlin]

did they delete the show and tell chat? I'm not seeing it

[plucky shoal]

it is under SDXL BETA BOT

[brazen quarry]

How do you use safetensors in AUTOMATIC1111?

[wary belfry]

artstation people can actually create "a picture of a beautiful girl, sitting on a spaceship on a magnificent planet", and not just write a text prompt 😉 It's just not the right place for AI art

[wary belfry]

same like normal checkpoints

[brazen quarry]

It doesn't seem to be doing anything. I loaded up some neon sign one, and it just doesn't seem to have an effect?

[wary belfry]

when using a lora? do other loras work ?

[brazen quarry]

I can select it from the dropdown at the top left

Not sure what you mean by lora

It's from civitai

It does say the base model is SDXL

And I think the model I have is 1.5

[wary belfry]

ahh ok, you have to reload the webUI after your download

you then you still have to combine 1.5 with 1.5 or sdxl with sdxl

[brazen quarry]

hm

I need to do that merge step?

It says the base model is SDXL. So do I need SDXL and the embedding and then merge?

[wary belfry]

no you just need a sdxl model and then you can use any sdxl lora with it

[brazen quarry]

What does lora stand for?

[wary belfry]

it's like an extension to your base model

[brazen quarry]

How come they're not just called "fine tunings" 😛

[wary belfry]

because that means something else already 😉

there also are embeddings and hypernetworks. A lot of names for somewhat the same effect, but different techniques

[brazen quarry]

yeah I thought these were embeddings

so the actual term is lora.. .gotchya!

Low rank adaption...

"...fine tuning..."

I wish they'd just call it fine tuning 🥲

[wary belfry]

it''s like some cars are called Tesla and some are called Ferrarie. Both are cars, but then different in some ways

[brazen quarry]

Right but both are called cars! 😄

In this case, fine tuning is the more general term right? I'm not crazy?

[wary belfry]

bot you won't go to the classic gas station with your tesla 😉

[brazen quarry]

haha

I get it I get it

[wary belfry]

yea you can finetune while using a lora, but also finetuning the model itself without using a lora or anything

[brazen quarry]

Will LoRA be the term forever?

[wary belfry]

no, someone will come up with another tech and then give it a new name

[trim magnet]

its called lora because its full name is Low-Rank Adaptation

[brazen quarry]

Yeah I mentioned that here

It'd be nice to stick to a more general term

[wary belfry]

there are also LyCORIS, LoCon and LoHa ... to get even more confused 😄

[trim magnet]

locon is LoRA on Convolution Network 🤓

[brazen quarry]

Ah I see there's "Checkpoint" too on Citivai

[wary belfry]

but in practice if you use ComfyUI then you don't have to care what it is exactly, everything works as a lora

yes Checkpoints are complete models

[brazen quarry]

So with LoRA I need the merging step right?

"LoRA's like an extension" like you said

[wary belfry]

no, you just can load both

[brazen quarry]

Ah ok

When do I need to merge? Never?

[wary belfry]

are you using Auto1111 or comfyUI?

[brazen quarry]

Auto

[wary belfry]

you can merge to combine it. For example you have 10 loras of old trains. Then you can merge them to just one model instead of loading 10 seperate loras

but of course you don't have to do it if you don't care to load the loras

[brazen quarry]

Ah ok, nice 🙂

[wary belfry]

if you combine too many loras they can mess up each other at some point. Because a lora can alter parts of the model that might be important for another lora at the same time. But also they can work well together. It's a bit of luck

[wary belfry]

maybe you have to load an extension to load the other lora variants. Haven't used auto in a while, so maybe it's already added to te main branch

[spice plank]

General question but, 1536x1536, how long does it take you and what gpu do you use?

1660ti is 10 minutes per image

Technically I am using 768x768 30 steps dpm 2m and 2x hiresfix @ 20 steps but idk if that matters a huge amount. Im assuming a 4060 would get it done in like 1 minute?

[trim magnet]

13 mins 1080p image + adetailer extension without adetailer 10mins on DPM++ 2S a Karras

[spice plank]

Really? Im assuming thats at like 960 x 540 and 2x hiresfix?

[trim magnet]

yea

not a 4060 tho on a 4060 its like 2mins

[spice plank]

Man that would be nice

Kinda holding out for the 5 series, I have a laptop so I cant really just swap a gpu in unfortunately

[trim magnet]

best card you could buy for cheap right now would be an used 3090 24gb for like $700

[spice plank]

Ouch, yeah thats a bit out of my budget

[trim magnet]

or u could buy an used 3060 12gb for $200

[spice plank]

Is there a good comparison for the dpm++ samplers? I never know which one to pick

[trim magnet]

u can prob find comparisons on reddit,i only use either DPM++ 2S a Karras,DPM++ 2M,Eular A or UNIPC because they are faster than SDE Karras

[wary belfry]

5000s is said to be released in 2025. Still a looong time 😉

a strong 4000 should make you happy right now, and then skipping the 5000. Who know how expensive those cards will be now that AI is exploding in all areas

and don't cheap out on vram. So your card isn't to weak for the next big jump

[spice plank]

Oh damn, I was hoping mid 2024, thats a long wait time for whats not even a new architecture.

[vivid dome]

just saw a thing about LoRAs and lycoris models merging?

========================= a1111-sd-webui-lycoris =========================
Starting from stable-diffusion-webui version 1.5.0
a1111-sd-webui-lycoris extension is no longer needed

All its features have been integrated into the native LoRA extension
LyCORIS models can now be used as if there are regular LoRA models

This extension has been automatically deactivated
Please remove this extension
==========================================================================

Anyone know how to call a lycoris model in this new case? do I just use <lora:NAME> or am I still using <lyco:NAME>?

[cobalt crag]

What are criteria or recipe to generate excellent images?

1.Checkpoint
2.Vae
3.Lora (optional)
4.Image Resolution
5.Prompts

??

[slim raptor]

with the new face restoration change, what is the setting that make it like the old face restoration?

i keep getting bad faces

[distant rampart]

Does using SDXL increase the time it takes to generate an image?

[spice plank]

From what ive seen, no. But the default resolution is higher, 1024x1024. Shouldnt matter if its SD or SDXL as long as both are tested at 1024x1024

[slim raptor]

it does for me but since i have to use lowvram mode, it cannot be accurately accessed on my part. i hope this help

[distant rampart]

Thanks, I tried to generate an image at 1024x1024 and it took nearly 5minutes on a 3070 8GB

ok?

[warm junco]

Hey, you need to use --xformers --medvram-sdxl --no-half-vae
To get the best speed

Then a 1024x1024 sdxl image should take around 1:30

[distant rampart]

8GB is considered medvram?

[warm junco]

No only for SDXL

[distant rampart]

oh ok

thank you.

Will I have to remove those arguments if I want to use anything other than SDXL?

[warm junco]

Nope. Xformers speeds up any generation. And the medvram-sdxl works only for sdxl.
No-half-vae is needed too

[distant rampart]

perfec,t thank you. I'll give it a shot again.

[warm junco]

No problem 🙂

[distant rampart]

wow ok, massive difference in performance took 20s to generate an image

[warm junco]

Wow nice! The 1:30 I get on my gtx1080 ^^

[distant rampart]

oh that's a massive difference in terms of architecture.

[warm junco]

Yep still impressed it can run sdxl that fast. A month ago sdxl on auto took 11 minutes per image

[distant rampart]

damn!

I was using a laptop with a 2060 6GB of ram, when SDXL first came out I couldn't really get much help everyone was in such a frenzy doing their own thing so I just gave it a shot and it generated a single image at 30min and it looked terrible so I just gave up until I upgraded.

[warm junco]

Oh okay. Would be interesting how long it would take now on 6gb vram

With the arguments

[distant rampart]

I didn't have any of the --medvram arguments, just let it do it's thing.

[warm junco]

Oh okay ^^ a 2060 would need to use --medvram --xformers --no-half-vae

[distant rampart]

Would you happen to know of any recent/updated guides for setting up controlnets? I would like to do some img2img, and was told controlnets would help immensely

[warm junco]

I dont know any recent guide rn.
But its pretty fast explained:

1. install the sd-webui-controlnet extension in the extensions tab.
   Then restart the webui.
2. Download all the controlnet models from here:
   https://civitai.com/models/38784/controlnet-11-models
   and put them into the folder:
   stable-diffusion-webui / extensions / sd-webui-controlnet / models
3. there are already .yaml (config) files in there for each model.
   You have to rename each model to fit its .yaml file name. But dont change the filetype.
4. Then you can restart the webui and use controlnet.

[distant rampart]

Thank you, I was 2/3rds of the way there. I don't see any yaml files in the folder.

[warm ridge]

hi

[warm junco]

Hmm thats strange then and you restarted the webui?

[warm ridge]

can anyone tell me how to generate image here

[distant rampart]

I had installed the controlnets extension yesterday and did apply/restart, yes.

[warm junco]

Hey, sure, you can use the bots in #1100170312106127410 - #1101178553900478464
Check the pinned message in the channel to know how to use it

[warm ridge]

thank you

[warm junco]

And your in the extensions folder and not in models/controlnet ?

[distant rampart]

I'm in D:\AI\stable-diffusion-webui\extensions\sd-webui-controlnet\models

there's two files in there, the file I downloaded, and put_controlnet_models_here.txt

[warm junco]

Hmm okay, is the extension also enabled on the extensions tab in the webui ?

[distant rampart]

"sd-webui-controlnet https://github.com/Mikubill/sd-webui-controlnet.git main 56d575e8 2023-09-06 15:24:48 unknown "with a little blue checkmark at the front of that.

[warm junco]

Very strange, maybe you should reinstall the extension

Delete the sd-webui-controlnet folder and restart the webui then install it again and hit apply and restart

[distant rampart]

nope no yaml files

it's alright, I'll figure it out tomorrow when I get some free time. Thank you for your assistance.

[warm junco]

Okay, np, you can also ask in #🤝｜tech-support about this issue. Maybe we can figure it out tomorrow or someone else can

[slim raptor]

Took 30 min on my quadro m3000m

[hollow tusk]

#justmaxwellthings

[slim raptor]

I looked at used quadro m5000m for upgrade but the thing cost more than a new 4060 16gb. Laptop gpu are too overpriced

[heavy bane]

does comfyui actually feels lighter compared to a1111 or am i imagining it

[cloud snow]

Hi, I am new here.
How do I use the bots?

[slim raptor]

go to civitai, there is a new bot there that can generate images

[cloud snow]

thanks

[plucky shoal]

@heavy bane still lighter. Have to try in A1111 turn of any preview. But in Comfui with AITemplate i have 3it/s not sure it is posible in A1111. Without AITemplate Comfui 2it/s and A1111 1,7-1,8 it/s

[meager kayak]

what happened with A1111??? Now we can not uncheck the HiresFix, the interface was very good before ...

[plucky shoal]

probably you need clean install? I have not such issue.

i had it on scale to 1 so it didnt works.

also if i had it on 2 in previous session, it didnt works.

[meager kayak]

well yes but that's the problem, before we just had to uncheck a square to activate/deactivate, now we have to move a slider and it's painful, each time you have to reset your setting. It's completely useless to have done that, it was very good before

[plucky shoal]

when packed, or how to say, it doesnt work, when unrolled it works. @meager kayak

[meager kayak]

Hell, you are right

i'm dumb x)

[plucky shoal]

no you are not 🙂 it is confusing

[mossy epoch]

Hello, I made a beginner tutorial to start with stable diffusion. First tutorial I ever made. Planning to do more. I hope it helps someone: https://youtu.be/Ts0rrb0Z6os?si=YTAog37fD30wzAcQ

[molten herald]

hi

[opal comet]

Hello. I having issues with generating images. I got a picture with generation data, seed etc. I want to create that image but I got different results for the same seed every time I generate. Any ideas?

[fervent thunder]

guys is there a limit to google colab?

[mild olive]

Can you share one of those image. you can dm me the image then i can get the metadata and see if i can recreate it also.

[opal comet]

sent it

[keen shell]

What are the programs, models and promps to can make Stable difusion nsfw?

[fervent thunder]

I'm sharing images made with sdxl in DevianArt and, unlike artastion, I'm getting many likes fast. But who owns the image? I have no idea what sdxl has done to create it.

https://www.deviantart.com/leoncio22/art/Landscape-975195394 this similarly created image for ex. It has a watermark. Yet it says "created with AI"

[frozen widget]

Hi I’m looking for some insight. I’m trying to build a world including art style. I’m doing this by generating a lot of images and hand selecting the ones that fit what I’m going for. For this I presume I should get more images then normal? Say upwards of 50. Then when I train is there a way that a keyword is not connected to certain aspects. That is I don’t want the model to think of putting humans in the photo but because I don’t want to “term” to be connected to artifacts like that.

[bright breach]

Hi, I have a question, I know a little bit of ai art, is there an extension or sth that would let me edit and redraw the mistakes in the image in the webui itself instead of doing it in Photoshop?

[frozen widget]

you can create a mask and have the model rebuild an area until you are happy with the results. Webui is good for this using inpainting.

I wonder if you can take general-chats like this and train chatgpt on it to create a chatbot hyberfocused on helping people with problems they have.

[cobalt crag]

Why most of the time when I generate stuff

I always get unlucky results?

Even my prompts and stuffs are correct and alright

They are still bad.

[bright breach]

Sometimes it's cause of a bad checkpoint or lora. Sometimes it's the settings and weights you're using (like maybe using a canny controlnet with the weight of 1.6 for example), promoting is very important too. But even if you do all that right you'll still get very random results in terms of quality. That's usually why I use Photoshop to fix the mistakes like bad hands or extra legs or sth.
Rn stable diffusion isn't a one click and done kind of thing, you gotta put in the work to get the results you want xD

[bright breach]

Inpainting is SUCH a nightmare, I'd rather just redraw those parts instead 😭🫠

[viral rune]

my lora is 9 mb
and its not working

[jade citrus]

Anyone possibly help me on a issue using comfy ui, I can't for the life of me get the reActorfaceswap nodes to work its pissing me off

[sour needle]

What prompt do i use for better fingers

🤌🤞

[grand quartz]

Hi, I’m new here, is there any way to get less time before generating another pic?

[fervent thunder]

right. Deleted my project from Artstation. Not only I was getting one or two likes which were later removed, it seems that AI art drives ppl away from your profile.

deviantart it's the opposite. Likes every hour lol. There are even profiles of people who only post AI art to sell

[willow coyote]

That progression of being into prompts to being into creating your own AI models.

Then when you show people stuff you've made, and they go like "It's just AI"

You be like, "No, I actually wrote that AI"

[fervent thunder]

.... I think I'll just delete my deviantart account. Just noticed that many of the profiles who are following and or liking me arts are... profiles with 0 published works, some are profiles created many years ago.

[trim magnet]

your mistake is caring what other people think of the stuff you post

[tender cove]

Fun prompt for SDXL: Jack Nicholson stuck in a deep dirty hole. Works really well, especially with photorealism. I've been trying to get him surrounded by Smurfs, but either the smurfs are weird or he turns blue

[spare fable]

#✨｜sdxl message

[ornate quest]

gg

[tender cove]

Better than anything I managed

I'm just happy about getting weird crap in pantheon, like #🏅｜pantheon message #🏅｜pantheon message

[violet matrix]

I'd like to ask how people control the compositional proportions of the image

[thin yew]

https://www.youtube.com/watch?v=LnqmmAH3m4U

How to install SDXL ☝🏻

[fickle epoch]

Hey, I'm new comer. What's the best way to run stable diffusion model in the cloud. I know we have vast.ai, just want to check any better/opinion on this ?

[exotic acorn]

https://reddit.com/r/StableDiffusion/comments/16ar9w5/how_to_do_this_animation/

How did he made it ?

[cursive fractal]

i feel so confused and looking for possible help specicially with comfyui and creating characters from a model that i trained already. Having bad anxiety doing this yet im trying to learn something new because i cant take my ADHD holding me back but mabye i should pick something easy cause this stuff is crazy hard.

the worst part is i dont think im understanding how peolpe use discord cause i never have successfully asked a questions and got a response that i have seen. Its like this app is an ADHD's nightmare way to have to communicate with the world

[hardy swift]

can i run SDXL on automatic111 with 6gb vram?

[warm junco]

Yes if you use --xformers --medvram --no-half-vae

[hardy swift]

i just download the .safetensors file and thats it right?

[plucky shoal]

Depends on size, models are about 6GB+

[warm junco]

Yes put it into the models/stable-diffusion folder.
Then edit the webui-user.bat and at the Commandline_ARGS= line you add:
--xformers --medvram --no-half-vae

[magic citrus]

hey friends, my weekly post. I just tested Image to video using the new I2VGen-XL. It is interesting, not really usable for real use cases tho. Let me know if I missed something or if you think there is any relevant use cases for something like this.
As usual it contains a step-by-step guide if you want to test it too
https://followfoxai.substack.com/p/testing-the-brand-new-chinese-image

[median arch]

Hello, let's say I want to generate a lot of images and pick one for further processing.
I would use 512x512 for performance and disabled Hires. fix.
Now there is a problem because now I would need Hires. fix to improve details but I probably have to use img2img (Hires. fix is not there) since I want a base image as input (not prompt).

How to improve image in this case?
I'm aware of <lora:more_details:1.0> and Refiner but I don't know where and how to really use and combine everything.

In case of upscaling I have some experience so I only need help with improving existing low res image to prepare it for upscaling.

[warm junco]

You can put the 512x512 image you like into the PNG-Info tab. Then press send to txt2img. Then you can recreate it using highres fix

[median arch]

how to quickly move it from txt2img to png info?

[warm junco]

You would need to drag it from the folder into PNG info

@median arch an other method is to use the ultimate upscale extension. This works in img2img.
So you can send the image directly from txt2img to img2img and use the upscale script

[median arch]

another question: how to add details that are cropped? for example I have house that is visible only in 30% and I want a full house (image size must be increased then)

[warm junco]

Yea it need to be increased

[viral rune]

what is usually the issue of Lora not applying style after trained ?

[median arch]

yeah but I could end up with image that has content and a black color around... how to make that black color a house?

[warm junco]

A black color around the image ? Like a frame?

[median arch]

yes

[warm junco]

That can happen if the resolution is to big. You would instead need to outpaint the image in img2img

[fervent thunder]

is there a node config in comfyui that replicates the same settings used by the discord bot?

[desert onyx]

hey guys is there a max amount of loras that you can have cause the last 3 i added in my files/map arent showing up

[bright nest]

what happens in during the generation when you dont have use a prompt?

[viral rune]

pc will explode

[marble portal]

is the generation limit really gonna be removed if you vote?

[sand holly]

how do i set the default values for Automatic1111?
Because whenever i launch it and open it theres already a prompt there with a non-default resolution and steps

[sand holly]

nothing, it will still generate

assuming you're using Automatic1111

[bright nest]

yup, but i was wondering, does it only use the seed when doing this? where does it get the no-prompt image info

[sand holly]

since it has no guidance as to what to generate i guess it will just use the seed and make the best of it

[bright nest]

thats what i figured, just wondered if there was anything else going on

[sand holly]

nothing out of the usual i would know

https://cdn.discordapp.com/emojis/805393285203099689.webp?size=48&name=OldShrug&quality=lossless

[sand holly]

nvm, i must've accidentally applied the defaults at some point

[deft sentinel]

hey hlo

i have a query

i used an expired credit card for free trial will it cut money from the new credit card if i did not cancel on time?

[fervent thunder]

this is nice. Sea art site can duplicate my images.

it creates images with almost no differences

the differences are so small that when you scroll from one to the other it almost feels as if they are different frames of some video.

[fervent thunder]

can you use extentions in stable diffusion api

[cobalt crag]

I can't find this on google

help

[harsh ether]

aint no way

the 3060 mobile is beating my rx7700xt

[jolly crystal]

What is the best stable diffusion website for mobile

[eternal geode]

mage.space but its paid for the most part

also leonardo.ai and playground.ai

playground ai is free tho

[jolly crystal]

Hi, please help.. When I run the google colab. and click on the Gradio link, it starts working fine for few less than 5 minutes then it stops working, it says Error, and when I check back the Google colab I see it stopped and it turned to red button. I have tried different Gmail accounts and different SD models, but still the same exact issue. Everything was working for some time before. What to do?

[eternal geode]

Google is becoming more strict regarding colab usage and it's banning users

Be careful if you want to keep your account because free colab is over

That's all I got to say

[jolly crystal]

Fck

[eternal geode]

Yeah that sucks

If something is good don't expect it to be free forever

But you can try stable horde online

[jolly crystal]

Where are you doing your ai arts?

[eternal geode]

Leonardo and playground

[jolly crystal]

It is good?

[eternal geode]

Good enough for me

[jolly crystal]

I'll try

[eternal geode]

Try playground first, leonardo needs registration and takes time

[hasty hornet]

Google banning only if you're using webui (of any kind) on a free account

They have something about it in the rules...

[vestal dew]

A1111 now works with controlnet?

sdxl, that is

[wicked badge]

I can remove this if it seems too much like advertising but...

This month the Unstable Diffusion discord has a competition for who can make the best logo and integrate it into a PC case. Winner gets the PC; New 4090 + i9-13900k rig.

If you're at all skilled with that type of thing and are looking for a hardware upgrade, it's the sort of task that takes just enough effort for most people to not bother attempting.

[mighty iris]

anyone know how i can run cpu instead of gpu in stable-diffusion-webui-directml with automatic 41111

i added # gpu = torch.device("cuda")

device = gpu if torch.cuda.is_available() else cpu

device = cpu;, to webui-user shell file but i dont think its overriding

thanks to anyone thats able to help!

this is for deforum extension also for videos 3d animation

also getting this error sometimes Input type (float) and bias type (struct c10::Half) should be the same. Please, check your schedules/ init values (fixed)

[mighty iris]

seemed to have fixed this. can any one guide me on where to put cpu only parameters

@untold ocean @hallow edge

[fervent thunder]

is the quantity of images that the AI can generate out of one prompt infinite? or is it like, some 10^x power?

[tender cove]

@hoary pawn lol

[cobalt crag]

How to make ComfyUI use GPU instead of CPU?

[cunning ice]

A random Dunkey in the wild! 😮

[vestal dew]

Does A1111 support controlnet for SDXL?

[gritty dust]

buying a new laptop, general tips? I know I should avoid AMD and get probably 6 or 8gb VRAM at the least. I want to be able to run AUTO1111 locally and I want to be able to train LORA.

[tame juniper]

They all the same, just pick one that looks prettier to you. As you said, nvidia, 8GB vram.

[tender cove]

Free API key from @frigid swan for use with Stable Horde. If you don't already use Stable Horde and have a preferred client, you can use it immediately with ArtBot using this link https://tinybots.net/artbot?shared_key=339e6c7e-bf3f-4a0f-84ef-90518c1eea2d, otherwise enter the key in your preferred UI: 339e6c7e-bf3f-4a0f-84ef-90518c1eea2d

[hasty hornet]

why does it sounds like some kind of weird scam

[left lintel]

Oh, i can't run stable diffusion web ui 🥲 Does anyone know why

[sleek oriole]

When rendering in A1111, the image shows the progression image but when it is done, the resulting image disappears from the live preview window, but it can be found in the output folder. Any ideas how to bring the preview back?

[placid arch]

hey guys, can you help me finding settings which can run dreambooth with SDXL on a single 3090? or is it possible to add a second 3090 to double up vram.

[knotty turtle]

How are you guys doing SDXL inpainting these days? Just using 1.5, or is there a secret SDXL inpainting model that I missed?

[placid arch]

@knotty turtle search on huggingface and u find it

[mental wolf]

I've been playing around a lot but I'm struggling to get any good results - I've got a masked image of a t-shirt and I'm trying to use inpainting to generate models wearing the t-shirt. Anyone got any tips or done anything similar?

[knotty turtle]

Oh, really? I was just looking. Didn't see anything official looking. I'm look again

[fervent thunder]

I am new here. Can i generate images there? If yes then how

[clear osprey]

you can go into #1100170312106127410 or any other bot channel and type / dream [your prompt here] (without the space between the / and "dream", i put it because otherwise message gets blocked)
but if you want way more possibilities, you can either go to some websites to generate images, they allow you to choose different models, etc (but they are paid most of the time), or you can install everything on your own computer, there's lots of tutorials for that

[warm junco]

You can ask in #🤝｜tech-support if you get an error

[forest vale]

SD ever gonna look into training any sort of nice animation/video stuff ?

from what ive seen in the open source space, its good, but it feels like the pieces are all their for Stablity to do some sort of SDVid/Sdmove/sdMotion lol :P

[shut plover]

hey, is it possible to add the bot to my discord server ?

[tired creek]

hi, is there a comfyui discord, or something like that?.

[bleak matrix]

Good afternoon, everyone!

How are we all today?

[molten quest]

@knotty turtle did you find the model? I am in need of inpainting for SDXL as well.

[knotty turtle]

not sure yet. i found this one, but haven't tested it yet. i've been asking around on other servers. it's a straight mixed bag of some people saying that there isn't one and other people saying to just go search yourself. heh.
https://huggingface.co/valhalla/sdxl-inpaint-ema/tree/main

[molten quest]

@knotty turtle yea it sounds like it isnt ready yet. I havent even seen any videos on it yet.

[viral rune]

is this a sign of overtrained ? #1019361238234443776

[covert sand]

should i wait to install a1111 until i have the gpu i plan to use? is there anything specific to the card in the files/drivers or whatever or does it not matter if i download and then replace the card?

[warm junco]

If you replace the card all you need to do is to update the drivers (or reinstall) and for a1111 you need to delete the venv folder before the first run

[covert sand]

just delete it, that's it?

[sly eagle]

hello, where is the artists chat?

[shut radish]

Which OS platform and what error messages?

[warm junco]

He tried google collab free

Google seems to block it maybe

[covert sand]

what is the venv folder and where is it? i saw someone say that they couldn't get a1111 working after they deleted it.

[warm junco]

The venv folder is located in the stable-diffusion-webui folder.
It will be created when launching the webui-user.bat the first time.
If you delete it. It will be recreated after launching the bat again.

[covert sand]

okay, thank you.

[spice plank]

Is there an easy way to get something that would allow me to take a lora of a character and render the same pose to like a 360 degree view. Background isnt needed, and clothes will be consistent with the model.

Like if I typed a prompt of, standing, front, from side, from back, but more steps between?

Like imagine you opened a 3d renderer with a model inside but just panned around the model, not moving the pose or anything. Sorry this is really hard to explain

An im drink

[dreamy igloo]

I have not been reading this server for a while, but there used to be a channel for colab or other notebooks. I seem to not be able to find it anymore.

I also read that google colab for SD is again not working today.

[tiny summit]

I'm trying to figure out how to access the event here: #1072240143521554592 message

it won't let me into the event??

[simple wharf]

What process do I need to read to learn how to feed images into the local bot then get it to create new ones with those images?

[simple wharf]

?

[vestal dew]

I dont think the bots do img2img

[simple wharf]

bro wtf is this chat

[simple wharf]

I mean in Stable Diffusion local installation

[vestal dew]

you just load an image in the image box

are you using comfy, a111, sdnext?

[simple wharf]

I want to train it on hundreds of images for a model

I have A111 installed that's what I'm using

In img2img all it does it make the same image again

[harsh current]

anyone who can help me reinstall stable difussion?

i keep geting static images like you used to on old tube tvs

[simple wharf]

how do I get the img2img to actually look like the model in the first image

It's just making random unrelated images

[harsh current]

you also need to add tags

and negatives

[simple wharf]

yea i did

it makes some weird as fuck completely unviewable colored garbage

[harsh current]

add more samples

change the sampling method

[simple wharf]

it only lets me upload 1 pic

[harsh current]

get extentions

[simple wharf]

what extension

which samling method

[harsh current]

idk

you have to play around with them

see what fits what you want

[idle needle]

i spent 10 hours in photoshop doing 1 image

[lyric estuary]

those are rookie numbers

[idle needle]

Rookie numbers for a deepfake got it

[sharp fulcrum]

stable diffusion innovation seems to have stagnated a bit

i haven't noticed any significant breakthroughs since loras and locons

[fervent thunder]

man

the absence of prompt weight in discord makes a huge difference

[fallow pine]

anyone here used SD Next alongside Auto1111? if you did then which you preferred the most about the functionality and outputs?

[stable aurora]

anyone here who can help with a simple issue using Comfyui? Im just looking for the location of the folder for the Controlnet Annotators. I have them all downloaded but comfyui what to download again. Can find the location to put them in. Much appreciated

[proven holly]

thing can't even come up with a specific image i want

[mental wolf]

I've been playing around a lot but I'm struggling to get any good results - I've got a masked image of a t-shirt and I'm trying to use inpainting to generate models wearing the t-shirt. Anyone got any tips or done anything similar?

[warm junco]

Set the denois slider to 0.7 or 0.5

[simple wharf]

It makes the photo out of a bunch of other small photos

It's image2random

[covert inlet]

Can't we just use ai to make logos?

And where can i find more info on this?
I already have a 13600K and a 4070ti but i won't complain if i get the 4090 😂

[devout totem]

i have a question

fot mods

@bleak matrix

[bleak matrix]

What is it?

[devout totem]

how do i make semtex

[devout totem]

yea how do i make it

semtex

[bright cradle]

How to give image strength?bot

Any way

[bleak matrix]

If you need help with prompting, I suggest heading over to #📝｜prompting-help

[fallow pine]

anybody here using sd next?

[plucky depot]

Don't know if this is the right thread to ask but

I'm in the process of installing (first time), and looking for the latest version, I see some stable diffusion v1.4 uploaded a month ago, and others v2.1 from 2022, and I'm extremely confused as to what to download

I run on a sim card with limited data at home, so I can't miss use my monthly gb >.<

[wicked badge]

Yes, it just has to look good enough to win the competition. Info is in the Unstable server's announcements.

[harsh bane]

good morning from West Virginia

[ripe gull]

Hello everyone, quick question. For those of you that have used Seaaart.ai, they have a feature called "creative upscale" wich basically runs the same pic but fixes some mistakes. Is there anything similar in Stable diffusion?

[knotty turtle]

Have you guys found a good way of finding celebs who look similar to the person you want to train? All the common celebrity look-a-like websites I've used don't seem have to have very deep databases.

[warm turtle]

hi

[warm junco]

Hey, 1.4, 1.5, 2.1 are base models. You probably dont want to use them. For the start check civitai.com there you find the community made models. These are based on the base model versions.
Check Dreamshaper v8 for example. Its a good start and only 2gb

[plucky depot]

Thanks a lot for your answer! I'll look out for those! ❤️

[warm junco]

Oh I had a typo. You don't need the base models. Go directly for the community made ones as they will output a better quality

[robust gulch]

I bought few credits and I remember being able to see the invoice on the account page. I want to get it now, but I don't see anything, it seems app (stabiility.ai) is not loading. Does anybody have the similar problem? Can I get the invoice?

My account is [e-mail]

[wary belfry]

https://starbyface.com/ try tis

[covert sand]

can someone explain the spaghetti thing to me

[pale latch]

ai could design logos that look like other logos, from just prompts. Logo design though should have more intention behind it. Consider the purpose of the company. What their brand is meant to represent. Consider how shapes invoke ideas. Compose together a few doesn't thumbnail sketches representative of your branding notes. Use those sketches to then direct image generation to add texture or details. But keep in mind that logos are about communication first and foremost. Being recognizeable is more about their sillohuette then their colors and filled in detail. Good logos should transcend all media and look great when printed in black and tiny, or imposed over photographs on a billboard

AI is a long way off from being able to encapsulate large ideas into a creative symbol on it's own

[spice plank]

Is 1.6 vastly slower or am I insane? It also sometimes just pauses for long periods at certain %

[vestal dew]

Anyone having controlnet working with SDXL in SDNext? On the github, vlad says it's working. However, people have left messages saying it's not working. When I try to install it, it installs but won't enable. It isn't running - there are no controlnet options in the UI, and it isn't checked as enabled even after I check enable.

Anyone have it working? Vlad claims it is working.

[fervent thunder]

Any logs?

[vestal dew]

Ill have to check it later because I am running out. I have the same issue mentioned in the github comments, it keeps disabling. I'll post logs later when I return.

[midnight crescent]

hey i'm giving a talk on generative AI and just wanted to clarify whether stability AI sorted out the disputes with getty images regarding stable diffusion

@karmic brook
@wintry stream
any ideas?

[spice plank]

Getty images always tryin' to getty that paycheck

[harsh current]

can anyone tell me why my stable diffusion started generating only this gray images?

its just blank gray images

[spice plank]

This is a long shot but is there a way to find similar looking checkpoint/models somehow?

Like if I input a model name I like I can see, "oh x amount of users that used this also like this other model."

[tawdry bridge]

Hello, just wanted to ask real fast - is it okay to use stable diffusion generated images for my personal portfolio website? It's not monetized and credit would be given to stable diffusion for the images that I use. I may end up cutting off the stable diffusion water mark in the process since these would be hero images, but I will credit SD somewhere on the site for the images anyways

[simple wharf]

bro the bot literally ripped off billions of creators to learn, just post it

[spice beacon]

I used AI to create a short film for the Biblical story of Jonah. Check out how I generated scenes with text prompts then converted them into video using the latest AI tools:
https://www.youtube.com/watch?v=p-mPZH8Y6pA
#ai #filmmaking #biblestories #jonah #texttoimage #imagetovideo

[vestal dew]

again I ask, is anyone getting controlnet to work in SDNExt. It will not enable.

[warm junco]

Can you post your image settings and an example in #🤝｜tech-support ?

[reef epoch]

so im assuming the music model wont be open-sourced?

[finite cloak]

https://stableaudio.com/faqs

Will this model be open sourced?
We will be open sourcing a music generation model soon, trained on different data.

[reef epoch]

ahh okay

couldnt find that written in the blog post.. so i was not sure

[lyric estuary]

moar toys to play with

[signal loom]

Im currently using ComfyUI for SDXL but A1111 looks a lot UX friendly, any way to get it to work?

[finite cloak]

auto does work with sdxl. But also if what you want is "comfy's capabilities but friendlier UX" check out https://github.com/Stability-AI/StableSwarmUI

[signal loom]

Oh, cool! I'll check that out first then 😄

I want to experiment with turning realistic textures into stylized ones with the help of AI. I was thinking of using a real image as a base then let SDLX do the rest, that should work right? 😉

[finite cloak]

as in 3d textures, like a game asset or whatever? Yeah that's doable, you'd want an init image and maybe also a controlnet to keep the shape together (there are SDXL ControlNets @ https://huggingface.co/stabilityai/control-lora)

You might also want latent tiling, which, uhm. @neon oriole how do tiling in comfy?

[signal loom]

Yep, I do a lot of 3D and wanted to experiment with turning photo textures into stylized ones, think of something like Warcraft or LoL.

[vestal dew]

I wonder if anyone has controlnet working in sdnext with sdxl

[sour needle]

I finally know how to use google colab without a tutorial

I thought it would be complicated and need a good pc ughh ...

[covert dirge]

ei guys, can xl stable diffusion API be used for commercial use? And the other APIs?
Because i have read the term of conditions of the site and the site is not free for commercial use, but maybe the apis yes? I do not know

[urban glacier]

I'm confused: Why do I still need credits for the self-hosted version of StableStudio? I have a beefy gfx card; I'd like to generate my own

[torpid ether]

hi everyone, i want to implement mask rcnn for instance segmentation. whatever i come across in youtube tutorial is old

can anyone help me

https://github.com/matterport/Mask_RCNN

[warm junco]

Then you need to install an other webui

Like automaric1111 or comfy or invoke

[pastel lynx]

Anyone having issues using Deforum with ControlNet ? NoneType object isn't iterable ?

[neon oriole]

https://github.com/BlenderNeko/ComfyUI_TiledKSampler

[finite cloak]

@signal loom ^ you'll probably want that

[finite cloak]

oh oop i finally actually looked at the link, wrong thing

tiling as in 3d model textures where it has to wrap around the edges

in auto webui it's a lil checkbox

[neon oriole]

oh that kind of tiling

I implemented it but it never worked well enough for me

so I didn't publish it

the tiling part worked but the images were bad

[pseudo hedge]

Anyone experiencing early disconnection in Google colab as free user?

I've been having a lot of those lately

Not sure if it's because I'm a free user so low resources priority, but why now when it was fine fora whole 2 monthes ago

[vital salmon]

I've been playing with the nightly branch of comfy ui, which uses pytorch 2.1, with significant speed improvements. This build uses the new pytorch cross attention functions and nightly torch 2.1 cu121 with python 3.11**.

How does one go about using pytorch 2.1 in Stable Diffusion Webui?

[finite cloak]

if you want comfy's power but an easier UI for it, check out https://github.com/Stability-AI/StableSwarmUI

[vital salmon]

Thanks, I will look into this further and give it a try. Curious though, does this use pytorch 2.1?

[finite cloak]

it uses comfy as the backend

so whatever you got working in comfy, works in swarm

[thin abyss]

Ok.. what is up... using the same prompt I was using yesterday getting great pictures.. now I'm getting cartoon like ones...

[vital salmon]

I was able to copy the nightly comfy build into stableswarm, so far seems to be working succesfully

[fervent thunder]

Stable audio don't work for me

[forest trout]

The teams working on getting it optimized for the increased traffic.

[karmic lion]

Can i invite @radiant meadow bot to my server?

[hushed quarry]

bot is only for here atm. We are using it as a research bot but you are more than welcome to keep using it 🙂

[warped hearth]

Hi!
I’d like to know some carbon footprint info for Stability AI. I’m trying to go carbon neutral so having the right data is really important. Thanks!

[pale latch]

profile pic respect engaged :nod.gif:

[cerulean kraken]

"hardcore" which is a genre of music gets removed from stable audio prompts :/

[brave matrix]

hey guys how do i download this controlnet tile model?

https://huggingface.co/lllyasviel/control_v11f1e_sd15_tile/tree/main

[fervent thunder]

Probably thinks your talking about a certain thing if you get what I mean.

[golden veldt]

hello guys

[pale latch]

try weird subgenres like "darkcore" or "speedcore" or whatever. there's a buncha weird core genres

[trim magnet]

https://www.last.fm/tag/obamacore/artists

[alpine tulip]

Hi everyone 🙂
I like the dude, that is posting on that site on insta: https://www.instagram.com/claracolor.ai/

1: Can anyone of you professionals here detemine, what model he is using most of the time? Is there a way to determine that? They have kind of a flat look, but still realistic in a way. Not hyperrealistic like other popular models, imo.

2: I especially love that pic of the girl on the bike. I tried to reproduce something similar like that with inpainting, bike loras.. without success so far. Would u have a advice, how I could do something like that? How did he do it?

[fervent thunder]

For some reason when my gpu gets hot or smth the process is slower

Is there a setting for it?

[pastel lynx]

Use low Vram mode

And be careful to not burn it lol

[hollow creek]

Can anyone explain how Clip G and Clip L of SDXL works?

[paper herald]

Hey community, I am looking to hire someone experienced in sd for a personal project of mine! Is this the right channel to ask in?

[bleak matrix]

Good morning, everyone! How are we all today?

[bleak matrix]

I suggest asking this in #🤝｜tech-support

[chilly dew]

Fed up with creating single-character images. I'm looking for SDXL templates that allow multiple characters to interact with perhaps a bit of additional controlnet. What could be the workflow to use? And to your knowledge, does this exist?

[lyric estuary]

image to image; use photo of bunch of person posing. Done this and SD actually recognize all of them and generate similar images

OR controlnet definitely, using openpose

[potent spire]

By working with imo best workflow possible for pure gen AI work: Layering and inpaint

Generate layer by layer

Beginning with background/environment

[clear osprey]

i have a friend that want to install a1111 and sd, where can i find a good tutorial for him please?

[lyric estuary]

google.com?

[clear osprey]

bruh i know but idk what tutorial to send him

[lyric estuary]

send nudes

[pale latch]

he's probably taking shortcuts. I don't see a lot of post work in his creations. Or detailing of any kind. my guess is he started with other popular photos and controlnet them into this sd15 year old girl he created.

[warm junco]

Here is a Install Tutorial (if he has an Nvidia GPU):.
https://m.youtube.com/watch?v=7auOOOqGyt8&pp=ygUJY3JlYXRpeGFp
Also read the Pinned comment under the video.

[hallow axle]

hey why is the enable hires fix button not in a1111 anymore

how do I turn it on and off?

[warm junco]

Its active if you click on it to set the hires fix settings and deactivated if you collapse it

[hallow axle]

oh

ty

[pale latch]

anyone got tips for doing img2img with the new v2 of animatediff?

[floral umbra]

wonder why there's no easy to use and fast .safetensor to tensorRT yet, as apparenttly, on a 4090, it speeds up generations by 2x, and like 30% on 30 series cards

[pale latch]

there's obviously issues if it hasn't been widely deployed

[wary nacelle]

<-- n00b. Question: can you run 1.5 and xl in same install or do you need to separate them?

[spare fable]

You can do the same install for both, but there are some "rules of thumb" you'll need to understand. SDXL works best with 1 megapixel starting sizes, so those 512x512px starting sizes you see for 1.5 don't do very well with SDXL. (You can find a table I created here for help with good resolutions for SDXL: #✨｜sdxl message

Also, you need to be cautious not to use something made for 1.5 on something SDXL The models, LoRAs, VAEs, ControlNets, etc., are not cross-compatible.

Also, the prompt behavior on SDXL is much better with natural language than 1.5 ever was.

There's more, but those are some of the bigger points. But yes, you can run both through the same installation instance.

[wary nacelle]

thank you soul...Ive been running it side-by-side in same install for a while, and...I wonder if running 1.5 models and xl models at the same time (in different browser tabs)....will cause some weirdness...

i typically have 5-10 tabs open at any given time, using various checkpoints/loras etc...and I "queue them up"

[spare fable]

Ah...that might bung up stuff as things get loaded into memory. Certainly with as much memory as SDXL eats up, you might run into something like OOM issues, depending upon your equipment.

[wary nacelle]

yeah....i think so too...i think i see models just flat out fail after a while and that even prompts gets "bleeded over" to the next run....

say i have a prompt that focuses on the color pink in one tab and its 1.5, then a xl image that focuses on mainly metalic stuff in another tab get pinkness in it 😄

i guess the best bet is to have them separated

i have a 3060 btw with 12GB

thank you for your help @spare fable

[nocturne tusk]

Hello

[normal pike]

@nocturne tusk hello

Hello everyone.

[nocturne tusk]

Hi

[normal pike]

What are your thoughts on Easy Diffusion? Anyone have experience with it?

[simple wharf]

Anyone know how to get Realistic Vision v3 to stop making 100% perfect looking people? I want some average looking humans not 10/10 babes every time

[chilly dew]

@lyric estuary and @potent spire thank you for your advices.

[potent spire]

You are welcome 😄 Thats how i do it as well when i work with only gen AI stuff

[simple wharf]

I even tried using 'ugly' and they all turn out stunningly beautiful

what gives

[lime field]

idk

[clever cradle]

What's a good AI Art app for mobile?

[lyric estuary]

Depends, good for what?

I use FaceApp and Remini a lot, other than that I use Pix2mix on Telegram too

They're all (or claim to) use AI, to some degree at least

[clever cradle]

Good for making original images

I'm looking for something similar to Automatic1111 in function. Something that will let you use your own models.

[lyric estuary]

lol good luck with that

[clever cradle]

Yeah I didn't think such a thing existed either. But never hurts to ask righr?

[covert sand]

someone told me if i get a new graphics card i should delete my venv folder. is that correct?

[pale latch]

generally not needed.

maybe if you're going from nvidia to amd card, but why do that? also, just remove --xformers, or let it fail to load

[spice plank]

Yeah, if you want average people try either analogmadness or lazy realistic amateur. Realistic vision isnt uh, realistic.

[covert sand]

nvidia to nvidia. . . also im not really sure what you mean

remove xformers and load it up, and then put xformers back in later?

also. . . should i even use xformers on a 3090?

[pale latch]

oh yeah use xformers . memory optimizers are good