🏞｜general-with-images

[dense nova]

https://community.amd.com/t5/ai/updated-how-to-running-optimized-automatic1111-stable-diffusion/ba-p/630252

[long folio]

That could be also just said as "how to make a photo into a drawing"

[weary urchin]

i have this

i remember the weird name

[junior sky]

Check out the cosxl edit model for that

[late sorrel]

that's why i stopped using that a long time ago LOL

[dense nova]

You need the --directml flag for AMD GPUs. Definitely.

[wild sorrel]

wait...I don't see --use-directml argument on this one too

[weary urchin]

uhh its broken now

[long folio]

https://huggingface.co/stabilityai/cosxl this?

[weary urchin]

wont even load now

[dense nova]

You have to add back in that flag...

[weary urchin]

i did

[late sorrel]

cosxl is very cool for editing, hopefully the editing stuff is improved in sd3

[weary urchin]

whoops

wait

yeah works again

[dense nova]

--use-directml

[weary urchin]

yea i did it

i frogot the use part

[dense nova]

Okay. Might have to give up on SDXL though.

[junior sky]

Yes. The edit checkpoint. Use it with ComfyUi. I don't think auto1111 supports that.

[weary urchin]

yeah thats fine

i just want to be able to make realistic looking thinbgs

idm how

[wild sorrel]

🫠 that's probably issue for #🤝｜tech-support then, I don't feel very helpful on amd part

[dense nova]

As long as you stick to just anime girls / asians, SD1.5 can probably do sort of a little bit of realism.

[late sorrel]

this is why nvidia > amd

[junior sky]

I think there's a comprehensive zluda guide somewhere

[wraith tundra]

Hi .. i cant find dreambot

[weary urchin]

yeah but i got it for £200 off

[wraith tundra]

Can you help me guys ?

[weary urchin]

so like

:DDD

[wild sorrel]

you're gonna get lots of problems worth more than those 200£

[wraith tundra]

Holaki can you help me ?

[dense nova]

Try this model out. https://civitai.com/models/7371/rev-animated

[weary urchin]

yeah im hella stupid

[weary urchin]

kk

[wraith tundra]

XD why ur saying that

[wild sorrel]

ain't bad model, but has nothing to do with realism

[weary urchin]

i just want realism 😦

[dense nova]

Argh. Hang on then. I'm looking...

[weary urchin]

cuz i aint even know how to fix basic issues

also i was blinded by the low low price of my gpu

so i went amd

[wild sorrel]

have you tried any other model if they work like...at all?

[weary urchin]

they all just tweak out when i go above 512x512

[dense nova]

Okay here: https://civitai.com/models/4201/realistic-vision-v60-b1

[wild sorrel]

weird

[weary urchin]

kk

[long folio]

where do I put it? SD checkpoints or something else?

[dense nova]

But you're going to need to render at at least 768x512 or 512x768.

[weary urchin]

which one

[junior sky]

In the checkpoints folder, yes.

[dense nova]

The first one. (It's a safetensor, 1.99GB.)

[weary urchin]

kk

[dense nova]

I'm downloading it now too. Will try to match results.

[weary urchin]

kk

[long folio]

Is the edit checkpoint capable of not preserving forms of objects? As in: original image had a perfect circle, output will have it distorted slightly

[hazy warren]

SDXL

[jovial tiger]

very neat

[junior sky]

Uhm, I have no answer for this question. You need to check that for yourself. I am trying to find a workflow for that at the moment. I am pretty sure I have posted an image here somewhere.

[long folio]

sure, no worries whatsoever, just asking

[dense nova]

Try these settings. Not great, but I want to know if you can render at over 512 or not.

[late sorrel]

wait they updated rev animated? i didnt even know that lol

[dense nova]

Yep. Just this month. 🙂

[late sorrel]

nice

[late sorrel]

yea this is a current problem, hopefully sd3 edit model is better 🙂

[junior sky]

Found the workflow. It's in the image.

[late sorrel]

the workflow fixes the cosxl problem? 😮

[junior sky]

Wich Problem?

[late sorrel]

oh sorry i thought we were talking about the same thing haha

[long folio]

Can I prompt the edit checkpoint with an image?

[late sorrel]

i mixed some conversations

[junior sky]

If you fire up the workflow you will see that it needs an input image

[long folio]

How do I get the workflow from the image? 😬

[junior sky]

Then you prompt what style you want and it should work.

[junior sky]

You drag and drop it into comfyui

[long folio]

Oh okay, thanks!

[weary urchin]

while im generating my fans start becoming really loud

is that normal?

or is it because im using my cpu or something?

[late sorrel]

oh but wait, i think we are talking about the same thing... you linked him that workflow because he was asking about cosxl distorting the form of objects right? and i asked if that workflow you linked fixes that problem, which i am assuming it does if you linked him that.

[dense nova]

100% normal. Probably won't need a space heater in winter. 🙂

[weary urchin]

xD

[languid pebble]

💖

[weary urchin]

oh my temps are both only around 55

[late sorrel]

that is terrible

for sd, my temps never go above like 35

[dense nova]

This is so not important... Don't you have a 4090?

[late sorrel]

nope

[dense nova]

IDK what my temps are.

[junior sky]

I think I never had a real problem with Cosxl edit. And that's pretty much the standard workflow 🤔

[dense nova]

RevAnimatedV2.

[late sorrel]

well i did notice cosxl does have some distortions sometimes

[weary urchin]

NEVERMIND

[late sorrel]

❤️

[languid pebble]

Sweet how much you people care for others! Had to send a heart!

[late sorrel]

hehe

how you say in german: meine liebe LOL

[languid pebble]

And sometimes it's working. One day I helped a person and next day i saw him here helping another.... 👍

[late sorrel]

yea

it's a very nice community

[late sorrel]

are you good now or still problems?

[weary urchin]

got a bigger than 512x512

@dense nova

[weary urchin]

im gonna try to go bigger now

my pc is SCREECHING

[late sorrel]

yikes

[dense nova]

Hang on. What you want now is hires-fix.

[late sorrel]

time to get a new pc

[weary urchin]

oh ok

[weary urchin]

i got it like 9 months ago

[weary urchin]

just that? plus what i already did?

[late sorrel]

it's just crazy to me that you have high temps and screeching with 16gb vram and 1.5 models

[weary urchin]

i perhaps may have 100V too little on my psu

[weary urchin]

kk

[dense nova]

Using hires-fix to go from 512x512 to 768x768. This upscale step is how you get fine details in there.

Not sure my settings are perfect though... I'm sure there's a better sampler for SD1.5.

[late sorrel]

that's why sdxl was not working for him, if 1.5 is barely working lol

[weary urchin]

how many sampling steps is that?

i cant see it well

[dense nova]

LOL sorry I used 50. Just use 35. Should be fine.

[weary urchin]

kk

my pc is crying

i thought it was good :((

[dense nova]

Did it die? 😦

Hang on...

[late sorrel]

your pc was good for video games... but not for this...

[weary urchin]

nash im here lol

[weary urchin]

yeah i get a ton of fps lol

[dense nova]

Okay. Try this.

[languid pebble]

As long as it keeps on working it should be fine ... some cheap airvents are just loud

[weary urchin]

oh

[dense nova]

Base size 384x512. Hi-res fixing up to 576x768.

[late sorrel]

ye but he got out of memory problems too

[weary urchin]

yeah these were £5 each

oh i have to go

tysm both of u for all ur help

[languid pebble]

Have a good evening!

[dense nova]

Well best of luck. Maybe AMD will have AI one day too.

[weary urchin]

yeah :/

[languid pebble]

I was teased to buy AMD first, too. Than I got the information about A.I. ...

[late sorrel]

A Major Disappointment (AMD)

[languid pebble]

Cheaper, more VRAM ... sounds good for a second ...

[late sorrel]

it's clickbait material

[weary urchin]

yeah ://

[languid pebble]

But for sure ... someone just starting with A.I. will not know ...

[late sorrel]

understandable

[languid pebble]

Thank god I had an Nvidia before so I got the information when I installed SD first time ...

[late sorrel]

same

i swear if 5090 is not at least 32gb, im gonna cry

[languid pebble]

I hope my 4090 will last at least one more year ... even though I have a 5k € coupon for ASUS store ...

[late sorrel]

😮

dude... discord is shit i swear.... i clicked multiple times that i want to hide the activity cards in the top right corner, but it keeps showing them, ugh

i guess it resets every time you close discord

[languid pebble]

I have problems with my sound configuration .... cause I connect my BT headset if I need and have to configure manual ... ^^

[long folio]

How do I get ip2p in ComfyUI?

[late sorrel]

let me link you a video

this covers all the current ip2p variants: https://www.youtube.com/watch?v=sP6CEx-UF70

you dont need to understand german, just follow what he is doing and you should be fine

[languid pebble]

😂

bookmarked ^^

[late sorrel]

nice

[long folio]

I don't understand something

Why where how

[late sorrel]

you need to update comfy i think

[long folio]

How do I...?

I might be stupid, but at least I'm not smart

[late sorrel]

from the manager

[long folio]

all I have is a folder with comfyUI

[late sorrel]

ok follow the installation instructions of this: https://github.com/ltdrdata/ComfyUI-Manager

comfy should have that node by default :3

[long folio]

### ComfyUI Revision: 1756 [d19de275] | Released on '2023-11-29'```
lmao 😭

[late sorrel]

yea now update within the manager

but thats some old comfy version, you dont update comfy?

i mean i guess it depends what you doing

[long folio]

I've left it in the dust for too long

I think I'm cooked, the manager ain't here 😭

[late sorrel]

you dont have comfy manager custom node in the custom nodes folder?

[long folio]

I do

[long folio]

I git cloned it

[languid pebble]

Did you restart comfy UI?

[late sorrel]

can you git pull comfy from within the comfy folder, it should update it

[long folio]

multiple times

[long folio]

like git pull https://github.com/comfyanonymous/ComfyUI ?

[languid pebble]

only "git pull"

in the comfy directory

[late sorrel]

https://www.youtube.com/watch?v=K9UuG9EuOFA

this dude does the git pull and shows your situation basically

[long folio]

It works 🔥

Now nothing can stop me 😈

I'm going to generate ROCKs and STONEs

[languid pebble]

... can break your bones ... ^^

[long folio]

no reason to why use this as an example...

[languid pebble]

It's a pretty nice toy I'm using, too

[long folio]

Can I prompt IP2P with a second image?

like if I have one image and a second image from which I want to take a style

[languid pebble]

You can have 1 pic for the style and another for ... missing word ... composition

[long folio]

so I can have Ryan Driven Droven Driver Gosling for composition and second for style? Where do I put the style?

[languid pebble]

https://www.youtube.com/watch?v=BCeCTKH0wKs&t=491s

Workflow sould be in description

[long folio]

Thanks a lot

[languid pebble]

You are welcome!

Gotta sleep now! Good night!

[long folio]

💀

[long folio]

Good night then!

[languid pebble]

In your manager you will find "install missing custom nodes"

[late sorrel]

of course you will generate rocks and stones... mhm :3 but awesome you got it working 🙂

[long folio]

What else there is other than rocks, stones and Ryan Driven Droven Driver Gosling?

[late sorrel]

so now dont forget to update 🙂

now you can generate wai... i mean rocks and stones

[long folio]

🤨

[late sorrel]

it's something only men of culture will understand, don't worry about it

[mystic path]

https://tenor.com/72NY.gif

[spark finch]

i did something wrong

and I am not sure what, cos the steve buscemi is mixed with ron perlman,

which is not at all what I intended

[summer scroll]

天空

[spark finch]

Not really, I would not say it is

[wispy nest]

tried to edit to make some changes to hand weirdness, then upscale. fixed the weird hands, but now theres random hair. fantastic,

[spark finch]

did you want it more like this ?

got rid of the hair, but still some bugs, I think

[spark finch]

so when 'm making firearms, the support and handle are often misaligned, how do I fix this, if I could fix this it would be perfect.

[late sorrel]

lol

[grim drum]

Please draw me a picture of a short sword, in the ancient Chinese style, with seven jewels set in the sheath

[pallid ruin]

I tried 😁

[hazy warren]

need some Warhammer images

[hazy warren]

needs some warhammer

[floral ledge]

this takes a while huh

[nimble mason]

Here is the image you requested.

Here is the image you requested.

[hazy warren]

should use Forge

[floral ledge]

what is that

sorry im new to this

[hazy warren]

a better A1111

[golden quail]

sometimes sdxl is generating image with multiple scenes inside, how to avoid that I only want single scene inside the image

[soft meadow]

Hi anybody know how to fix this ?
Something wrong with my AnimateDiff , v1.9.2 using --medvram and --xformers.
AnimateDiff doesn't come out on my UI

[nimble tree]

@shut sinew

Idk who to tag that's 100% sus

[frank scarab]

一个老人站在海边

[nimble mason]

hit the exclamation emoji, it'll disappear immediately but trigger an alert to the staff

[clever oar]

i download alien model copy promt from civitai and get only shit)

so bad quality

[languid pebble]

You need cats on space ships!

[clever oar]

i have MJ version

I want SD to listen the same way

[languid pebble]

New Enterprise crew member: Xena 😄

[late sorrel]

is that comfy or something else? i personally can only help you with comfy, not familiar with other UI problems sadly 😦

maybe you need to update either your UI or extension

[soft meadow]

Automatic 1111

[late sorrel]

did you try updating a1111 or extension?

[soft meadow]

Ya latest v1.9.2

[late sorrel]

also check the github issues page for the a1111 version of animatediff, maybe someone has the same problem as you and there is a solution

did you update the extension?

[soft meadow]

Ohh okay

[late sorrel]

cause sometimes with newer versions of UI, you also have to update the extension

or maybe animatediff is currently not supported with latest version of a1111

so you have to wait

[soft meadow]

I download from the weiui that and also check the version said latest already

[late sorrel]

ok then maybe the animatediff extension needs to be updated for latest version of a1111

[soft meadow]

Ya I think both of them already latest version

[late sorrel]

idk what to tell ya, aside from telling you to switch to a better UI 🙂

[soft meadow]

Ok find btw tQ for answering

[late sorrel]

haha gl

[storm jackal]

any ideas why every picture is so grainy and pixelated?

ComfyUI, using the usual prompts I used in A111

[nocturne oak]

Is that the actual size of the image you were generating?

[storm jackal]

nope, just cropped

but everz image results the same

[nocturne oak]

Depends on factors. What model, what size, what CFG, etc.

Looks like it might be your VAE setting. But you need to share more information.

[storm jackal]

yes, it was

I loaded sd vae while using sdxl model, didn't notice it

[nocturne oak]

There you go.

[nimble mason]

fuck yeah flying fish at cruising altitude that's my style

[late sorrel]

look at that lad

[covert pagoda]

Look at that lad

[umbral mortar]

Flower pot

[nimble mason]

Here is the image you requested.

[late sorrel]

lol

[clever oar]

[cat: dog: 0.65]

[clever oar]

fish day

[deft bison]

good work ! ...

[clever oar]

thx

[clever oar]

they have evil faces😄

[clever oar]

you know name this fish?

[deft bison]

Yes ! ... very strange! ... 😀

[clever oar]

[cat: spider: 0.5]

[cat: turtle: 0.4]

[late sorrel]

the stuff of nightmares

[deft bison]

surely new species that scientists will have to name! ...AI never ceases to surprise us! ... 😄

[clever oar]

no this is an old extinct species of ichthyosaur, this is lora 😃

[deft bison]

😀

[strange wave]

how to draw a AI picture

[clever oar]

i want know too

[torpid olive]

two eggs in the sea

[unkempt nebula]

image of a poor indian boy who is a rag picker searching for something in garbage site

[arctic laurel]

sdxl finetune on voxel art

ok i’m done

[wraith nebula]

站在高山之巅，脚下是茫茫云海，云朵像波涛一样翻滚着，远处是连绵起伏的山脉，山峰被云雾缭绕，若隐若现。太阳的光芒透过云层，照射在云海之上，形成一道道金色的光柱。

[storm jackal]

did upscaling myself this time and combined 2 tutorials together (1 image where the worflow was shown)

[late sorrel]

nice

[languid pebble]

Hi! I'm new to SD. Anybody got an idea? 🙂

[late sorrel]

you are not new to sd 🙂

[languid pebble]

Dang! Got me 😄

[late sorrel]

❤️

but i sadly cant help you with your problem 😦

what are you even trying to do?

[languid pebble]

https://github.com/if-ai/ComfyUI-IF_AI_tools?tab=readme-ov-file

ComfyUI-IF_AI_tools is a set of custom nodes for ComfyUI that allows you to generate prompts using a local Large Language Model (LLM) via Ollama. This tool enables you to enhance your image generation workflow by leveraging the power of language models.

[pearl bolt]

Hey all, I hope I'm at the right place to seek for some help.

I'm having trouble with sdxl turbo, after adding it to the webui it produces quite weird low quality images. I've followed a youtube tutorial how to set it up, the guy gets awesome results and i get garbage. Normal SDXL works perfectly fine (but its slow) What am I doing wrong?
(on mac studio m2 ultra)

[languid pebble]

Always have a look at the model description on civit.ai ...

[pearl bolt]

sdxl, and turbo

[languid pebble]

They will tell you about the right settings for the model you use

Youtubers sometimes forget 🙂

[pearl bolt]

what exactly should I look for there?

[languid pebble]

cfg scale, steps, sampler ... you will find examples in Text or in the picture information

[pearl bolt]

actually I'm seeing similar comments on the page 😄 like: "Respectfully, I can't get this model to produce anything but burned-out nonsense. Please post some guidelines"

yay! fixed it? I think. default CFG was the issue

now it looks fine, its just weirdly generating two heads for everything 😄

[languid pebble]

Wrong picture size?

[pearl bolt]

512x512 is what it likes

[languid pebble]

Ups ... tbh I haven't played a lot with turbo models but that sounds pretty small for a XL Model

try 1024*1024?

[pearl bolt]

i mean before upscaling - smaller file size just speeds it up (a bit)

[languid pebble]

Most XL Models are trained for 1024*1024 before upscaling

[pearl bolt]

trying it, although it does not seem to affect the image result itself

[languid pebble]

Let's hope for someone else with an idea. I just forget my mistakes so I'm not really good in reverse engineering 🙂

But you could try another turbo model ...

[pearl bolt]

😄 haha, thanks anyways
Turbo is an official release from stability ai, and the only thing it should do is improve the speed of sdxl - which is quite handy on mac as even with an insanely powerful mac studio ultra it is slow (not really optimized for macs)

sdxl generation wit every speed up option is minimum 30-40 seconds

[languid pebble]

One of my problems ... I have absolutly no idea about Macs 🙂 There's also a lightning technology to speed up things ...

[pearl bolt]

turbo 2-3 seconds

[pearl bolt]

ive seen something, theres a ton of features and nice things that are faster on pc, but for everything else I use mac, so i have to stick with the platform for this as well - with all the drawbacks mac has when it comes to ai image generation

[languid pebble]

I do understand that. But it's one problem for Mac users: harder to find support, cause many don't have the experience. Maybe ask again here later. I'm sure you will find a solution.

[pearl bolt]

the main appeal was that its XL and faster

[pearl bolt]

yep, thanks, will do that

yeah im not the only one experiencing double heads, read a bit and it seems like its a common issue, and the problem is that negative prompts are not working at all with sdxl turbo

[languid pebble]

Awww... it's a problem I've had when I wanted to create pictures bigger than the model was trained for. But that's not your problem. You can also try here: https://discord.com/channels/1002292111942635562/1002602742667280404

[pearl bolt]

good to know 🙂 yeah thanks, im new here so thanks for pointing me in the right direction

[grizzled sage]

thats a custom node problem. outside of the scope of just "stable diffusion". If you're having trouble with comfyui's visual code editing, then you might be better off trying a simpler UI experience.

To solve these problems and ongoing comfyui workflow compatibilities, you're going to need to learn some idea of coding. A node graph is basically a script. If you have no idea how to code, you'll have no idea how to debug a node graph

[languid pebble]

Thanks! Got the idea for this workflow here ... and sometimes it's just a little problem with me beeing stupid 😄

[grizzled sage]

i just realized you said you're not actually new to the scene. mb.

new users should stray from comfyui imo. its more advanced. you seem like an advanced user though

[languid pebble]

Just not really used to comfy that much yet ... but starting to learn basics at the moment. Cause I think SD3 will change a lot again ....

And sorry for my humor ... I should watch it a bit more 🙂

Welcome to NeoTokyo Highway ...

[lost valve]

Asian Swift

Asianest Swift

[long folio]

Anyone has a ComfyUI workflow for IPAdapter Compomsition + Style?

[wispy nest]

Woah. Are you him? I feel like I am in the presence of royalty.

[arctic laurel]

HiDiffusion does indeed work and it’s impressive

[wispy nest]

Illustration for “Pickman’s Model” by H.P. Lovecraft

[crisp stream]

it as well includes a checkpoint + loRA merger + hi-res fix

had to quickly swap out the prior brightness + contrast after a Comfy update and the new ones are being quite strong even if you change it by the smallest amount, so you might wanna look for an alternative

It's IPAdapter plus btw

[jovial tiger]

I think because of the high contrast nature of your pics, they work well as style images for ipadapter

[crisp stream]

yes, could be, looking good 🙂

[languid pebble]

🙂

[unique condor]

A lot of these remind me of Stephen Gammel's work with all the drippy stuff, but more toned down waterpainting version of it(he's a huge inspiration to both my real art and ai art). Pretty cool stuff though

[crisp stream]

Thank you, I see, he isn't part though btw 🙂

[crisp stream]

The sun always shines on TV 😄

[languid pebble]

running with oil?

[crisp stream]

it's the essence of oil 🙂

[languid pebble]

Learning Comfy ... dang! I should have done before!

[crisp stream]

it's ok, just started using it intenesly a few weeks ago. After all, it's about the result anyway, at least here 🙂

[languid pebble]

But even if you are not really a fan of it you learn so much more about A.I. And you might become a fan 🙂

[crisp stream]

was that a thing for you?

not the Image btw 😀

[languid pebble]

Well ... I have been fine with Forge. But learning has always been a target. Even if I'd return to Forge I think I'll be able to handle it better.

[crisp stream]

yes, I see what you mean

[languid pebble]

For sure you can create pictures with Midjourney. But that's not really what some of us do. We have to learn and understand. Cause we don't wanna create nice random pictures.

[crisp stream]

Confy is certainly an extension to possibilities, yet not necessarily necessary for pleasant outcomes the way intended

[languid pebble]

It has an influence on your ideas and intensions ... I'd think.

One of the reasons for me to be here is that there are people with different or more knowledge 😄

[crisp stream]

Yes, additional possibilities arising incorporate into work flows, just saying even without it´s already fairly doable

[languid pebble]

Sure and Forge is pretty easy ... generate picture, send it here or there or there again 😄

[crisp stream]

🤭

[languid pebble]

🙂

Welcome to Ocstost!

I need a new idea to play with ... this white stuff only looks so nice for a few views 😄

[covert pagoda]

Mmmm

[jovial tiger]

@worn ermine can you paste a screenshot of which node is red showing some surrounding nodes

?

[worn ermine]

yes

those are them

[jovial tiger]

you could just replace it with a regular ksampler with those settings.

[worn ermine]

k thanks

🙂

[worn ermine]

not working with that and i can't find that exact ksampler

i meant to say a regular ksampler doesn't work

its cool don't worry about it. i'm just gonna let it go.

may be working!

[wispy nest]

ITS MY BIRTHDAY! 🥳 🎉 🎈 🎂 🎁

[wispy nest]

These two were made from NASA images

[unique shuttle]

I'm new using SD and I'm trying to figure out where to start on improving a piece of art i have using SD. can anyone assist me?

[languid pebble]

Let's hope someone with the needed experience coming in ^^

[languid pebble]

Maybe you can show us your problem and tell what you use?

[unique shuttle]

i am trying to do something with this picture i was basically following guides on youtube but it didn't come out right.

[languid pebble]

Nice one. It's not really what I work with a lot but my Idea would be an Image2Image using controlNet. You give the A.I. your image as scribble, add a fitting prompt and it should combine both.

Ohh... should work with Text 2 Image, too. I can't create such nice art with my hands so I don't really use it.

[unique shuttle]

this was a gift to me a few years back. i have been trying to figure out what kind of prompts to put in to aid in good results.

[unique shuttle]

holy crap

[languid pebble]

An example, prompt: Painting of a young asian male with long hair

[unique shuttle]

that was from image as a scribble?

[languid pebble]

I've added your image to ControilNet, enabled it, turned off the preprocessor, and told it to resize and fill ...

In Text2Image Mode

A good idea for prompt could be to find a concrete name for the hair style and if you like add something for the clothes ...

[unique shuttle]

i know the name of the hair style. i heard that some words the ai cant pick up on or input well

[languid pebble]

I alway mix zhem up ... Rasta and dreads ... and so on

dread locks ...

rasta locks

[unique shuttle]

I'm still doing something wrong. i have been getting random images not even similar to my image.

[languid pebble]

What's the UserInterface you are using?

[unique shuttle]

honestly not sure i only downloaded it a few days ago and been playing around with it

[languid pebble]

In Stable Diffusion Forge it looks like this:

I don't really know todays Automatic1111

It's just the ControlNet stuff on the picture. It will not be perfect ... more a fast shoot in the dark cause I don't really use it that way 🙂

[unique shuttle]

looks the same

[languid pebble]

Starting Control Step is pretty high ... so I think the influence of the scribble will be pretty low

So it will start doing it's A.I. Mojo and work with the scribble at the end ...

[orchid schooner]

why did you turn off the preprocessor?

[languid pebble]

Because it's already a scribble?

But I am not a pro for this case. I'm just trying to use the knowledge I have in general. Sure there will be better solutions.

[unique shuttle]

Is there anything I need to get from civitai?

[languid pebble]

Still not working?

[unique shuttle]

Nope not even close to what you sent

[wispy nest]

Whats best inpaint model for anime

[languid pebble]

Have a look in youe Model/Controlnet Folder ... something in there? You might have to download them manually

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

https://stable-diffusion-art.com/controlnet-sdxl/ could be interesting. I need to sleep now. Good luck!

[wispy nest]

Anyone good at inpainting can help me

[crisp stream]

for inpainting I can very much recommend Easy Diffusion. It's got quite a pleasant Inpainting an also drawing interface (both can be combined). It's currently not being developed anymore, yet what is there already (including plug-ins) is quite sweet, yet I don't know yet, whether it's going to support SD3), otherwise a pretty good easy entry level interface with a pleasant overview of the generated images, sort of minimalistic focussed UI for the settings like sampler, seed,i2i, etc. leaving lots of space for the images themselves. If you are talking Comfy, haven't done inpainting there yet 😀

oh, and it's a 1-click install

https://easydiffusion.github.io/

[arctic laurel]

this one sorta has jon stewart energy for some reason

[arctic laurel]

i like this a lot

[crisp stream]

I quite like this one 🙂

[wispy nest]

ty!

[crisp stream]

yw

[gritty wyvern]

The grandeur of fantasy reigns supreme as three resplendent queens make a breathtaking entrance into the WWE arena. The first queen, donned in a regal green and blue ensemble, adorns the prestigious WWE belt. The Queen of Hearts captivates in her pink attire with the AEW belt, while the royal queen exudes elegance in her majestic costume, proudly wearing the Team Tag belt. Each exudes an air of dominance, ready to conquer the ring.

[crisp stream]

This is not the Image you requested.

[arctic laurel]

Sad text

[clever oar]

after the lake with piranhas

[frank wolf]

Ok real talk guys, do y'all really save your png files instead of jpg because of "higher image quality"? Honestly I cannot tell the difference. The only difference I see is the jpg being 1,70 MB and the png being 24,1 MB

1st is jpg, 2nd is png

Huh, strange, when I look back and forth on both files here I can clearly see a difference. But when I do the same on my photo program, I can't tell them apart The fuck

[nocturne oak]

Well, this isn't really the best image to show off a great level of detail since it just looks like a poor upscale of a smaller image.

Looks fine zoomed out, but then at 1:1, you can see a ton of moire-like pattern in both images.

Also, I don't think you have the workflow baked into the .jpg.

[frank wolf]

Point taken. I only save both file types when I upscale images, so I don't have the jpg of the original 1024x1024

Maybe somehwere in appdata, not sure

I'm actually quite new to this if you couldn't tell lol. Not sure what "workflow baked into the .jpg." means

[nocturne oak]

The workflow, aka settings, of the generating data for the image can be stored within the image itself.

[frank wolf]

If I'd take a guess its like baking in all the settings into it

right

[nocturne oak]

In any case, .png uses lossless compression and .jpg does not. You can output whatever you want; that's the beauty of doing it on your own machine with your own settings. If space is your concern, use .webp and don't embed the workflow. Then you'll get down to the smallest size without much compromise.

[frank wolf]

That's very helpful soul, I learn't something new! Thanks mate

I think I'll need to look more into how I upscale my images also

[nocturne oak]

Things for you to look into regarding upscaling:

• Generative vs Non-Generative
• Types of upscalers geared toward different use
• OpenModelDB (https://openmodeldb.info/)

[frank wolf]

I'll look right into it! Appreciate ya

[nocturne oak]

After all of that, check out SUPIR.

[frank wolf]

Never heard of it. Will do!

[unique shuttle]

Any know how I can turn this image into ai art?

[wide violet]

https://tenor.com/view/finding-nemo-darla-glass-tap-hello-gif-7964192757336767793

[silk rock]

oh what the yoshikage kira did I do wrong?

[vagrant dust]

Like this?

Works with controlnet line

[unique shuttle]

@vagrant dust
Yea though male but besides that how did you do that every thing I try I get something deformed or the exact same image I put in

[vagrant dust]

sorry was guessing the gender. 🙂
Simple Workflow with comfyui. Just need the preprocessors installed. Then load the controlnet for line art and use it with a higher strength (0.9 in this case).

Loaded your image, switched the latent size according your image proportation and added the prompt a woman with a leather jacket 🙂

[unique shuttle]

I’m using SD Forge how different are the two?

[wispy nest]

hi, where i can gener images from prompt? wich chanel?

[keen turtle]

#🏞｜general-with-images Woodcutter's Cottage

\

[wispy nest]

make

[vagrant dust]

sd forge got also an extension for controlnets as far as i read. so it should work the same. Load your image, extract lines with a preprocessor and apply the controlnet with line model

[languid pebble]

Your controlnet starting point should be 0

[vagrant dust]

Mine? i just tried to help. Discussion is now moved to tech-support as Deus Lupus needed Forge support

[languid pebble]

Yes ... sorry ... it's been the last mistake I've seen at his setup yesterday ^^

[vagrant dust]

oh this seems a long story 🙂

[languid pebble]

It was 🙂

[clever oar]

i try adetailer today-best extension in my collection

very help fix distored faces on distance

maybe no need on xl models

[languid pebble]

Looks like A1111 copied the bad UI from Swarm

[clever oar]

what is it swarm

[languid pebble]

New UI

Latest shit but I don't like ... not everything on your fingertip like SD Forge

[unique shuttle]

@languid pebble in forge are there base models you can't use?

[clever oar]

i think a1111 not bad

[languid pebble]

I don't think so ...

[clever oar]

ui not big important i want so that everything works

[digital hornet]

you cant use cosxl in forge and probably anything else really new like ella

[languid pebble]

Latest version you gave to extend stuff to see the settings you need like sampler. Not everything there where you need it from the start

[clever oar]

what you want see on start in a1111?

[unique shuttle]

i ask cause the one i had that stopped working was a SDXL base model and the other just didnt work and it was a pony base model

[digital hornet]

shouldnt have issues with those

[languid pebble]

Had to search for Karras first for example ...

[unique shuttle]

what would cause them to stop working?

[clever oar]

i have it

[digital hornet]

I dont know, it depends on in what way they werent working, what the error was etc

[languid pebble]

But you don't see it directly when you start. You first have to extend an option. Maybe you can save that but I had enough after first trial ...

[languid pebble]

Best idea here ... if you have problems: Don't tell them, show them

[clever oar]

is what i see on start

karras how you want

[languid pebble]

Looks like old version?

[clever oar]

1 7

[languid pebble]

Maybe I messed it up. It's getting a bit confusing:

[clever oar]

is automatic ?

[languid pebble]

Looks like it's SD Next ... whatever this is now again lol

[clever oar]

never see this ai 😃

[languid pebble]

But see what I dislike? You don't really see all the stuff you need. SD Swarm was the same ...

[clever oar]

but is not automatic)

[unique shuttle]

that is the error im getting when trying to use the base models

[clever oar]

but i understand what you want

[clever oar]

you can have broken model

[clever oar]

if you click this?

[languid pebble]

Sure that works. But I don't want to do that for everyday options ...

I think it's been A1111 -> SD Next -> SD Forge

[unique shuttle]

how do i go about fixing it (please don't say re-download its a long one)

[languid pebble]

I thought you have a lot of models already? Maybe just try another one first? If they work yours might be broken. I never used that pony stuff, sorry.

[clever oar]

only one way

or use not base

[languid pebble]

But respect! You don't give up so fast!

[unique shuttle]

yea i'm persistent. i've been up about 24 hours on this

[languid pebble]

Those who tell you it's easy have been working with it for months or the only do the easy prompting stuff ...

[unique shuttle]

i have specific visions so im not looking for easy

[languid pebble]

And you have seen what's possible. It's just pretty hard to guess someones problems here and to explain everything. Basics are ok ... but often there's so much more.

[unique shuttle]

ive noticed

[clever oar]

my experience - if strange errors change or reload the model

[languid pebble]

Quick'n'dirty for the fun ^^

[jovial tiger]

That's pretty neat

[languid pebble]

Some anis didn't came out like I wanted ... but didn't want to waste all the time 😄

But pretty cool we can do stuff like that today ...

[remote surge]

A digital illustration of a steampunk library with clockwork machines, 4k, detailed, trending in artstation, fantasy vivid colors

[clever oar]

in the end i dont want kiss this girl if i was this guy 😃

[languid pebble]

Her face becomes a bit long 🙂 At first I thought it's his nose ... but must be an animation problem

She looks a bit like the Red Queen 😄

Down with his head!!!

[clever oar]

she like cocaine 😄

[languid pebble]

I bet he closed his eyes when they started kissing!

[clever oar]

want some animation from kadinsky?😃

i make day ago

[clever temple]

what is that skull structure

[languid pebble]

Maybe another alien thingy?

[languid pebble]

The morphing stuff could be good if used short in music videos

[clever oar]

i want static objects

without waves

[languid pebble]

I think the only good available thing at the moment is video2video

Otherwise you just lose control the longer the video becoomes

[languid pebble]

Good nite!

[clever oar]

good nuts

[clever oar]

moofi i found for you art style

[austere meadow]

Can someone help me with some input. Would you say these three images are relatively consistent in like style and framing? It's for a visual novel. So these images would be placed side by side as sprites. But I don't want them to feel so obviously different.

[analog goblet]

some of the better pictures from this nights run, i hope you like some 🙂

[languid pebble]

Made me think of this even if I've done them some weeks ago 🙂

[hazy warren]

I'd jump in a pool of this

[languid pebble]

Looks a bit like candy 😄

[gusty lance]

A colorful birthday celebration with balloons, streamers, and a delicious cake. The cake has the words "Happy Birthday Sharvi" written on it in frosting.

[languid pebble]

https://discord.com/channels/1002292111942635562/1047610792226340935

[restive phoenix]

man in moon

[languid pebble]

Here's the picture you didn't ask for:

[gusty lance]

A majestic winged lion with the shimmering scales of a koi fish. Think powerful body and mane combined with vibrant colors and flowing fins.

[hazy warren]

now make it cyberpunk themed

[languid pebble]

It's more drug themed 😄

[hazy warren]

now make the Wizard cyberpunk themed

[restive phoenix]

man in moon

[vagrant dust]

here you go a man in moon

[deft bison]

interesting theme, the magician, this struggle between evil and good, this reality that we try to erase from our memories in favor of a sprawling, interesting, but soulless technology! ...

[hazy warren]

😮

[covert pagoda]

No! In the message above, #1047610792226340935

[covert pagoda]

No

[robust path]

SD3 soon?

[languid pebble]

Tiny house?

[clever oar]

yes i try controlnet lineart

[clever oar]

original

[languid pebble]

Cool!

[clever oar]

thx i think is good cn model

[languid pebble]

I'm not really good with CN ... I do my fights when I have to 🙂

[clever oar]

i try model for make video but is so bad result

[languid pebble]

I think we'll have to wait some more months ... and hopefully can do them local!

[clever oar]

second version

[languid pebble]

Dunno wich one is better 🙂

Have you tried Video2Video with ControlNet?

[clever oar]

no i try animatediff

and cant start

[languid pebble]

Every start is hard ...

[clever oar]

I agree, but there must be a limit to suffering, here the limit has exceeded what is permissible 😄

[languid pebble]

Take some times off ... the best ideas come while you do something complete different!

[clever oar]

you're right that's why I come up with interesting ideas after the toilet 😃

[pastel root]

https://tenor.com/view/skibidi-toilet-gif-2096158443957078234

[clever oar]

yes yes yes

[late sorrel]

what am i looking at :3

[unique shuttle]

@languid pebble i finally got it but but im trying to figure out how to use the seed i have to lighten up the skin tone a bit

[thorn shale]

Quién sera el próximo presidente de Colombia?

[tropic shell]

@languid pebble Please review the #✍🏼｜rules-and-tos regarding images. Thank you!

[languid pebble]

You are ... whatever ...

Thank you!

[pastel root]

[languid pebble]

Better don't show a nipple here ... or spam this place with nippled nipples ...

Could be prompt material ... nippled nipples ...

[unique shuttle]

how would i lighten the tone of his skin but keep this seed

[late sorrel]

men nipples? :3

[tall plinth]

A perfect circle

[tawdry fjord]

A perfect circle with crypto

[fluid quartz]

#🏞｜general-with-images Draw a beautiful girl with a pencil with blue skin with bitcoin badges instead of eyes with a hood in the form of a toad with fangs on her head without a background of thin nose black white beautiful lips on the hood a lot of fur picture is cold exclusively blue colorBackground, wild west, picture that says watch out, colors brown, white, yellow.

[crisp stream]

habe you tried Deforum?

[crisp stream]

thank you 🙂

[crisp stream]

Photoshop 😄

[calm imp]

an airplane with one base and two wings. The fuselage is the base, representing the smart city. The two wings are industrial intelligence and intelligent network connection. There is a runway below, which is divided into three stages. From far to near, they are solutions and soft platforms. ,hard terminal

#🏞｜general-with-images an airplane with one base and two wings. The fuselage is the base, representing the smart city. The two wings are industrial intelligence and intelligent network connection. There is a runway below, which is divided into three stages. From far to near, they are solutions and soft platforms. ,hard terminal

[crisp stream]

This is is not the Image you requested.

[calm imp]

#🏞｜general-with-images an airplane with one base and two wings. The fuselage is the base, representing the smart city. The two wings are industrial intelligence and intelligent network connection. There is a runway below, which is divided into three stages. From far to near, they are solutions and soft platforms. ,hard termina,8k,high quality

[crisp stream]

This is not the Image you requested.

[late sorrel]

kek

[clever oar]

maybe i try

[potent onyx]

Equestrian and aviation campus logo

[faint silo]

Here is the image you requested.

[crisp stream]

This is not the Image you requested.

[late sorrel]

some abstract art eh?

[late sorrel]

nice bird

[late sorrel]

i mean... that is certainly an image that exists

[crisp stream]

yeah, that´s actually a photo

[ivory glacier]

guys, using AUTO1111 it seems i can't get any model on any sampler to draw 2 faces with opened eyes. when 1 face in the image the eyes are great, but with 2+ faces all the eyes are closed or deformed. how can i fix this please ?

[unique condor]

It's probably the model or lora you're using then. Also, diffusion usually has a hard time with more than one person in a scene. Just inpaint the broken stuff once you find a composition you like

[ivory glacier]

hard to inpaint. i'm testing the prompts so i will use them into deforum

the same result i'm getting via deforum

[unique condor]

Inpainting is easy, and just use differential diffusion in comfyui, it turns any model into an in paint model and is god tier at seamlessly blending it back in. Just make sure to blur the mask so it has gradients

[ivory glacier]

i'm not using comfyui anyway. not familiar with it

Thanks though!

[unique condor]

Well it's a lot easier to use than it looks, there are a ton of good YouTube channels you can check out for tutorials

[late sorrel]

dont worry,, he will convert one day :3

[unique condor]

Oh and when you're trying to inpaint to correct the problems, you are making sure to change the seed, right? Otherwise, it will just fill it in with exactly the same thing that generated the image

If the seed number is the same

[nimble mason]

Here is the image you requested.

[vagrant dust]

still up for the equestrian aviation...

[jovial tiger]

I was responding to a reddit thread tonight where people were putting in prompt requests for sd3. the sheer volume of requests of directly conflicting stuff... people just love torturing the system.

[languid pebble]

Justin Bieber ... an angel on earth 😄

[late sorrel]

what is that between his legs?

[languid pebble]

Zipper?

[late sorrel]

yea but i mean that bulge lookin sus

[languid pebble]

Thought the same but prompt was pure

[late sorrel]

:3

[languid pebble]

Maybe ... it just ... fits to Justin? 😄

[eternal dome]

guys , what upscaler works best for real images ?

[unique condor]

Try all the major ones and pick the one that you like most. There really isn't a one size fits all upscaler and it really is a matter of trial and error

[eternal dome]

coool

[thin sable]

SD3?

[wide violet]

A man stands on the city wall

[vagrant dust]

here you go, an egg man standing on a wall

[deft bison]

a little mix here and there like for example: XL, biomechanical style, Ideogram 1.0, ...

[deft bison]

Thank you ! ...

[clever oar]

the strip between the legs evokes certain associations....😆

[languid pebble]

Will become a trend in 5 years ^^

[clever oar]

in Thailand? :3

[languid pebble]

Be nice or I'll send him to your place ^^

[clever oar]

i not scare with my friend

[languid pebble]

No brain ... not very helpful ^^

[clever oar]

he have ai brain

fast

[languid pebble]

Running on 4GB 😄

[clever oar]

128!

[languid pebble]

A.I. has become so crazy good ... I wonder how it will be with SD3

[languid pebble]

Because I haven't heard anybody asking today: SD3 when? 😄

[clever oar]

Do you think it will live up to expectations?

[languid pebble]

Maybe not at start ... but later with better workflows and usermodels ...

[clever oar]

I don’t expect special quality, I just want a picture without distortion of distant objects and so that the text can be understood

and follow my promt

and all it on 4 gb vram ^^

[languid pebble]

I think one of the benefits will be something we won't see on first view ... more complex prompted scenes ...

[clever oar]

my model can't even draw 3 bottles arranged by color 🙁

[languid pebble]

This will become easier

[clever oar]

sometimes I smile when I remember the crows bathing 😄

and when i translate it

[languid pebble]

Yes 🙂 I still have some SD1.4 or 1.5 prints here and love to redo some og the prompts

[rapid spear]

Love that third image

[languid pebble]

https://media.discordapp.net/attachments/989166678233464863/1178791361651167423/Its_magic.gif?ex=65776e01&is=6564f901&hm=7a08021cfaeac58d3a956589f3b275e27ddf21ceeea29df6bdaf01ed93ecdad7&

It's magic!

[clever oar]

have crows there?)

[languid pebble]

Not really 🙂 But a friend loves bats ...

[pastel root]

Do not trust strange women lying in ponds

[languid pebble]

Sometimes it's different ...

... than it looks like!

[inland gale]

style trasnfer

[languid pebble]

Messing around with Clownsharks Workflows leads to ....

[zenith bison]

#🏞｜general-with-images dog

[vagrant dust]

Here you go

[idle vale]

a dog in sea

#🏞｜general-with-images a dog in sea, happy, running

[vagrant dust]

here you go happy dog in sea

[clever oar]

promt head crab from half life 2

[steep sleet]

Some SDXL from ComfyUI

[nimble mason]

Hahaah nice eyes

[clever oar]

the liquid terminator took the wrong shape 😃

[clever oar]

thats your Workflows what you requested 😃

[languid pebble]

I have added this Perptubed thingy Node 😄

[languid pebble]

Wiggle wiggle wiggle ....

[languid pebble]

Who has stolen my eggs?

[jovial tiger]

hah neat. is that svd or animatediff?

[languid pebble]

SVD 🙂

[jovial tiger]

what were your motion bucket and augmentation level for that one?

[languid pebble]

Motion 135 cause I add a slowmo and uprender later, Augmentation I never really changed ... it's 0

[languid pebble]

You are looping?

[jovial tiger]

interpolater

gonna try a third stage of ksampled upscaling to see if it works intead of using upscale with model

[languid pebble]

Sometimes it works pretty good! the one video I had to look twice 🙂

[jovial tiger]

yeah the upscale actually looks words

worse

[languid pebble]

I'm using Topaz ... would like to try supir ^^

[jovial tiger]

yeah that probably helps

[languid pebble]

Old trick when we didn't have good upscalers yet 🙂

[jovial tiger]

finally got a good result

[languid pebble]

Haha ... cool one!

Playing with reality ...

[clever oar]

very nice

[late sorrel]

dem red cheeks :3

[languid pebble]

Worse things happen ... 😄

[late sorrel]

wait.. what even is this pic tho.. i mean the people and the front car are so small compared to the cars on the left LOL

[languid pebble]

Feeling old? Get a new one ...

[silk shoal]

guys please help

please give some ideas what this pic needs to be fixed

[late sorrel]

this kinda reminds me a little bit of Detroit Become Human game :3

[languid pebble]

Or Altered Carbon ^^

[late sorrel]

never seen that haha

[languid pebble]

Give it a try ... pretty clever

[nocturne oak]

well shit...that was several hours of hunting, trying, failing, rinse, repeat...

The new Proxmox 8.2 install comes packed with kernel 6.8, but DKMS is borked on it so the GPU passthrough made it look like the hardware was there, but Plex wouldn't hardware transcode. Ultimately, had to pin a 6.5 kernel for use to get the thing back up. I think I rebooted my node like 8 times today to fix this shit. But it is done and working now. PVE 8.2 running with hw transcode. No more manual maintenance for a while.

Well...none once I'm done updating the backup server, but that's far easier.

[cyan shoal]

gosh

can you show the options of the VAE

OH

width -> 1024
height -> 1024

[cyan shoal]

then just do this

change the resolution

yeah it's SDXL, you need to use 1024 instead of 512

and it should work well

you're welcome

you can also use aspect ratios like: 1334 x 768

1152 x 896

if you prefer the quality of juggernaut, then sure

base models are usually inferior

I use all sorts of models, but I would recommend juggernaut

which version of juggernaut do you have

oh yeah the normal X one