🏞｜general-with-images

[nova quail]

test image I did yesterday

[north garden]

hello

[languid pebble]

Wooden model ...

[languid pebble]

We forgot the fairies ... 😄

[candid surge]

fairies are forever, but sometimes they are in other places

[languid pebble]

Bed is waiting. Have a good time!

[candid surge]

thanks, night!

[nimble mason]

i call this work "The Concussion"

[heavy slate]

wow

[merry iron]

Thank you for using comcom analytics.
"comcom analytics" supports all community managers (moderators and server owners) by stats, visualization, and analytics.

If you have any questions, feel free to ask us!
Your dashboard
Help
Support server

Other languages
en: help
ja: help Japanese

[wispy nest]

I really love these wide sci fi images
what sort of model is this? thanks

[wispy nest]

kolors

kolors

[torpid wave]

A futuristic, abstract representation of a human brain fused with circuit boards and digital neurons, set against a deep space background. The brain should be partially translucent, revealing pulsing energy and data streams within. Incorporate vibrant, electric blue and purple hues to represent cognitive activity. Add subtle, glowing lines connecting various parts of the brain, symbolizing neural networks. The overall shape should resemble the letter "C" for CogniZone. The style should be sleek, high-tech, and slightly ethereal, conveying the concept of advanced artificial intelligence and cognitive computing.

[wispy nest]

SDXL but using a kolors image via IP adapter

[wispy nest]

SDXL but using a kolors image via InstantID node

it actually failed to replicate the face

which is the whole point of InstantID

however it transferred the style well

[clever oar]

you cant attack him from behind 😃

[languid pebble]

Have a good day!

[robust path]

Hello earthlings!

[languid pebble]

Hi 🙂

[robust path]

Hi 😊

[languid pebble]

Lunchtime here 🙂

[robust path]

Back to work here

[nocturne oak]

I'm using kolors for these. But it's my prompting that really brings out a specific look. 😉

[languid pebble]

Kolors is pretty nice! But some kinda prompts it can't handle ...

[nocturne oak]

Like every other model out there, essentially. That's why it's great that we can use what is appropriate for the thing we want at the time we want it. 🙂

[jovial tiger]

aura

[nimble mason]

bwahaha

[languid pebble]

Lord Lemonhead ...

[gentle blaze]

kolors

[stoic lark]

Guess who.

[eager wasp]

AuraFlow

[eager wasp]

Noticed no other model (sd3, kolors, pixart) could make sense of this prompt (tomato warrior creature eating a dragon that lies like a roasted chicken) and then with the ideogram discussion in mind, tried ideogram,it's close 😂

[stoic lark]

It's a bit sexy, but I finally got him wearing pants. Please remove if NSFW. This is developing at an okay pace. If anyone has any suggestions for how I can hike his pants up without photoshop, I'd appreciate it.

Trying to get Dream's hair right is killing me. He keeps coming out like Sideshow Bob or Goku

[eager wasp]

More AuraFlow (and so the waiting begins, model shows a ton of promise, when's the final out, now soon pls 😂 )

[wispy nest]

SDXL

kolors

SD3

[olive robin]

create a beatuiful girl

how can i create image

[dusk flume]

Hello, I am new here. Can someone please guide me on how to generate images in this channel using Stable Diffusion? Thank you!

[hearty violet]

hi, Omost is a extension for SD?

I install but dont found it

[dusk flume]

It's a bit sexy, but I finally got him wearing pants. Please remove if NSFW. This is developing at an okay pace. If anyone has any suggestions for how I can hike his pants up without photoshop, I'd appreciate it.

[celest sigil]

It's supposed to only be a 1024x1024 model, so far 🤣

[fleet goblet]

Hm.

What's the current best prompt-adhering model

Is it just any model using t5

[nimble mason]

it honestly might be aura, or close to it

sigma is pretty good too but aura is nuts

[fleet goblet]

Guess I'll try aura.

I've heard aura has an issue with a specific image

the nsfw filter cat image thing

[fleet goblet]

Do you have the link to the model?

[nimble mason]

nope

[fleet goblet]

rip

[jovial tiger]

the cool stuff comes out when you're refining the results to get even greater fidelity out of aura

[fleet goblet]

ope I can't even run auraflow

Lmao

[jovial tiger]

I think there's public demos on huggingface

[fleet goblet]

I have 16gb of vram

but the auraflow 0.1 safetensors itself is 16.1gb

[jovial tiger]

well if you can think of any prompts let me know and i'll run it

[fleet goblet]

Actually...

What is the designed resolution for this model @jovial tiger

is it 1024x1024?

[jovial tiger]

correct

[fleet goblet]

also what's the cfg settings etc

I need to find the settings lmao

[jovial tiger]

https://huggingface.co/fal/AuraFlow/tree/main

go there, and get the comfy_workflow.json

[fleet goblet]

ye

👍

thank you

Also, it seems I actually can run this model.

Intel arc can run auraflow.

How fast are you guys getting?

[jovial tiger]

it's kind of slow. at 30 steps with euler, it's 21 seconds on a 4090

[fleet goblet]

Well it's far slower on my card, 5.11s/it.

Lol

I mean it does work though.

I'm confused as to why it's this slow, though.

I'm going to look through the code.

[jovial tiger]

very cool

[fleet goblet]

The problem here isn't even the model

[fleet goblet]

It's the fact that they have the t5 encoder merged int he safetensors

[jovial tiger]

yeah definitely. they should outboard that.

[fleet goblet]

wait

21 seconds per it on your 4090?

or 21s per image

[jovial tiger]

per image

[fleet goblet]

ah ok

lol

jesus I got confused there

[nimble mason]

per 4090

[jovial tiger]

be good to your hardware.

[nimble mason]

only diff between the last two: perlin vs gaussian noise for stage C

[ivory fog]

how do i fix this >:

[unique condor]

Probably a wrong noise schedule, sampler or vae. Or option D) Not enough information

Show your settings and model, that will help narrow it down

[ivory fog]

hmm

[unique condor]

Like for instance, if you're using SD3, you can't use ancestral or SDE samplers or it can end up looking like this

[ivory fog]

its was talking about

NotImplementedError: No operator found for memory_efficient_attention_forward with inputs: query : shape=(1, 4096, 1, 512) (torch.float32) key : shape=(1, 4096, 1, 512) (torch.float32) value : shape=(1, 4096, 1, 512) (torch.float32) attn_bias : <class 'NoneType'> p : 0.0 decoderF is not supported because: max(query.shape[-1] != value.shape[-1]) > 128 device=cpu (supported: {'cuda'}) attn_bias type is <class 'NoneType'> flshattF@v2.3.6 is not supported because: max(query.shape[-1] != value.shape[-1]) > 256 device=cpu (supported: {'cuda'}) dtype=torch.float32 (supported: {torch.bfloat16, torch.float16}) tritonflashattF is not supported because: max(query.shape[-1] != value.shape[-1]) > 128 device=cpu (supported: {'cuda'}) dtype=torch.float32 (supported: {torch.bfloat16, torch.float16}) operator wasn't built - see python -m xformers.info for more info triton is not available cutlassF is not supported because: device=cpu (supported: {'cuda'}) smallkF is not supported because: max(query.shape[-1] != value.shape[-1]) > 32 device=cpu (supported: {'cuda'}) unsupported embed per head: 512
Time taken: 5 min. 48.3 sec.

like wut does that even mean

[unique condor]

That you're probably on windows, since triton doesn't work on it without using WSL(think of it as a virtual Linux install for windows)

[ivory fog]

hmm

so what should i do?

[deft grove]

im new to stable diffusion, how can i get started

[unique condor]

again, i can't help you if i have zero clue what you're even using or attempting to do. this is what would be option D) on a multiple choice test: not enough information.

[ivory fog]

well im using python and git?

[unique condor]

what is the app... what are the models... etc etc. i can tell it's some kind of gradio app. is it a1111? is it some random diffuser thing? you're still being far too vague

[ivory fog]

a1111

[unique condor]

okay, that's a start. did you follow the correct the installation from the guide? if you're on an Nvidia GPU, follow this guide: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-NVidia-GPUs. if it's some kind of plugin you're trying to use, you need to follow their guides as well. a lot of addons require their own dependencies and not all addons are compatible with windows

[ivory fog]

hold on gimmie a min

Found no NVIDIA drive on your system

uhh

im not that smart

but i think i dont gots one

it told me to look for one

aww sick dude

i think i fixed it

thanks @unique condor

[arctic cargo]

I'm looking for an AI developer to develop my project.
If u are interested, contact me.

[languid pebble]

Coffee! Good morning!

[languid pebble]

Hi MikersoftPop

[haughty roost]

So happy my thing works now. So basically the sd3 medium I downloaded a while back just is broken I guess. Don't know how others make it work but mine is just dead so I use this realism sdxl

Look at this thing

It looks like a real photo so crazy

[languid pebble]

Yes cool! I'm happy it works for you now!

And I think it makes more sense for you to use SDXL until there are finetuned SD3 Models

[haughty roost]

Yeah, just kinda confused on one thing. I saw someone mention they used sd3 ultra as a model but can't find that post and also it looks like that doesn't exist.

[languid pebble]

Maybe they where talking about the model you can use inside the artisan channels here ... I think it's a bigger version of SD3 they are still working on

[haughty roost]

Oh ok

Anyways, what's the place where I can find other models?

Or the lora things?

[languid pebble]

civitai.com

[haughty roost]

Also, how do I get the lora settings?

This workspace I installed didn't come with a gui for it

[languid pebble]

You are using ComfyUI?

[haughty roost]

Yeah

[languid pebble]

You will need a special workflow including a LoRa-Loader Node ... should be available at civitai, too

[haughty roost]

I just got it I think

You know how to link this and make the wires?

Figured it out sorry, just realized the colors mean something

But there is one problem

[languid pebble]

https://www.youtube.com/@sedetweiler has a damn good tutorial for comfy.

[wispy nest]

bird

[haughty roost]

Which of these are positive and negative, and why is there 2 model spots?

[languid pebble]

Left side input right side output

[haughty roost]

Ok I think I got it, I was trying to follow this image

[languid pebble]

Looks good!

So you are building your own workflows. I think it's way more clever than what I did when I started ...

[haughty roost]

I just wish I could search for loras without seeing this

[languid pebble]

You can filter them out ...

[haughty roost]

I just tried

I literally turned off 18+ mode and I still get this

[languid pebble]

No idea why people are so much interested in censored stuff ...

[haughty roost]

I'm filtering by most downloaded to hopefully get the good ones that are like realistic and what not, but I've seen so many NSFW ones including a ||mile high club flight attendant||

[languid pebble]

🤣

Talking about models it can be fun to use one for adult stuff for something complete different ...

AuraFlow

[unique condor]

no you can't, they aren't deemed nsfw, so the filter won't block them

i was joking about this the other night that i scrolled through like 200 rows with 5 columns, so 1000 loras, and only 30 of them didn't contain some female waifu image

[lavish garden]

Prompt: https://civitai.com/images/17967621

💜 "Mesmerizing" - Novakin (Starbound) 💜

[quaint meteor]

guys is there such thing as the ReActor but for short form videos?

[robust path]

Adding the following tags hides a lot of those
ANTHRO

FURRY

HENTAI

PONY

PONYXL

[lavish garden]

👀 these are the tags that appeared on my end

did you perhap mean the prompt?

[robust path]

No I added thes in my filters

[lavish garden]

ah 👍

[robust path]

hidden tags

[mental terrace]

tiny problem

missing dpm++ 2m karras, sde karras and such

how do i get them?

had it on previous pc, idk if i messed something up with my install or something

[austere crater]

What software do you use to create models ?

[mental terrace]

ah just trying to upscale stuff with SD1.5

[royal monolith]

go download them?

[languid pebble]

Is there a small software to read out my prompt from pictures created with comfy?

[royal monolith]

???

[languid pebble]

Pictures created with comfy include the workflow and prompts. I'd like to read out the prompt. Some programms I've tried don't wanna work with PNG, others tell me the metadata is in a wrong format ...

[royal monolith]

have you tried just uisng an EXIF viewer?

[languid pebble]

Some ... only one worked but very cryptic ...

[royal monolith]

okay then, as far as I know, you're stuck opening the images in comfy and getting the prompt that way. curious why you want to do that any other way?

[languid pebble]

Starting Comfy takes some time and sometimes I want to reuse a prompt in another workflow.

[royal monolith]

so if you're using it in another workflow, you have comfy running, right? i mean, i save my prompts in a spreadsheet as I work, but you could always load the image, copy stuff down, and go back to the other workflow or something

[languid pebble]

Hope for a better solution 🙂

[unique condor]

Wait, people actually have civitai accounts? Hell naw, I'm not making one lol

[languid pebble]

Got some Credits 🙂

[unique condor]

My point is, I'm not making some account just so I can set filters for what shows up

But I don't ever really get on civit unless I'm looking for something specific or click a link that takes me there

[languid pebble]

You can use it without ...

[unique condor]

I'll have to try because the endless walls of thousands of waifu thumbnails drives me insane

[mint prism]

"/imagine prompt: A delicate baby pink rose with soft petals, with a sparkling diamond emerging from its center, representing elegance and strength"

[gentle blaze]

kolors

[languid pebble]

Using Kolors at the moment, too

[gentle blaze]

I am really liking it. Try this custom sigma: 14.61482048034668, 9.277286529541016, 6.797628879547119, 5.1426873207092285, 3.7699766159057617, 2.6711347103118896, 1.7243311405181885, 0.9766815900802612, 0.42036789655685425, 0.12058357775211334, 0.01

[languid pebble]

I have no idea about custom Sigma 😄

[languid pebble]

Ohh... complete new Workflow. I'm still with the sample ...

[gentle blaze]

workflow kolors unet + ipadapter
// In this workflow, I am testing Euler. In the previous version, I was using DPM_2_Ancestral with CFG 1.8

[languid pebble]

Thank you! I'll try!

[clever oar]

🖐️

[languid pebble]

Good evening 🙂

[clever oar]

hello

[languid pebble]

Hope you've had a nice day!

[clever oar]

i was on lake😎

[languid pebble]

Sounds good!

[clever oar]

weekends are always good

[languid pebble]

Yeah!

[languid pebble]

https://youtu.be/7Itjx4w0t9U?si=g7qgeiHPAdqacrFu

[ripe cedar]

What's the best realistic inpainting model for skin and muscle?

[eager wasp]

(Auraflow gens)

[jovial tiger]

kolors

[unkempt spindle]

testing out a marble lora Im training

[eager wasp]

Kolors too (and got something of a refiner working in my poor little discord bot \o/)

SD3 2b

[jovial tiger]

auraflow/outpainting/denoised

[royal monolith]

but... the pasta? dirty shoes standing in it??

[jovial tiger]

5 second rule?

[royal monolith]

i don't think he's just been standing there for 5 seconds...

[dense brook]

hamburguer

[lucid parrot]

/ imagine um dragão voando sobre uma montanha ao pôr do sol

[rugged dagger]

@lucid parrot and cap, Stable Diffusion doesn't work like that. You have to download software or use a cloud based website like Leonardo or Playground.

[robust path]

I don't get it,i f you're dl'ing from them why not making one?
It just takes any email, no personal info

[languid pebble]

Yeah! Cool!

[gentle blaze]

He is trying it on... 😅

[languid pebble]

Human OS 😄

Loading ... ... ...

[wispy nest]

im guessing the way you get super high quality results is by inpainting?

whenever I do that it never seems to work out great, in comfy ui

even with 0.01 denoise it will make too much

im using comfy ui, if there is a very simple workflow i can take a look. I know you are supposed to use inpainting checkpoints but im not sure that I have one specifically desgined for that, just normal checkpoints

[royal monolith]

if I inpaint, i use photoshop's gen fill for that

[wispy nest]

checkpoint? is that on hugging face, github or civitai

oh wait photoshop

I just use gimp

[royal monolith]

get krita then, it's free, and get the stable diffusion plugin for it

[wispy nest]

ive heard of that, ill look into it

[royal monolith]

https://krita.org/en/

[wispy nest]

but isnt it possible to make an image in comfy?

or is it just inevitable that like 1 in 10 photos will have a lot of imperfections, due to nature of SD

[royal monolith]

certainly, i just don't do my inpainting there. i'll do that later, much later, down the line if i decide to use something i've created for a project and need it

[royal monolith]

quite frankly, 1 in 5 photos taken with cell phones or cameras by hobbiests and casual users have a lot of imperfections in them

and even quite a few taken by pros

[wispy nest]

yeah thats true, girls take 100 pictures and delte 90 of them lol

so the "crisp" results come from photoshop after image generation?

[royal monolith]

humans themselves don't look perfect. our faces aren't symmetrical for starters

[royal monolith]

no, you use the right models, you can generate that. you just have to work with the software long enough to know what you are doing. and what not to do

[wispy nest]

I dont mind imperfections but SD occasionally goes insane, im eager to see how it evolves

[royal monolith]

it's an AI - and the only thing an AI wants to do is nothing. so it'll go off down any rabbit hole it gets the chance to go off down

[wispy nest]

I do feel like im better but a long way to go. There is so many dials, switches, variables, takes a lot of experimentation.

This is something I made a few weeks ago, instead of using a 4x upscaler I just ran it through 8 ksampler checkpoints lol. I have moved to a different work flow now

this is my new workflow I think its better

but I dont really know how to outpaint yet

i started messing with svd_xt_1_1 but that is super touch and go

[lavish garden]

Prompt: https://civitai.com/images/17108726

💜 "Urban Wrangler" - Starlo (Undertale: Yellow) 💜

[wispy nest]

sci fi always welcome

[languid pebble]

Thanks! It's a remake of the first A.I. picture I've saved 🙂

[lucid parrot]

00020101021226850014br.gov.bcb.pix2563qrcode.brb.com.br/pix/cobranca/a71e6abd0e9b484d9d3a4c97e9040d955204000053039865802BR5914BRB-MOBILIDADE6008BRASILIA62070503***6304C34B

[clever oar]

witch😮

[eager wasp]

AuraFlow

[eager wasp]

A huge corpse, in dynamic view, In its eyesockets are bleeding glowing crystals., its forehead has a celtic cross carved into it, its chest wide open with an eldrich horror in the smokefilled chestcavern. With intricate, dark, sky and horror details, insane level of details, hyper realistic, sense of awe and scale, cinematic, composition, artstation HQ, by Zdzislaw Beksinski and Jan Urschel .

(yeah chestcavern, oops, turned out to work better than cavity after fixing, eldritch was worse too, lol, and those tags "artstation HQ, by Zdzislaw Beksinski and Jan Urschel " do close to nothing for AuraFlow)

[nimble mason]

Here is the image you requested.

[wispy nest]

okau

why delete

[tawdry pecan]

Because delete

[wispy nest]

i already saw it

[nimble mason]

Here is the image you requested.

[wispy nest]

okay lets see how I can do

[tawdry pecan]

Do i care? Hell no.

[wispy nest]

so you wont mind then! thats great

[slate jackal]

That’s the base album cover

This is the one I saw of starfire, but I was hoping for one of Raven and one of beast boy

[wispy nest]

eh

lot of work and involves trade marked characters

[nimble mason]

Here is the image you requested.

[slate jackal]

Dang man, I’m sorry. How much work do you think it’d be?

[wispy nest]

more work than an almond lol

i would probably see if they have a beastboy lora cuz it wont just make that naturally

im sure they do tho

right now I dont do much img2img personally, I need to create a workflow for it in comfyui

its easier to do in automatic1111

[slate jackal]

Ok. Do you know someone that could possibly do that? My parents haven’t gotten a new pc in ages lol

[wispy nest]

probably more than a few people who know how to do it

[nimble mason]

Please specify the subject and prompt you require.

[wispy nest]

img2img, he gave the images

https://civitai.com/search/models?sortBy=models_v9&query=beast boy

heres a model of what he wants

"beast boy"

[hearty violet]

its possible to me add maks to segmentation table? fore example if I wanna that it segment something that don't exist on this tale
https://docs.google.com/spreadsheets/d/1se8YEtb2detS7OuPE86fXGyD269pMycAWe2mtKUj2W8/edit?gid=0#gid=0

[wispy nest]

side question, im guessing AI art is popular on tiktok

[slate jackal]

Yeah lol. My girlfriend saw it and wanted to do that since me and her are big Ken Carson fans lol

[nimble mason]

Here is the image you requested.

[wispy nest]

not even close lol

honestly I almost feel like it might be easier to do with photoshop given its mostly a black and white photo

[slate jackal]

I didn’t think I’d be going through this much to get these for us 😭

[slate jackal]

Dawg 😭

[eager wasp]

More beast boys for all 😆

[wispy nest]

ai art is not that easy!

[slate jackal]

I didn’t know man, I’ve only known just the tell chat gpt and it does it

[royal monolith]

if you're going to do that, then you're actually using dall-e. if you're going to use dall-e, then use microsoft's version for free instead of paying openAI to use it

[slate jackal]

Could that actually get me closer to what I’m looking for?

[royal monolith]

i don't know what you're looking for, but i have a thing against paying for something you can get for free

[slate jackal]

Is it ok if I show you?

[royal monolith]

sure. you're welcome to DM me if you like

[slate jackal]

Ok, I’ll send you a dm

[slate jackal]

That’s from the exact guy that made it I think

[pallid ruin]

Probably, I still don´t know how he made it ... F

[wispy nest]

thats amazing what checkpoint and prompt did u use to make

[wispy nest]

oh why hes peretrating

[pallid ruin]

Its not my image, I saw it online, im still don´t know if they used controlnet or what, tomorrow i´ll search a bit more

[languid pebble]

Let's have a good morning coffee 🙂

I shouldn't prompt stuff I have no idea how it looks in reallife ^^

[nocturne oasis]

Pls can someone help me out here

I need to run and edit on a picture
How do I go about it

[wispy nest]

canny edge and depth map control net
a few ip adapters
and then inpaint the changes

[nocturne oasis]

Please I don't really understand

[wispy nest]

how many of the things in my reply did you understand?
because you could try just one of them

[nocturne oasis]

Nothing at all

This the picture I want to edit

[nocturne oasis]

I want to edit it to be in a space background and the monkey put on a space helmet

[wispy nest]

okay
ignore control net and ipadapters for now
download Fooocus and try inpainting there
https://github.com/lllyasviel/Fooocus

[nocturne oasis]

Ohhh okay 🥲
Thank you very much 😊

[nocturne oasis]

But I don't really know how to run an edit

[wispy nest]

if you are lucky it will work

if it doesn't work there are more things you can add later

IP adapter can be a bit like "style transfer" it transfers the style of an image

Control net is like, you upload a drawing of what you want and it helps the model

[wispy nest]

I don't use fooocus a lot personally so maybe wait for the others to wake up
there will definitely be someone here who uses fooocus as it is very popular

I personally use comfy ui SEGs-detailer node for inpainting, which is too complex for you at the moment

if I remember rightly though on fooocus you just upload the image into the inpainting tab
it should be relatively clear

[celest sigil]

Here's a start for you

[stoic hemlock]

hi

[languid pebble]

SD3?

[eager wasp]

AuraFlow, tried it before with sd3, it bled the concepts dog mask / cat

[languid pebble]

Nice work ... Would have to downgrade Torch and XFormers to make it work ... last picture took me 30 Minutes 🙂

[eager wasp]

I'm not using it locally either. I'm a proud AMD GPU owner 😵‍💫

[languid pebble]

Awww... 4090 here ... but software issues 🙂 It would be boring if it's to easy. Finished my lunch ... have a good time!

[sterile kiln]

Hey guys, anyone know when the new SD3 model promised in the last announcement is coming?

[nocturne oasis]

Ok

[nocturne oasis]

Thank you Very much 👍

[nocturne oasis]

Wow thank you 🙏
Pls can you help change the colour of the monkey aswell and wear it a space helmet
And maybe increase the picture background 🙏

[celest sigil]

No

[nocturne oasis]

🥲

[nocturne oasis]

Human or machine?

[sharp pier]

Guys, I'm a newbie, tell me pls, why don't I have these parameters (Controlnet) in my interface? My interface is in the second img

[hearty violet]

where are you running it? controlnet is a extension if you don't have it you have to install

[fossil charm]

facedetailer thought hair was an face lol

[wispy nest]

lol

[jovial tiger]

kolors

[languid pebble]

Lumina

Kolors

[wispy nest]

lumina is another good looking one

[languid pebble]

Yes 🙂

Lumina ... and it works with a higher resolution quite fast ...

1536x1024

Ohhh... took the same one 😄

Spice Storage 🙂

[wispy nest]

yeah the higher resolution abilities of lumina are really nice

[eager wasp]

After the cat #🏞｜general-with-images message, had to see where the dogs went 🐶

[languid pebble]

Working early to prepare for XMas

[jovial tiger]

do not taunt reindeer bot.

[languid pebble]

: ) :

Schrödingers Smiley

[frozen hull]

Hello, everyone. I am very new to stable diffusion. I really hope you could help me.
There is an app called AI Morph (the one that was made by Daily Joy studio). (It's not and add, cause this app sucks)It was great, until now. This app lets you transform any picture in anime style. It was the best amount the other, accurate in everything and detailed. Very beautiful. But now it doesn't support NSFW. When i put some NSFW there it says "The result may not meet our guidelines. Please try something different." Can you please tell me, if apps like this are using something like stable diffusion or are they creating their own ai? Cause their pictures was the best in everything. And now I want to understand can I make something like this in Stable Diffusion?

[eager wasp]

Back to kolors, kolors is easy mode for coherent images

[hearty violet]

wich is better, a 5.1 HYPER or 6.1 (VAE) B1?

[hearty violet]

very beautiful, do you use upscalling or Adetailer?

[clever oar]

🖐️

[hearty violet]

a extension .yaml is usefull or must be .pth?

[hearty violet]

man I thought that T2I-adapters are only to color grid, PID and CLIP-vision, what are those all contronelts again?

[languid pebble]

No ... nothing yet 🙂 But I wanna include later 🙂

[languid pebble]

There's a new All-In-One Model ...only downloaded it cause I don't use ControlNet that often.

[languid pebble]

Für SDXL -> https://huggingface.co/xinsir/controlnet-union-sdxl-1.0

[celest sigil]

There's not actually any way to control this controlnet yet. It uses new architecture, so needs to be written into the existing UI's

[languid pebble]

https://www.youtube.com/watch?v=UBFEw1IUX_I ?

[celest sigil]

That preprocessor node isn't related to the Union model though. That create a controlnet image, but you need to be able to select the correct image type from the union model.

At the moment it just defaults to one type every time.

The method for selecting is explained in this post #✨｜sdxl message

However, that's not implemented anywhere yet.

I'd happily be proven wrong though! 😄

[languid pebble]

Thanks! Didn't try on my own yet ... just remembered that video an me downloading it. Looks like the Vloggers are faster than reality sometimes 😄

[celest sigil]

It does work, to some degree, but not as well as it should do...yet

Better off using the correct models for the type until it's ready.

Multi controlnet from a single model is most interesting. That'll hopefully save some resources 🙂

[languid pebble]

Some clever people will have a solution soon 🙂

[hearty violet]

I know, thanks! But it don't exist for sd 1.5?

[languid pebble]

AFAIK not ... you might not need all modelds. Depends on the way you wanna use ControlNet ... maybe just download one model if needed?

[hearty violet]

do you use promax or normal?

yes, but its very very helpfull if existe controlnet union for sd1.5

@languid pebble

[hearty violet]

coffeeee

[hearty violet]

@languid pebble controlnet union works : D

with SDXL, sure

well, my final image don't work, my prompt wa "A modern living room"

I will try more

[hearty violet]

• why my juggernaut sdxl ever result a image with this noise aspect? the example images on civit ai are smooth

• what is the problem here? I2T with style clip vision

oh, the image reference is not there, must be it

no, same message =/

[robust path]

Pixart > Mobius i2i, upscaled with Upscayl

[wispy nest]

I don't know Mobius, it looks good though

[robust path]

It's one of my favourites
https://civitai.com/models/490622/mobius?modelVersionId=545562

[wispy nest]

ah yeah it looks good

for SDXL I always use jugger

[nimble mason]

yeah mobius is outstanding, def recommend it

[wispy nest]

at least from the civit previews, Leosam Hello World always looked good to me

[clever oar]

I2T open pose better than default open pose?

👋

[languid pebble]

Congrats! Yes ControlNet already exists a while. A pretty cool tool. But not really for me cause I don't like the idea to work with something already existing.

I have to check Mobius, too! Thanks community!

[celest sigil]

It'll be very difficult working with anything that doesn't exist already. 🤔

[languid pebble]

Not for you! Not for you! 😄

[wispy nest]

hopefully this new super controlnet will be even better for multiple types of conditioning at once

cos they trained it with more than one applied at the same time

[hearty violet]

I can use Omost only in hugginface? Can't be used with SD?

[clever oar]

help

i dont see all t2i preprocessors

[hearty violet]

I think is only it

for 1.5 or xl?

but you can take then here
https://github.com/TencentARC/T2I-Adapter

[clever oar]

1.5

how you think is good thing or i can replace this

youtube said that t2i openpose better

[hearty violet]

well I started study t2i yestarday lol

I'm trayng understand why exist some preprocessos like openpose that already exist on controlnet

I'm using only style clipvision and color grid for now

[clever oar]

advise what cool things can be done with it

[hearty violet]

sure

one minute

[clever oar]

thx

for answer

[hearty violet]

https://stable-diffusion-art.com/controlnet/

here are a section about T2I is where I'm studyng

and a lot of other cool things

with examples

[clever oar]

oh i want try too

what diffrent between colors and lineart

[hearty violet]

colors?

[wispy nest]

it cant do colors

ip adapter can though

what you are seeing are depth maps

and normals

[hearty violet]

colors is a controlnet?

[clever oar]

yep

[hearty violet]

can you print it, pls?

[clever oar]

color grid

i mean

[hearty violet]

colorgrid you sintetize the porportion of colors of a image in squares and create a new image rearrange this square according your prompt

make sense?

[hearty violet]

like this

but dont work on mine, dont create a new image XD

[clever oar]

👍

color grid test 😃

[hearty violet]

how you do it works?

[clever oar]

doesn't work for you?

[hearty violet]

no

mine

[clever oar]

click on right image

in result

it work fine i think

[hearty violet]

what?

where is result?

ah, ok

the map is correct but my new image dont

I asked for a contemporary house

[clever oar]

not how i want

dont understand it works or not

[hearty violet]

color grid make a new image based on the color proportion of the image reference you use have

what exaclty you wanna do?

[clever oar]

it works i see

not how i imagine

huh lineart do anime to realism

[hearty violet]

what exactly is lineart used? why use it instead canny or scribble?

you are doing great images

[clever oar]

thx

[clever oar]

i dont know but i like lineart effect 😎

[hearty violet]

well I will try more ^^

[clever oar]

i use canny beffore but lineart for me better or for different tasks

[hearty violet]

I dont understand the propose.... it's not clear on my head

sometimes the differences are small for me

good lol what was the prompt?

[clever oar]

neon light

[hearty violet]

nice

[clever oar]

thx

i will try lightning strike

oh not good how before 😎

[languid pebble]

Trying to upscale some old pictures 🙂

[hearty violet]

I think I understand better about Omost 🙂

this is to teach us how construct a prompt to sd? to inspire?

[languid pebble]

It's adding stuff, LLMs think we would like to have cause statistics say they fit ... that's how I understand it ...

[hearty violet]

but I can copy and paste this promp on SD?

[languid pebble]

Sure ... Leonardo has something like that ...

[hearty violet]

really? cool, looks a lot of things but if it works, great. I will try some time ^^

[languid pebble]

But might be a better idea to chat with an LLM like ChatGPT about the picture you wanna have and ask to create a prompt later

I think you can do it with Bing including creating the picture for example ...

[hearty violet]

I think no, Omost must be better approppriate

because was create by Illyasviel to it, chatgpt is a instruct LLM

my opinion

look these prompt, it perfect

[languid pebble]

Omost only has your prompt and adds stuff ... Bing can ask you for specific more information and not just add them by checking statistics ...

[hearty violet]

I get your point

I like Omost, will try

but maybe use booth? can works too

[languid pebble]

Omost has several functions ... I'd prefer to have a prompt really fitting to what I want ... and not what people ask for often ...

[hearty violet]

well, for me I see better how construct a prompt by attention

help me

[languid pebble]

For example you can say ChatGPT: "You are my prompt creator. I'll tell you a topic and you will ask for details till you are ready to create a prompt for SDXL"

Haven't tried that yet ... but should work ....

So you will learn with every question from chatGPT what you should have added to your first idea ...

[hearty violet]

yes I understand

you can dialogue with chatgpt to construct your prompt, is a good idea

[languid pebble]

My idea is always to get the picture you already have in your head ... and not a random cool one ...

Midjourney is giving you more random but often cool stuff ...

Sometimes it's even more interesting to try to get what the A.I. don't really want to give us ...

[nimble mason]

trained a model to add details to sdxl latents leveraging its vae and the 16 channel VAE

[hearty violet]

I like it, dont change the form only better the image, add more detailed how you said

[nocturne oak]

prolly garbo, really...
who are you even?

[hearty violet]

how can I get it? 🤗

[languid pebble]

You don't know? Nothing more to say ...

[nimble mason]

not yet is the answer... this is the prototype

proof of concept works pretty damn well

[nocturne oak]

I bet he doesn't even know how to triforce.

[nimble mason]

so now the next task is gonna be gathering up a mountain of images of every style imaginable lol

[languid pebble]

Aren't you a moderator here?

[nocturne oak]

No. But I play one on TV.

[languid pebble]

Sun always shine on TV ....

[nocturne oak]

Also, I know who CSBW is. I was giving him shit.

[languid pebble]

I think he knows more than me ... I'm just trying not to completly miss the train ... and share ideas ...

[nocturne oak]

If you really want to have a good time, ask him about math and sigmas. 🤣

[languid pebble]

A.I. has splitted into so many directions ... nobody can get them all ... Hazard has been asking for ControlNet ... I've had some youtube input but it's not really my thing ...

I think you are kidding me 😛

[languid pebble]

Sun always shine on TV ... #

[clever oar]

i dont remember i can show company name?

[royal monolith]

got their permission to use their traemark?

[clever oar]

ask about rule

discord group

[wide grove]

When trying outpainting using ComfyUI, it just doesn't work like I'd expect. Here is the result. I do have a 2nd KSampler, but I guess I'm doing something wrong.
What other information should I provide?

[jovial tiger]

you need a denoise setting on your secondary ksampler. so swap in another regular ksampler with a denoise of 0.7 or higher. could try all the way up to 1, but you'll probably start seeing stuff happen above 0.7

[wide grove]

swap in the regular one instead of the advanced?

The only reason why I had the advanced was to do "start at step", but I guess it's not relevant in that context

[gleaming hinge]

Can someone suggest me model, styles, etc. to recreate this ai image?

[celest sigil]

https://tenor.com/view/shut-up-and-take-my-money-futurama-fry-take-my-money-money-gif-15195954

[clever oar]

uberrealistic porn merge

[celest sigil]

Less precise recreation

[lavish garden]

Prompt: https://civitai.com/images/20194470

"Serenity" - Biker ('The Wolf' - SIAMÉS)

[steady loom]

Cartoon style, cute golden retriever, bright colors, , made by yukisakura, awesome full color, picture book style, illustration

[quiet current]

Friendly reminder to keep it classy with the imagery. That's all.#✍🏼｜rules-and-tos, rule 3 (No NSFW Content)

[celest sigil]

Should I remove my last posted images? I thought they'd be ok if she was dressed.

[quiet current]

Let's say it's very much on the line. In my book it's still ok but I prefer to gently remind people about rules whenever I see a slippery slope.

[celest sigil]

Noted, thanks 🙂

It was borderline for me too, but nothing was said about the original post so I thought it must be ok.

[quiet current]

Mods are not robots (yet 😜) sometimes stuff goes under the radar. Or they have hard time deciding when to pull the trigger. I'm pretty sure this is one such case.

[robust path]

We need a bad taste filter

[hearty violet]

ufing Omost to generate the prompt, "psychology clinic with a bat cave concept"

[languid pebble]

But do you want a bat cave or a batman cave?

[hearty violet]

batman cave, thanks XD

here in brazil we say bat cave

[languid pebble]

Yes ... but not sure whether A.I. understands it as wanted. you could also try to add the movie into the prompt ...

[hearty violet]

nice ^^

[languid pebble]

A.I. often knows movies pretty well ...

[hearty violet]

this one 😎

based

@languid pebble Omost is a LLM you can talk with it

like chatpgt

I asked to change the batman cave to a pokemon paradise, let see

[languid pebble]

Cool! I saw the instructions part later ...

[languid pebble]

😄

[hearty violet]

time to wok!

[languid pebble]

Better don't try Spongebob 😄

[hearty violet]

😮

must be cool huehuehuehue

[languid pebble]

This become ... strange ... 😄

[languid pebble]

Hi dicordos 🙂

[languid pebble]

Reminds me of the "Hello Moto" campaign ... 🙂

[clever oar]

good morning coffee

[slate owl]

can you guys help me create this type of semi realistic art?

[wispy nest]

this is dreamshaper style

and put "render" in the prompt

sometimes "unrealengine"

[slate owl]

Thanks. ill try in that.

[wispy nest]

wow

[wispy nest]

just bare in mind those two examples you posted
especially the purple one
have had detail boosting

either tiled upscale or noise injection

or something like that

[slate owl]

oh i didnt know that , sure

can I find any tutorials online ?

[languid pebble]

You can search at civitai.com for pictures you like and look at the prompt and used model and settings ...

[languid pebble]

My prompt has been: "a wallpaper hd quality Naruto Wearing a Pan Coat Rich Personality Name Faiez, photo, illustration, cinematic, typography, 3d render, unrealengine, poster, fashion, anime, product" Model: Kolors

[slate owl]

ill try similar for gojo

[hearty violet]

sassukeeeeeeee!!!

[languid pebble]

I don't really know Gojo 🙂

[drifting quest]

:

[clever oar]

mojo mojo😃

[clever oar]

is real device

[languid pebble]

We do have a Mojo Club in Hamburg 🙂

[clever oar]

you have personal language!

[languid pebble]

And Austin Powers lost me once and didn't like it ... there's a whole movie about that 😄

[clever oar]

how many things you have😃

[languid pebble]

But this name is pretty old ...

[clever oar]

what is g..y?

[languid pebble]

A place in Secondlife to meet people ... I don't really remember it ...

[clever oar]

meet with gys?

😮

[languid pebble]

sure ...

[hearty violet]

I found Omost to comfy, dont exist it for A11?
https://github.com/huchenlei/ComfyUI_omost

[languid pebble]

Comfy often first to get new technologies. And not sure whether all new technologies will work for A1111 or SD Forge ...

[hearty violet]

realeasea new forge did you see it?

[languid pebble]

Ohh.... I have to check!

[hearty violet]

https://github.com/Panchovix/stable-diffusion-webui-reForge

[languid pebble]

Thanks!

[hearty violet]

you are welcome ^^

[gleaming hinge]

i would like to generate images with similar art style or model. I've have try some models and style but couldn't get results similar to this or near this.

[steady cradle]

重新生成

[wispy nest]

the lace design is amazing

[buoyant plaza]

Кассир из пятерочки в депрессии, работает с утра до ночи, устала очень, мало зарабатывает

[wispy nest]

ыы

[jovial tiger]

auraflow / kolors

he's eating it apparently.

[summer viper]

hhh

wait

[languid pebble]

Flowers ...

[gentle blaze]

best sound

[hearty violet]

why when I use TI2 with CLIP-apapter give me this error message?
RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 2048 but got size 768 for tensor number 1 in the list.
Time taken: 0.5 sec.

[balmy jacinth]

Flower

How to get an image with commands?

[fossil mason]

The Last.fm user hugetoad has no listening history on their profile yet.

Just created your Last.fm account? Make sure you set it to track your music app.

Using Spotify? You can link that here. This can take a few minutes to start working.

Please note that .fmbot is not affiliated with Last.fm or Spotify.

[balmy jacinth]

message hello ¯_(ツ)_/¯

[balmy jacinth]

message how to generate an image with commands? ¯_(ツ)_/¯

message could you help? ¯_(ツ)_/¯

[hearty violet]

hi, what do you need?

[jovial tiger]

witness what happens when the cat storm hits

[summer viper]

angry cat

[jovial tiger]

he better get his chocolate cake

[sinful sage]

where can we get a free chatgpt4.0

[royal monolith]

you can't. but you can use a free opensource LLM if you want

[languid pebble]

Good morning Capucino 🙂

[stoic lark]

Been working on this for f* months. Censored, spoilered, don't ban me.

Not done yet. Not even close.

Open in browser. Isss bigg.

Reason I post it, I can't for the life of me prompt the elements to the left or BG for clarity. They just get more indistinct. I could use some advice.

[stoic lark]

Well, took almost two hours, but this is a composite I made from 20 different gens that I guess approximates the guy in the top left.

[languid pebble]

Only fools say it's easy ...

[celest sigil]

Any better?

Another attempt

[haughty roost]

It's kinda slow cause I'm on the base model for the m1 macbook air 13 inch, but it's still awesome and I get great results with this. I'm trying out drawthings for the first time and it's actually my favorite thing ever. It has a great ui in my opinion, let's you download models, lora, and other things straight from the app, and then also it preselects settings which I've always needed as I can't ever find good ones no matter what. These two are made with sdxl turbo as well as I guess the other things that the app adds onto the images

One thing though, I did have to use firefly ai for the first one on that middle deer as it originally was made without a head, but the body still looked impressive for a "turbo" model, which I'd expect to not produce the most quality results if it's aim is to rush

[atomic jay]

made this, quite like the texture

[hearty violet]

gemini is free and his API

[stoic lark]

Prompt? I'm taking it in tiles for accuracy and detail.

[unkempt spindle]

trying to make a LoRA that apply marble texture to objects, well first version was..lol

[languid pebble]

Learning by doing ... my first one was unusable 🙂 So we all feel with you ...

[unkempt spindle]

haha, yeah. But atleast it is improving on version5 atm

[languid pebble]

A mistake is the best thing we can do to learn ... see it's working!

[unkempt spindle]

totally agree👍

[languid pebble]

I remember rendering a video 12 hours using a wrong VAE 😄

[unkempt spindle]

oh god..

[languid pebble]

German?

[unkempt spindle]

swedish 😄

[languid pebble]

Cool! Zicke sounds so German 🙂

[unkempt spindle]

hehe hellsicke is slang for "darn it, dang it" in swedish

[languid pebble]

Kinda bitch in German 😄

[unkempt spindle]

hahaha

[languid pebble]

I'm living in an IKEA Store ... pretty hard at day but the nights are calm and it's for free 😄

[wispy nest]

wow even a little bit of movement adds a lot

[languid pebble]

Only SVD but still fun!

[wispy nest]

ye its still so much better than a static image

[vagrant dust]

May i ask why you choose to create a lora instead of using the ipadapter for style and composition ?

[royal monolith]

faster

[unkempt spindle]

It's good to learn this way, what works and what not. I recently went from kohya to onetrainer and I wanted to test out when you combine different concepts. Im gonna go comfyui and ipadapter, video and all that stuff later 🙂

[vagrant dust]

I see because the image above took only 20 seconds after i saw your post, so you are talking about learning speed and sort order of different topics.

[unkempt spindle]

when the lora is done, Im hoping I can do 20 images of different marble subjects in those 20 seconds 😉

[wispy nest]

anything like IP adapter, control net or text prompt CFG can degrade your image quality
so lora is good because it lets you use less of the above

[celest sigil]

It was a workflow that created it, not a prompt.

[unkempt spindle]

I assume ipadapter is good when you want marble on stuff the LoRA dont have a friggen clue where to put it on, I know it will be close to impossible to have marble texture on every item in existence, especially if they are not in the dataset 😄

unless you use inpaint then ofc

but this LoRA is alsa about training a certain type of texture of the marble, black veins, grey veins, gold veins, glowing veins etc

[vagrant dust]

It is more or less if you just need a fast combination of styles and compositions, or the integration of multiple images into one with nice transistions etc.
I would create a lora for a specific style of images i would generate with the ai.

[unkempt spindle]

but this is a specific style, its not about marble cars 😄

[wispy nest]

in general training, lora or checkpoint, is way more powerful.

[vagrant dust]

Cars with different Marble coating. Ok the blue one is created from a lightning image ...

[unkempt spindle]

this is from the lastest trained version and starting to look what Im going for

[stoic lark]

Do tell, my good man. I'll fire up the comfy and finally start learninating it.

[celest sigil]

It's one of the ones available on my Patreon, "Creator's Lounge" tier.

[languid pebble]

SVD Music Video ... https://www.youtube.com/watch?v=vxLnqgTCloY wouldn't do it the same way nowadays ...

[unkempt spindle]

lol catgrass dealer 😄

[languid pebble]

There's an english version, too. Catnip Dealer ... but without video 🙂 To much work ...