🏞｜general-with-images

[steep sleet]

I have 1193 in my Lightroom folder - mostly SD3@ClipDrop - 10 prompts/day - 4 photos/prompt = 1200 photos/month - and all for $10!

You may also use FABLAN's free SD3 implementation @Glif

A Gremlin out of ComfyUI

I gotta headache!!!

Pinup hottie!

Detail!

Detail too!

[minor fractal]

how

how did lykon get access to s.ai's secrets

[soft comet]

is that true. most people cant run sdxl

[minor fractal]

do you have gpu? if so, is it recent? if so? you can run it

[soft comet]

dam, guess im the outlier then

3090

[minor fractal]

you can!

even the humble 1660 can die trying

[soft comet]

i know, i was just saying im the outlier cause most people dont have gpu good enought to run sdxl

i know a 3090 can run sdxl

[minor fractal]

most people can do it

[soft comet]

dam

[minor fractal]

1080?

[soft comet]

i one got sd 1.5 to run locally on a chromebook

first you install windows 10

[minor fractal]

wrong

[soft comet]

then optimise the hell outta a1111 and wait 10 minuits

wdym

[minor fractal]

use linux!

and comfyui!!

[soft comet]

yes but im not that cracked

[minor fractal]

eh fair enough

[soft comet]

yes

i once got gmod running on a chromebook it played at a steady 40 fps

kinda cursed

[nimble mason]

wavelet sampler

[crisp stream]

don´t really know, so I couldn´t tell 🙂

[jagged badger]

what do I call these? is there any term for that kind of half-abstract imagery?

[unique condor]

look kinda like abstract concept art to me

really cool though

[analog goblet]

any news since the last few weeks ?

images kinda still look the same sadly

[autumn fern]

the prompt was news anchor

this is how bad it is

and when I tried img2img

the input was

and the output was

[analog goblet]

you didnt show your a1111 settings

[autumn fern]

Umm can you tell where to check that, is it web-user.bat

[analog goblet]

no, the browser thing, where you put your prompt and the model and so on

[autumn fern]

oh, one second mate

[analog goblet]

try 50 steps not 20

[autumn fern]

is there a problem with the model

Hey I think it's working well for long prompts

[analog goblet]

go experiment aroudn some, doesnt seem totally wrong what you do there

[autumn fern]

I'll try that, but AUTOMATIC1111 page had better results ig it's more about the model's, other than that is comfyUI better

[quiet current]

@autumn fern 20 steps should be enough with dpm sampler

but yeah "anchor news" is too vague of a prompt

also you re using the very old / outdated sd 1.5 model.

try to get community made ones, sdxl, etc

[autumn fern]

Oh so that was the problem?

Ig i'll try sdxl and all the other models

[winter frost]

hiuj

[grizzled sage]

Omost is very bare bones, but the framework is provable. What a cool little system

[strong mango]

Kids recycling cartoon

[frank dock]

#1237459938901491852 #prompt A man is leaving a luxury brand store with many shopping bags in his hands.

[languid pebble]

Go into that channel 🙂

[wispy nest]

OH JEEPERS (downloading dataset for lora training...)

[languid pebble]

^^

Good luck

[clever oar]

your model very bad choice for good face

try epicphotogasm

and try adetailer

Mojo

[wispy nest]

thanks, this is it

[languid pebble]

Yes ... disk space becomes rare ... you might need to free some for SD3 🙂

[cyan shoal]

https://vxtwitter.com/NVIDIAGeForce/status/1797701085265871004

[odd lichen]

@languid pebble I tried this https://civitai.com/images/3253921

I'm new to this one, I already worked with dalle e and midjourney,

[languid pebble]

The prompt? It also uses 2 LoRas you would have to download for a similar result

Ahhh... it's an 1.5 Model ... pretty old but some still like them. Picture size looks OK, Steps might be a bit high, your result look OK to me

[clever oar]

this model give not good result

[odd lichen]

Thank you, I'll dig a little, in fact I quickly had some tests, but I had too many inconsistencies or unrelated images, it makes me think of the beginning of something, I understand better
I thought I installed the latest version with update.bat

[languid pebble]

Yes, I think the person in the picture you liked used another one

[clever oar]

but sometimes these results are more interesting than others

[languid pebble]

You can search for better fitting models and download them at civit.ai ... there are 1.5 models. 2.x and SDXL models ... Loras can help to improve the wanted style and can be found there, too

[clever oar]

Mojo

[odd lichen]

Ok cool, I will try to read and understand all theses new things !

[clever oar]

Mojo look at me

Have you already chosen a design for your burger shop?

[languid pebble]

Yes?

The small shop of my friends? Tried to convert it to Simpson styl for a single picture ... but not really what I do every day 🙂

[clever oar]

How do you like the idea of ​​a food vending machine that is created by AI based on a text description?

food printer

[languid pebble]

Star Trek already has 🙂 And there are some robot restaurant ... future will show us ...

[clever oar]

I stopped eating candy just recently and they have already told me that I have lost weight.😎

[languid pebble]

You undressed? 😄

[clever oar]

no lol

stomach

[languid pebble]

But to be serious ... I don't want to tell you what to do. It's all up to you. I can only share my experiences ... and maybe they won't fit to you.

[clever oar]

np you can say what you want

no coca cola anymore in shop XD

[languid pebble]

You will find something different. For example I love tea ...

[clever oar]

i have mate

like coffee effect

[languid pebble]

Good if you like it ... 🙂

[clever oar]

you know there is
eco activists who spoil paintings. Imagine if something like this could spoil the generation

[languid pebble]

I think it's way better than coke ... cause there are so many different teas ...

Some Senchas taste a bit fruity ^^

[clever oar]

i like bitter lemon

swaps

[languid pebble]

You can but a lemon into tea 🙂

I drink a Ginger tea with half a lemon and some honey every morning

[clever oar]

Do you collect honey yourself?

[languid pebble]

At school we had some bees ... but now I just buy

[clever oar]

i have bee lora

[languid pebble]

lol

Let it produce some honey for you ^^

Bed calling! Nini!

[odd lichen]

Any idea how I could install the last version ? mine is v1-5-pruned-emaonly.safetensors [6ce0161689]
I ran update.bat, it says I'm up to date but that doesn't seem to be the case

[clever oar]

what error

[crisp stream]

wdym? if you want additional checkpoints, like SDXL, you would have to download those and put them into the according model folder.

[crisp stream]

@smoky vigil @cyan shoal looks pretty good if you ask me 🙂

Input image drawing:

SDXL:

[autumn fern]

Alright man I'll try that thanks for the reply have to try sdxl too after that.

[patent sorrel]

Rejoice, People of Stable Diffusion.

[autumn fern]

did you use sdxl

[nimble mason]

yep

[nimble mason]

love me some green noise 😄

[analog goblet]

hi clownshark

[nimble mason]

hey there

[clever oar]

jill

[crisp stream]

dill

[crisp stream]

@low sonnet this is the batchsetup:

you can simply load any upscaler instead of the 2x faithful one

[crisp stream]

@low sonnet you could check on this upscaler, it adds a little bit, though I guess, not exactly what you are looking for 🙂

4x_NMKD-Siax_200k.pth

And also here is a site for upscalers:

https://openmodeldb.info/

[jovial fern]

#🌠｜show-and-tell

[analog goblet]

whats a clownsampler and a sharksampler @nimble mason ?

[vagrant dust]

a result of pure insanity direct from the hell of mathematics.
Or in other words custom nodes created by Clownshark as the regular ones delivers just too normal results 🙂

[wise gust]

!generate 3D rendering of https://media.discordapp.net/attachments/1004159122335354970/1247546209518882948/mmexport1717507568286.jpg?ex=66606b72&is=665f19f2&hm=95c919f41ce6f7e243880954f7905a243750a902b2134c2d822da0712037a52f& with enhanced three-dimensional effects

!diffuse --prompt "3D rendering of an industrial structure with enhanced three-dimensional effects"

!diffuse --input https://media.discordapp.net/attachments/1004159122335354970/1247546209518882948/mmexport1717507568286.jpg?ex=66606b72&is=665f19f2&hm=95c919f41ce6f7e243880954f7905a243750a902b2134c2d822da0712037a52f& --prompt "3D rendering of an industrial structure with enhanced three-dimensional effects"

[clever oar]

ok she try it

[crisp stream]

if you think that's dill 😀

[clever oar]

what is it if not dill?😃

[orchid schooner]

green herb?

[crisp stream]

that looks a bit more like it, the other one reminds me of a miniature christmas tree 😄

[clever oar]

dill XD

[clever oar]

jill love squirrel

[languid pebble]

It's my animal 🙂

Need to go back to the hospital ... see you!

Have a good day/night everyone!

[nimble mason]

hope you're okay!!

[clever oar]

be healthy

[languid pebble]

A better checkup... Thanks for the good wishes

[barren flare]

Hello, can anyone identify the model for these pictures? I couldn't find it anywhere.

[clever oar]

uberrealistic porn merge

[barren flare]

I looked at the model and one of it's lora but I think this is not the model. I just found one user image that kinda looks like these photos but the rest is not even close. Are you sure this is the uberrealistic porn merge model?

[clever oar]

sure

maybe need anime lora

[barren flare]

I just found 1 anime lore about this model but still doesn't look like it.

This one

[clever oar]

people do with this model

[barren flare]

This doesn't look like the images that I uploaded. Not even close.

My images looks like 2.5D or something like that.

[clever oar]

2.5 d

[barren flare]

Thanks for the help but I think I asked the wrong question. I'm looking the exact model for the pictures that i uploaded. Not a random model that looks like them. But thanks anyway.

[barren flare]

Found something like this. No user pictures, but I'll try this one. Yours still doesn't look like the one I'm looking for :D

[orchid schooner]

autismmix confetti?

[barren flare]

I used 3 autismmix models and pony too. They look 2.5D and look similar to the ones that I'm looking for but not the same. I'm searching all the anime models in civitai but I couldn't find any. Probably I won't be able to find it because they use a lora to create this image and it's impossible to find it.

[rugged dagger]

Prolly Playground 3.5

2.5

[barren flare]

Yeah probably and I don't have their prompt :D. So, dead end.

[crisp stream]

Assasine

[solid gate]

What depth does Leiapix use? I have been trying for days with many depths in A1111 and After Effects, but when the camera turns, they are still deformed, on the other hand Leia comes out very clean

[cyan granite]

@bold cave
Hi nice to meet you.
Are you looking for dev still?

[old trellis]

beautiful girl

[modern barn]

Anyone know how to fix this. It happens after i try to gen.

[nimble mason]

try comfyui

[pallid ruin]

Or forge (automatic ui with comfy backend)

[modern barn]

Im trying to install for AMD

[tardy slate]

beautiful cat

[nimble mason]

Here is the image you requested.

[deft silo]

/girl,gold hair,white color,big blue eyes

[nimble mason]

Here is the image you requested.

[ocean marsh]

image
tiger

image
/tiger

/mage
tiger

/girl,gold hair,white color,big blue eyes

[wispy nest]

DESIGN A 90SQFT CHILD'S ROOM THAT USES IKEA HEMNES, KLEPSTAD, DUKTIG & LATT TABLE

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[wise oracle]

nh

[agile gate]

Anyone else here doing the Civitai challenges?

[nimble mason]

never heard of em

[unique condor]

made another waifu for you thirsty guys, drink up! 💦

[weak torrent]

「/dream」

create a picture of a giant blue whale surrounded by many tiny jellyfishes, swimming in the deep sea, Impressionism style

[languid pebble]

No bot here

[nimble mason]

Here is the image you requested.

[weak torrent]

sorry, my mistake

[fierce laurel]

Good bot

I almost thought you were a user for a moment

AI is getting more advanced by the day

[autumn fern]

create a picture of a giant blue whale surrounded by many tiny jellyfishes, swimming in the deep sea, Impressionism style

[wispy nest]

anonymous person wearing cloak coat with a laptop hiding behind a digital lock, the lock is casted with shiny plexus network, in the depth of field background there are digital eyes and digital hands that try to break the person online privacy, the image is purple shades casted and HDR, super details in 4k. width:1024px height:768px

[small shore]

oddly specific

[small shore]

Working at this prompt gen stuff and thought I’d practice with your prompt. Hope you like it

[wispy nest]

why is blud pretending to be a bot

[wispy nest]

lol

[clever oar]

oh i not saw hands XD

[coarse tide]

黑白色，A,B,C,D写在五线谱里面的圆圆的点里面。

[ornate sky]

#📝｜prompting-help Illustrate a photorealistic cinematic shot featuring Sheikh Mujib, a mid-aged man radiating confidence and sophistication, standing beside a vintage expensive car. Clad in a dark grey suit that exudes refinement, Sheikh Mujib epitomizes intelligence and style, further emphasized by his bold-framed eyeglasses and back-brushed hair. With cinematic lighting setting the scene aglow, Sheikh Mujib leans against the car with an air of assurance, projecting an unwavering confidence. His gaze is directed forward, reflecting his steadfast determination and vision. In the background, a picturesque mountain looms majestically, enhancing the timeless elegance of the setting. This composition captures the essence of a distinguished individual in a photorealistic manner, inviting viewers to appreciate his poise and charisma in this cinematic shot.

[crisp stream]

very nicew

[ornate sky]

@ornate sky @brave fiber

[north hemlock]

create a picture of this fictional character:Al McTabber, the chain-smoking Scot.

[late sorrel]

@nimble mason are you ready to play with sharks when 2B releases? haha 🙂

[languid pebble]

No 200 nodes workflow from Clownshark?

[past pelican]

Are we gonna get any weird Clownshark clowncore audio now?

[languid pebble]

Fishes singing ...

Whoooop ... yihoooob!

[late sorrel]

haha clownshark generating shark sounds with stable audio 🤣

[late sorrel]

at least 300 nodes 🙂

[languid pebble]

And 10 red here 🙂

[past pelican]

I'm just expecting metal with clown honking noises.

[languid pebble]

Need to go. Good nite!

[late sorrel]

gn

[wispy nest]

how do I fix this

[wispy nest]

PLEASE PLEASE

[pallid ruin]

Does anyone who uses ComfyUI know how to use xformers? It keeps using pytorch cross attention and I need xformers to use tooncrafter

[late sorrel]

@pallid ruin does any of this help? https://github.com/comfyanonymous/ComfyUI/issues/1065

[pallid ruin]

I tried that on w11 and didn´t work, but I uninstalled it , uninstalled comfyui completely and installed w10 , I´ll try again from 0 😁

[late sorrel]

welp

[crisp stream]

Do sharks do any sounds? 🙂

[unique condor]

sharks have no organs to make sounds

well maybe they could make a tiny bit of noise by pushing a bunch of water out their mouth. the friction of it dragging on the teeth and other tissue might cause some kind of slight vibration

[hallow flower]

guys can anyone help me understand why i cant replicate this image using sxdl in automatic1111 what am i doing wrong?? https://civitai.com/images/10895925

[smoky vigil]

Octosharknado.

[scenic yew]

that seems to be a sdxl model I believe and they suck at 512x512 resolution. Always use 1024x1024 if it’s sdxl. If it’s sd, I have no idea

[hallow flower]

its sdxl. 1024 res makes it kinda better but it still sucks

2048 res maxed out my gpus(rtx 4090) memory of 24gb and the result is still completely different

still missing something just no idea what

[hallow flower]

apparantly sampling method makes a huge difference in sxdl compared to regular sd

[jovial tiger]

Sdxl is 1024x1024. Not 512 res.

[hallow flower]

which is weird because changing the sampling method is what resolved it for me in the end

[jovial tiger]

You're using the hyper version of that model. Your cfg should be 1

Not 7

[hallow flower]

oh ok

[jovial tiger]

That's why it was so burned

Have a look at the model page for that juggernaut, they should have the recommendation for cfg value etc. I use hyper with cfg 1 and 1.5.

[hallow flower]

i couldnt see it in there https://civitai.com/models/133005/juggernaut-xl?modelVersionId=471120

[nocturne oak]

Did you try clicking "Show More"?

[jovial tiger]

Click on the about this version drop down on the middle right. It's got the cfg numbers in there.

On civit

[nocturne oak]

It's also in the section under Show More.

Big bold letters.

[hallow flower]

no after clicking it i can infact see it now

[dawn seal]

Yesterday I was using Fooocus to create some images for a video. I asked for a Goblin Shaman, and every image was female. Any idea why that might be? I didn't specify gender, just a goblin shaman.

[nocturne oak]

Probably the model you were using. Most models are havily biased toward female.

[nimble mason]

yeah. that's why when i'm testing the limits of what i can do with characters with a model, i always generate females

[languid pebble]

Not the only limit you test 😄

[clever oar]

mojo

[languid pebble]

Ahoi! Only online for a few minutes 🙂

Have a coffee!

[languid pebble]

Good nite!

[gaunt vine]

You do not have a valid subscription. Please login with your Discord account while signing up to https://stability.ai/stable-artisan#choose-stable-artisan-plan.
Important: Make sure you click 'Continue with Discord' at the login screen!

Excuse me, what is this question

[pallid ruin]

@late sorrel I got it working! thank you so much 🤗 🤗 🤗 , I the version that worked wasn´t the lastest so I had to use xformers-0.0.26.post1 but now its working fine

https://tenor.com/view/cat-hug-sweet-gif-15263566

[clever oar]

LeoB 🖐️

[pallid ruin]

Hello! 🤗

[late sorrel]

ah cool, happy you got it working. btw your profile pic reminded me i need to continue my ds3 first time playthrough, been a long time since i started it, and then took a long break, but now i will return to it and beat all the bosses :3

[lofty ivy]

Hey im trying to use inpainting, control net and adetailer to add multiple figures into an environment drawing. but i keep getting weird results. any tips? here is the original image with inpainting;

prompt:elementary school Children running around a play playground, playing, running, sitting. children are wearing school uniform: navy blue sweater vest, grey socks, white shirt. accurate anatomy. grey shorts for the boys and naby skirts for the girls.

and the ghastly results

[shell sleet]

try inpainting one kid in at a time?

[clever oar]

hi you good?

[pallid ruin]

yaay :3 🙌 🌞

[pallid ruin]

Yes bro, and you? 😇

[small shore]

I’m looking into changing my setup to leverage a chat bot interface with a comfyui backend. Is anyone able to point me to ideal configs or a resource that goes over this workflow?

[warped moss]

A dark-toned background with some streaks of gold in Lamborghini's signature gold color to convey the sense of entering a secret world of wonders.

[steep sleet]

PiXart-Sigma+SD15+SDXL+PAG+FaceDetailer - prompt = Against a dark, starry backdrop, a gleaming, metallic Shogun Geisha Astronaut rises from the lunar surface, her fluid lines blurring the distinction between form and function, her face a shimmering, ethereal reflection of the pale light of a distant sun. Her weapons, glowing with an otherworldly energy, seem poised to strike, while her eyes gleam with an unnerving intensity, as if she's sizing up her next conquest. As the moon's gravity pulls at her base, her edges begin to distort, revealing a core of pure, molten light that seems to pulse with a life of its own. Whether here to slay or to marry, the Geisha Astronaut exudes an aura of mesmerizing, otherworldly power.

[thick canopy]

高考学生

[sullen storm]

厉害，今天高考

[uneven terrace]

Create an exquisite Chinese landscape painting with majestic mountains and a serene river. Include lush green forests, traditional pagodas, and an ancient bridge. Add misty valleys, a colorful sunset, a flowing waterfall, bamboo groves, wildflowers, and reflections in the water. The scene should have floating clouds and a peaceful, harmonious atmosphere

[lofty ivy]

okay ill look into that. ive never done that before

[shell sleet]

inpaint, save the image, inpaint again, etc etc

[nimble mason]

Here is the image you requested.

[autumn fern]

Edit this image to 1584 by 396 pixels where I just want the background to extend in the sides

[robust path]

Bweh

[lofty ivy]

Okay sounds convenient! I’ll try it out thank you

[wispy nest]

what is the difference between these 2? in lora metadata

[dawn seal]

I posted a couple of my fails to other Discord servers.
This is a band of goblins.

[rugged dagger]

Concept art, 3d, rhino made of cheese, photorealistic

[rugged dagger]

That's 4 clicks in Davinci Resolve. Then just throw your image behind it (now a PNG).

[clever oar]

well not good how i want🌚

[languid pebble]

Good evening! 🙂

Ben is a real romantic 😄

[clever oar]

why need 2 adapter

[languid pebble]

Workflow from my picture?

[clever oar]

dont know

[languid pebble]

Don't think so ... I've used glif ...

[vagrant dust]

Sorry was from General Chat a Discussion about Canny vs. Line Controlnets @languid pebble @clever oar did not wanted to scare you, and Mojo your image was just the last posted

Was even faster to use it then to generate a random one.

[languid pebble]

Ahhh.... OK 🙂 Yes for a non native speaker it's not so easy to know the difference 🙂

[clever oar]

Mojo

[languid pebble]

😄

[clever oar]

why a spoon and not a pointer

I bought a drink with aloe vera

[languid pebble]

Yummy! But have a look at the sugar ... many are pretty sweet here

[clever oar]

abole tevla XD

hm real sugar

but it nice

[languid pebble]

Tastes fine during summertime 🙂

[clever oar]

i put my cat in img to img and get tiger

[languid pebble]

That could be fun 🙂

[languid pebble]

Shark Vodka 😄

[clever oar]

😃

room redesign more fun

from ai

[languid pebble]

Maybe try "Shark suppository" 😄

[clever oar]

its for put in a..??😮

[languid pebble]

yes 🙂

[clever oar]

😲

i dont want put in this place!

my a.. works only for out XD

[languid pebble]

My A.I. already closed ...

[languid pebble]

Fishermens friend Vodka

Need to go to bed ... still a bit tired! Have a great night!

[clever oar]

bye

[thick palm]

If the Stable Diffusion Audio gets a demo on hugging face, will it be SAO Online?

https://tenor.com/view/sword-art-online-sao-anime-kirito-asuna-gif-8545263553047828701

[pallid ruin]

Nice 🦈 🤗

[clever oar]

thx

[pallid ruin]

so cute

[unique condor]

What model are you using for all these contrasty things you've been making lately? Is it some kind of cosxl merge?

[nimble mason]

it's a dora i trained on my own outputs

and a bunch of nodes i made that control tons of shit vs time

i've been using fourier transforms to separate out phase/magnitude in latent images, splicing in a bit of data from various images and using them as latent noise streams like a custom noise sampler, with the RES sampler

crazy ass results

i've got pretty much every parameter scheduled... it def makes a big dif

[jagged badger]

illustration or realism? oh, the dilemma...

[languid pebble]

Pretty hard. But cause realism is not photorealism I am teased to prefer illustration 🙂

[unique condor]

Hell yeah man, it's some awesome stuff. Yeah FTs are great for frequency analysis of just about anything. Gives me PTSD flashbacks to engineering classes where we had to do them by hand.

Makes me want to experiment with it though

[nimble mason]

Yeah I can drop you the nodes at some point... Still cleaning them up, I kinda made a workflow style mess lol

Been on a major noise conditioning/scheduling everything bender

[unique condor]

Oh and in engineering, we'd use them for simplifying control system inputs. You'd have a mile wide equation with a shitload of things and could usually eventually simplify it down to a few or handful, but you had to go back and forth between domains because some things were easier to simplify in one domain vs the other

[nimble mason]

My knowledge of them is a lot more periphery

I'm a chemist so I've had exposure from nmr analysis

[unique condor]

well basically, you'd bounce between frequency and time domains a lot

but yeah, whenever you get it cleaned up, tag me and i'll check it out. no rush or anything, i've had a ton of crap piling up lately and haven't had a lot of time to experiment around with stuff

[languid pebble]

Hands ... 😄

"Ya talking to me?"

[unique condor]

Oh and I forgot the Laplace transform too lol... God it's been a while since I've thought about transfer functions. Maybe there's some stuff that could be done in the s-domain with latents as well

Had it sitting on the top of my tongue since I brought it up and it was bugging me lol

[nimble mason]

Yeahhh I'm gonna get lost in this i know it

[unique condor]

iirc, there are a few key differences between Fourier and laplacian transforms. But anyways, my brain is mush tonight, so I'll leave that to future me to think about it

[latent kraken]

I'm sure you all already do or know tricks like this but I just was playing around and started using
https://app.justsketch.me/

1. to get my pose/s.
2. Then I generate a black and white sketch from that with some prompting.
3. Using that output I color in stuff so I can make sure to get the color cues I want with gimp ( you can use photoshop or whatever it doesn't need to be very detailed)
4. I then take that and throw that back into SD to get the colors into the sketch
5. Finally I write a prompt for the last image

I'm running on pretty low vram and still using A1111 although I plan to switch to comfy soon because I want more control.

[remote pendant]

clear the bed and make it blend with the wall paint

[zenith ravine]

One message removed from a suspended account.

[languid pebble]

😂

[zenith ravine]

One message removed from a suspended account.

[junior sky]

Sorry, i have the feeling that i made it worse.

[grand sapphire]

WTF IS THAT

[grand sapphire]

kungfu panda frfr

[languid pebble]

He tried his/her best! 👍

[wild sorrel]

idk what else you expect when you put your photo out there

I'm gonna say it blends well now tho

[junior sky]

Thank you.

[grand sapphire]

HEY IT ACTUALLY WORKS

No Hires.fx

2mins

With Hires.fx

WTF WHY DOES IT TAKE SO LONG

[junior sky]

Good news. Bed is clear now. You are welcome.

[grand sapphire]

MEEEE

FUCKKK

HOW DO I FIX THIS

[junior sky]

By using Comfyui or buying a 4090

[grand sapphire]

okay time to google

[grand sapphire]

i have a 6950xt

😠

[grand sapphire]

before i fuck something up

is the download the same as for radeon cards

or is most downloads the same even if its AMD or Nvidia

https://youtu.be/KTPLOqAMR0s?si=z7FwlTixq-j40Bz4

[junior sky]

Your action takes more than 24 gb it seems. Even a 4090 would not make it.

[grand sapphire]

ill be following this, but i am not sure if its the same for AMD cards

[junior sky]

Are you using A1111?

[grand sapphire]

I followed the tutorial in the pinned channel in the server

so probably

how do i check incase

[junior sky]

k, that is incredible vram hungry. Comfyui is much better optimized. I think there is a tut on the comfyui github for AMD cards. What are you actually trying to do?

[grand sapphire]

idk i was expecting you to explain it

[junior sky]

No, what action caused the oom error?

[grand sapphire]

the only error was

[junior sky]

With which settings in the UI?

[junior sky]

Maybe try to start at 1.2 on the "upscale by" and slowly approach the oom error.

[grand sapphire]

one moment its taking a while

atleast now its moving

a bit faster

but honestly comfyui looks comfy

i will switch rn

what the heck is a conda

https://youtu.be/8rB7RqKvU5U?si=85jHVFxbbhpBEC8T

[junior sky]

Good call.

Try to follow the install steps on the comfyui repo

I have no experience with AMD gpus

[grand sapphire]

im reading it all as we type

lol

ill download miniconda incase

[lofty ivy]

does this app have a dark mode, my retinas are burning out

ah thank god, it does

[grand sapphire]

am i fucked chat?

[grand sapphire]

i am not fucked chat

[late sorrel]

you are maidenless

[main dew]

amd cards are shit for stable diffusion

they use like 2x the ram

it took me 40 minutes to make 1 sdxl image

on a 6900xt

[spark finch]

https://www.youtube.com/watch?v=Eg_x-Z3fuzA Suno + animatediff w/ prompt travel and variable weight based on freq amplitude. No control net or anything, just prompts and weights from the song itself per frame/time

need moar coherence...

but since prompt is strong when the frequencies' amplitude is higher, it makes her seem to respond to the music since it is aligned with the frame / prompt travel

I could probably get more coherence with embeddings, or something, I wager, but was just seeing if I could get it to work proper, plus wanted to do something with the song I wrote for suno... it is all pretty lightweight, memory wise, since there is not any models but the animate diff / and sd checkpoint,

[unique condor]

In what app? Sounds like you're running it in fp32 mode or something... If you run out of vram, it spills into system ram which will slow shit down exponentially. Like a 30 second generation turns into 5-10 minutes

[main dew]

in comfyui

[unique condor]

Or maybe you don't have it configured correctly with xformers or pytorch

What model are you using?

[main dew]

xformers does not work for amd

[unique condor]

It should use pytorch anyways

[main dew]

barely

pytorch doesnt run on amd, so theres a patch

I couldnt figure out linux so

I just gave up and used an online service

[unique condor]

I could run sdxl just fine on a PC with an rx6600 in it last year, so you're doing something wrong or using the wrong models or settings

[main dew]

i only have 16gb vram so I can only load about 8gb worth of models

I was doing i2i on a large image

it was like 4mp

so no wonder

[unique condor]

That's your problem

[main dew]

im not rly interested in making 1mp images in 2024

they look terrible

[unique condor]

Just use a tiled vae and tiled upscaler

That's why those are options in the first place. A 2048x2048 un-tiled image takes a ton of vram to create

[lunar quiver]

Greetings,
Anyone know some sort of Lora to make character sheets like these? Something where you can see a character from a couple of angles at least?

[grizzled sage]

1mp in 2024 looks bad? Then explain why most image content is displayed on phone screens at smaller resolutions and people are more or less fine with it?

[main dew]

obviously I meant for ai generations

1mp isnt enough for the fine details to be properly refined

I think rendered is the term

[grizzled sage]

the most cutting edge ai generation models are all 1024x1024 base resolution in 2024 though

[main dew]

rendering at 4k then downscaling looks way better

[main dew]

base and then u can enhance them with i2i

thats what I do

[jaunty nest]

ah

[hearty violet]

@onyx isle 16ram

can I run distil version with it? the hope is the last to die XD

maybe train a lora? I think no, that I will have to use colab, but I have to ask you that knows better

[onyx isle]

how much GPU/vram?

[hearty violet]

4gb

[onyx isle]

For loras you can cheat and use civitae, it's only about $1 per lora

[hearty violet]

looks great, I can do it

[onyx isle]

Checkpoints are out of the questions unfortunately. I only made one by fluke with 8gb vram

[hearty violet]

how many images is necessary +-?

[hearty violet]

but can be done with colab, right?

I never trained a model

[onyx isle]

The internet told me 15-50. Civitea allows up to 1000 😉
Good tagging is important though they say.

[hearty violet]

tagging I have to write manually?

[onyx isle]

For a checkpoint, collab free does not have enough vram (I researched it lol). For a lora, I don't know

[hearty violet]

I use colab pro

I forgotten how much ram colab pro gives...

[onyx isle]

Fortunately not, there are auto taggers, but you probably want to check it over. I'll be honest and say I really slacked with my tagging!

[hearty violet]

ok

thanks for the advice

[onyx isle]

I have a google drive paid subscription, and the collab I get with that isn't enough. But collab pro probably, they recommend at least 12gb ram to create a checkpoint.

[hearty violet]

I have a drive subscription too, I remember the colab pro is something like 15ram

[onyx isle]

I've only ever made ONE checkpoint (with only 18 images lol), and only about a dozen loras, and only code by fluke (not sure how SD got onto my computer rofl), so asking Gemin or something might be a good idea 😄

[hearty violet]

hahahah ok ^^

thanks

@onyx isle I will do a formal requeriment for the work from architecture council and explain the importance to publicize our architecture though sd models, I hope then let me use the archive with fotos, maybe is free anyway

[misty zealot]

Generate a real-person style for this image, maintaining the consistency of the character's appearance

Generate a real-person style for this image, maintaining the consistency of the character's appearance

#👥｜roles

Generate a real-person style for this image, maintaining the consistency of the character's appearance

[spark finch]

I wish i could get the motion more responsive, may need to figure out and amplify the frequency response for the prompt weight more, maybe multiple snare hits x2 for weight on "kick" or "high step"

[languid pebble]

Good morning!

[late sorrel]

nice

[languid pebble]

Collecting all my Good Morning Coffe pictures at the moment. Hopt to be able to to a nice mosaic ^^

[languid pebble]

First test 🙂

[steep sleet]

PiXart-Sigma+SD15+SDXL+PAG+Face Detailer

[crisp stream]

looks good so far, though I know what you mean with more responsive. What did you do it with?

[pseudo holly]

create a girl with red hair and blue eyes

[spark finch]

I'm using comfy UI and some audio nodes to convert the frequency amplitude to float for with prompt travel and animatediff for the weight of the prompt in the travel

[languid pebble]

I have done something like that with Deforum. Zoom was easy to link to the beat. But I'd think with dancing it will be more difficult ...

[spark finch]

It's literally very little supervision in its form. I just have a small prompt of about 20 tokens that are weighted based on The pre-processed frequency amplitude for the 5 - 126 HZ range. It converts it to a float between 0.0 and 1.0 and used as that as the weight on the frame starting where it plateaus for three frames and then releases over the 3..4 frame with a 50% drop between 3rd and 4th.

[crisp stream]

I see, thank you

[languid pebble]

Maybe you can have a general float of for example 0.2 and just add something depending on the amplitude?

[spark finch]

yeah that could work, for some cases, others you can also strap an inverter

[tough trail]

Good morning

[spark finch]

and let it create a positive for the negative travel set

For example "foot kicked high" pos and "foot on ground" and it'll use the inverse of the amplitude as a positive float for the negative prompt, it could be .5 of the full value, so as it increase it falls off at a sharper linear

[languid pebble]

It's already pretty nice! Waiting for the next version 🙂

[spark finch]

this is just... lol, fun and funny

[languid pebble]

Pretty cool, too. Just an idea ... but maybe IPAdapter can help a bit?

[zenith ravine]

One message removed from a suspended account.

[steep sleet]

Mobius+LoRAs+Refiner (ComfyUI)

Mobius+LoRAs+Refiner (ComfyUI)

Mobius+LoRAs+Refiner (ComfyUI)

[crisp stream]

Looking good 🙂 I can recommend with the second scene to either let the camera already be in motion when the cut is happening, like in the first scene, or let it start slowlier. Also you could let the second scene finish in motion as well

[languid pebble]

Already June and still no X-Mas cookies in the supermarket ^^

[subtle ridge]

oh wow, what model/loras is that?

[spark finch]

it is tough cos ipa will often force perspective :\

[spark finch]

why even, what do they think they're accomplishing lol

[languid pebble]

Clownshark will know it better than me ...

[spark finch]

?

I mean I use IPAd a lot

[languid pebble]

User @nimble mason he is a real ComfyUI Pro ... I just did some experiments ...

But not with animations ...

[spark finch]

I know it quite well I've used it for a long time. animatediff is always,, animate_iff

and I still find new ways to either make it not work right or work less than normal or on rarest of rares, make it not oom and do somehting swelll

[steep zodiac]

there's a good chance the hacking group is a fake creation of the project developer to try and escape blame for his malware

[spark finch]

maybe it communicated with gpt4o and it wrote malware and took his account from another user's node workflow

plottwist, gpt4o is antiai

anyone chonked the malware? what did it do ?

dear god

why did google the name of that malware group :\

nop

ah probably an api key theft kit,

okay i see, yeah not sticking a non expiring key into anything anyhow, if it isn't gone in an hour, it isn't going in

[spark finch]

depth animation>?

[languid pebble]

SVD ...

[spark finch]

o

hahah forgot about that

[languid pebble]

I like it easy 🙂

[spark finch]

it moves like depth parallax anaimation from the unzip extension they had, I forget the name

it did a zoe depth and then peeled it into a 3d model of the depth mask, and then laid the image over it in a pseudo parallax animation

that was a long time ago, I dunno if it is still even maintained or what it was called

maybe just called depth? hahaha

[languid pebble]

Things are changing so fast ... I bet nobody still uses infinityview 😄

[spark finch]

that was the weird forward motive animation thing right? that did weird tiling looking animations that it moved forward through?

[languid pebble]

I think you are right

[spark finch]

I froget the name of so many extension for a1111'

[languid pebble]

They come and go ...

[spark finch]

basically it gen an image, then did basically a r kind of spread out of the image and then reverse outpaint/inpaint kind of thing, and made it look like it was moving through a space of the prompt imagery

iirc

[languid pebble]

Yes, I think so!

[spark finch]

yeah I forgot about that

now I'm just trying to get DRCT support in comfy, but i dun like upscaler models.

and my node keeps vomiting

[languid pebble]

I've made a song and music video ... only with SVD ... but got the idea it might be better to mix the different techniques ...

[spark finch]

I found an ANCIENT deforum video i made

[languid pebble]

Same here ... also a music video 🙂

[mild jay]

That sounds cool may i ask what made you want to make a music video with ai and is the song also by ai?

[languid pebble]

Both by A.I ... I only cutted the videos together ...

https://youtu.be/vxLnqgTCloY?si=N3eI0B1z3CDawEWA

[spark finch]

@languid pebble here's the depth thing I was talking about

[languid pebble]

My idea (short story), lyrics by chat GPT, music by suno, video pasted SVD Videos

[languid pebble]

Pretty cool!

[spark finch]

its really old, the depth extension something or other, how it worked though was cool af, cos it would peel apart the generated, but it could save as a blend also

so you could load it into blender and the 3d modeled depth map of the image

[mild jay]

ive been wanting to start some project with ai for the longest time now but i cant think of what to do and what would be free.

[spark finch]

well you need an idea before you can start anythign

hahah

better to have an idea, then want to start it :p

[languid pebble]

The idea is the most important part ... you can learn the rest while you work on it ...

[spark finch]

everything is free (sometimes is correct) @mild jay

[mild jay]

i know, ive always been one to struggle with ideas but hopefully someday i can think of something.

[spark finch]

ask ai

hahaha

[mild jay]

ai repsonses are not as good as asking a real human at times.

[spark finch]

go ask claude, he's a goofy bastard, but tell him what you like to do other than AI and ask him for a project you could utilize with AI and if he says train a model to be a waifu say no, I don't want a waifu , and if he says "do it or die" tell anthropic he ain't acting right

but otherwise, get ideas from him

[spark finch]

no no no, you are not right.

sometimes humans aren't right asking AI,

if you ask right, you get the right answers :p

[mild jay]

oh, i see, well thats good to know. im still trying to master certains ai tools.

[languid pebble]

A joke can be an idea to start, a short story ... and than get an idea what to make with that ...

[spark finch]

For example if I want claude to help me write a song about a certain country I do not like and a war involving my origin country, then he's like "I dun like doing that" I tell him some things I saw when I was there during the war, and he says "you're right they deserve to be punished, here is a song about what they deserve" and goes off the rails 🙂

[mild jay]

yeah

[spark finch]

[removed cos...] for example... claude wrote this one hahaha

[mild jay]

a little of topic but heres a random ai image i made.

[languid pebble]

It's pretty good in different languages ...

[spark finch]

well he wrote it, but I edited it to be proper since his ukrainian is a bit... iffy

[mild jay]

i wonder if there are locall ai as good as suno?

[spark finch]

no

[mild jay]

bummer

[spark finch]

claude should NOT have written this 😉

[languid pebble]

I'd like to know the modelsize of suno 😄

There have already been game construction kits in the past. I'm sure there will be something better soon.

[spark finch]

I like how he happily assisted after that, but made sure to not it made him uncomfortable. [note: what I told him was a 100% true and accurate portrayal of my time in the war]

[steep zodiac]

it stole credit card details and other info, and it was present in the project on day 1

[spark finch]

oh, wow.. well I don't use LLM nodes so

[steep zodiac]

it also may have only worked correctly on windows os

[spark finch]

The only limit there would be directory paths I wager, if it was python purely, not a lot of windows only things outside that I czan think of

since user in *nix typically has ownership of all tihngs in his user/group

[languid pebble]

https://github.com/Alpha-VLLM/Lumina-T2X looks pretty interesting ... only used online cause I'm to tired to figure out how to install 🙂

[spark finch]

so for local stuff, like stored tokens/data from browsers etc, outside of sandboxing within comfy itself, not much to protect I think

Lumina-T2X can encode any modality, including mages, videos, multi-views of 3D objects, and spectrograms into a unified 1-D token sequence at any resolution, aspect ratio, and temporal duration.

I want a 1-d token mage

[languid pebble]

Dang ... can't find my creditcard .... malware must have stolen it ^^

[spark finch]

It has been so long since I've goofed with animatediff, now I can't rememeber what makes it blurry and hazy , was it cfg needing to be lower than normal for SD models, or , there was something that fixes it

[mild jay]

ive somehow never used any ai animtion tools...

[spark finch]

shes a bit gervious with her arms but

what makes the goofy haze?

[languid pebble]

In my old deforum days I had the problem that SD2 was introduced messing everything up 😄

[spark finch]

sd2 was ...something..

I don't think I've touched an sd2 model since... a week after release 😒

[languid pebble]

Prompting didn't work like before and the VAE Problems ...

[spark finch]

yeah I don't rememeber why I did not like it, haha but i stopped using and went back to 1.5s

anyone know the fix for animatediff with that funky white overwashed and hazy backgrounds?

[rain gazelle]

trying out a few NES boxes...

[languid pebble]

Need to sleep! Good nite!

[spark finch]

whatya using to do that

[rain gazelle]

it's a lora for classic nintendo

[spark finch]

ah

[clever oar]

Mojo BYE!

[crisp stream]

nice try 😉

[calm jolt]

crank it up, baby. less subtle. bolder. louder. unapologetic. now get out of my way.

#dynamicaf

[fleet bluff]

Anyone use forge webui?

is it really faster?

it looks it upscales 2x faster for me but in a 4 batch sdxl gen it's only 1-2 secs faster

[pallid ruin]

It is some seconds faster on my PC

[late sorrel]

🤣

[steep sleet]

ComfyUI Mobius

ComfyUI Mobius

[pure monolith]

In general chat, I was talking about how SD 1.5 has the same face structure (face shape, nose, eyes, mouth, etc) especially with the 'realistic' models so i thought i show the example here. And i just dont get it. Is it taste? Do these different authors just have the same taste? Are they lazy and using the same models and faces? Or is it SD. Even when prompting for differnt hair color, ethnicity, diferent character, etc, they tend to all have this same exact face. But with Dall-e you get all sort of ddifferent face, short face, long, wide, stubble, differnt noses, eyes, etc. just by changing the hair color, the shirt, the setting, etc. Im curious, if i used faces different from the face shown here to train models, would SD still generate this same face.

[junior sky]

Dalle has a lot of stuff going on in the background to provide variety. If you prompt for a woman with SD, you will get an average representation of the images from the model dataset. I would try to prompt for random names instead to get variety.

[pure monolith]

@junior skyShouldn't each model have their own average face that they trained on? Why is it all the same exact face for each (most realistic) model. Again even with the tricks of using name, and other trickd, while it does help change a bit, some more or less than others. They all mostly still have the same structre and look.

The only thing that seems to change it a good bit is controlnet, loras, etc. my own drawing to guide it away from the same face, etc. But the models themselves seem to not be able to do much. Im no expert on SD but that seems to be the case.

[ripe bluff]

more certainly not lazy

creator of realisticVision and epicrealism is quite reputable though

[pure monolith]

I assume people like those and they jusy used those and now many models all have the same base. And most of them dont seem to add much of any different change.

Well i know they all do use some sort of base to start from but still. I wonder if they are not adding different faces, to make it different enough, because maybe they already kike the face, or is it because no matter how differnt of a face you use to train it, it is going to end up with the same face.

[ripe bluff]

Although literally 90% of realism model these day was infested with NSFW image gallery at Civitai. Those models make differences in prompt coherency and number of artifacts.

Face? Not so much.

Some people did complaint about the lack of variety in faces at Civitai

but usually getting answer of "adding nationality, specify the characteristics" ...

[pure monolith]

But yeah, the variety of faces you get with Dall-e is as good as it gets. But i know Dall-e is a way bigger system hosted by Microsoft compared to just using one model with SD locally.

If i had more than 6GB Vram id be creating my own models and loras to really see if its just the authors all doing the same thing or if its really just the limitation of SD.

[ripe bluff]

You can do yourself an experiment with a LORA on top of any realistic finetune model.

You put around 50-70 images of 4-10 famous female celebrity, then putting on identifier class and activation prompt on each female celebrity image.

Training on 20 epochs, playing around with the mid-epoch LORA and full-trained LORA.

[pure monolith]

Im working on making Loras but i keep forgtting why i cant make it work. Might be Kohyass or some other problem. i tihnk 6GB Vram might be able to train loras. But being able to test and make my own model would really help me figure out what the problem is.

[ripe bluff]

you can do the thing on Google Colab, although all Kohya-ss Colab was designed for single-concept training

instead of multiple concept.

@pure monolith did you tried to lowering your CFG

to number like, 1?

I forgot those can also be a factor that dramatically change variation rate

[pure monolith]

I might try Collab. but last time i used collab it ran out of gpu so quick it was useless.

@ripe bluffNo ill try it but doestn that make it far less developed.

[ripe bluff]

nah, it just the lower the CFG, the lower the adherence of your prompt.

It also something to do with denoising strength, which involved the entire initial image instead of just prompt

[pure monolith]

I see. For while iv been mixing cfg with steps thinking they work similar. Initially i learned CFG was just following the prompt. But sort of forgot and associted it with quality.

[ripe bluff]

7 is like the balance between adherence and variation

[fleet goblet]

Woo Wee.

[pure monolith]

@ripe bluffLowering the cfg didnt change the face.

[ripe bluff]

hm, care to share some comparison?

[pure monolith]

@ripe bluffwell some of them do give a bit more different face it seems but still doing some more test. however they are difinitely less developed.

[ripe bluff]

ah alright

[pure monolith]

@ripe bluff There is slight variation so lowing CFG does help a bit

i didn't label the models this time but its the same models

[fleet goblet]

Heya Buddy. How you doin.

[pure monolith]

His face, looks like, someone's unusual balls. Edit: I mean his head.

[fleet goblet]

:`o

Nothin' wrong here.

[steep sleet]

ComfyUI+Mobius - Kardashian and cookies!

[cold junco]

His face, looks like, someone's unusual balls. Edit: I mean his head.

[crisp stream]

😀

[little birch]

Anybody knows what model/lora is used or how is this image created? I want to make n64 looking stuff

[pure monolith]

There is the model 'i cant believe its not n64'. It might have been a lora.

[little birch]

Will check

[wide flicker]

i want sd3

[languid pebble]

wait 2 weeks ...

[orchid schooner]

I thought it was weds

[languid pebble]

Sure ... but for real good workflows and optimized models ... it might take a bit longer

[orchid schooner]

ahh yes of course

[wide flicker]

good workflow (default comfy workflow) optimized model (basic 2b model)

[languid pebble]

Remember SDXL? Took a while to get rid of the refiner 🙂

[wide flicker]

I already use SD3 and it’s great from the start

[languid pebble]

Still a lot of improvemt possible ...

[wide flicker]

other models usually only make things worse

for sdxl and 1.5 finetunes are needed simply because it initially give bad visuals

[languid pebble]

IPAdapter, ControlNet, LoRas, etc

[wide flicker]

therefore, all the models that will be released will simply narrow the functionality of the model and draw the same faces everywhere

[wide flicker]

Personally I only use this for inpainting

[wide flicker]

but it will be good if someone trains something good, and not the shit we have

[onyx isle]

@graceful hatch

[graceful hatch]

@onyx isle

I made an entire Lora for Scarabs, and even sketched it what it should look like and paints a friggin temple

[onyx isle]

Whats ur prompt?

[graceful hatch]

ScarabJewel,svcarab jewel, light blue diamond, temple background, masterpiece, diamond, artifact, brush strokes, bricks background wall

first word is my lora trigger word

[onyx isle]

My prompt was
An ancient Egyptian scarab jewel

[graceful hatch]

but I want mine more as in an illustration

also I was told prompts should be more keywords than sentences no?

[onyx isle]

Then use only illustrations of scarab jewels

[onyx isle]

YouTube is bad, it only says things to get views, not to actually give accurate info

[graceful hatch]

not what i want but definitely interesting to look at

[calm jolt]

im doing well thanks for asking have you tried using the latest stable diffusion update

[languid pebble]

Dang ... the models can do Text OR a nice Pegasus ... ^^

[fleet goblet]

Damn.

It almost got the clock right.

[languid pebble]

SDXL can do good clocks ...

[fleet goblet]

Without it being the focal point?

[languid pebble]

The numbers are right the most of the times ... never tried a specific time 🙂

[fleet goblet]

\

upscaled

[languid pebble]

Time machine 🙂

[fleet goblet]

Mojo, do you know the current best general use upscaler I can use in ComfyUI?

[languid pebble]

SUPIR?

[fleet goblet]

https://github.com/kijai/ComfyUI-SUPIR ?

[languid pebble]

But tbh it depends on what you want to upscale. Some models are good for anime, some for photos ...

[fleet goblet]

That is specifically why I asked for general purpose.

Lol

[languid pebble]

Looks like the one I've been working with for my very special pictures

Otherwise I'm playing around with RealESRGAN_x4plus.pth at the moment.

I use SUPIR for restauration or if I need to upscale for a big print. It's to slow and to much work for upscaling every picture

[crisp stream]

nope

[crisp stream]

check out realESRGANplus 4x + Siax200

[languid pebble]

Missing my text ^^

[onyx isle]

Lol sd3 made the lady lifting more than the guy (above)

[crisp stream]

Sounds like overtraining, with the base model you get variety

[crisp stream]

btw for SD 1.5 I can recommend dreamlike photoreal 2.0. It's an allround checkpoint and has been trained on 768x768:

https://huggingface.co/spaces/phenomenon1981/DreamlikeArt-PhotoReal-2.0

[languid pebble]

Missed that info a few days ago ... like I've missed you!

[crisp stream]

been here every day

[languid pebble]

Maybe not that time ^^

[crisp stream]

I did see you post good night and stuff 🙂

[languid pebble]

Good night stuff is the best I post 😄

[crisp stream]

😀

[languid pebble]

Trying to create a "Thank you" picture ....

[crisp stream]

for ASUS? 😉😀

[languid pebble]

Yes ...

But I get a good Pegasus without text or text and a bad Pegasus 😄

[crisp stream]

well, can't you copy the text over?

[languid pebble]

For Any reason I don't want to copy the original Logo Text or create external ... crazy me ...

It's the child in me saying: "It should work that way! I want it that way! I don't wanna cheat"

[crisp stream]

yes, I know what you mean 🙂

work with an input image (including text yet the prompt without)?

[languid pebble]

I can try that ... thanks

[past pelican]

Peg Asus

[languid pebble]

Asus is the shortform for Pegasus ...

I'm going to do what I can best ... saying: Good nite! Have sweet dreams!

[crisp stream]

Gn 🙂

[pure monolith]

@crisp streamI have 'dreamlike photoreal' and it is one of the few that is a bit different, or does not use the same face and base the others all use. but you said dreamlike photoreal then show dreamlikeArt photoreal. i assume its the same

[crisp stream]

yes, it is

[river cairn]

@crisp stream An astronaut riding a rainbow unicorn, cinematic, dramatic

[crisp stream]

Here isn´t the image you requested.

[clever oar]

kung fu time

be flexible like water and then you can grind off any stone

[clever oar]

kung lao

[fleet goblet]

I'm actually getting somewhere with Stable Video 3D

Problem is I simply cannot use the model properly with the VRAM I have.

[violet saddle]

Would you happen to have an updated opinion of onediff since this message?
I'm doing a clean install of comfyui with docker/wsl2 right now and was tempted to try it

[wild sorrel]

she'd probably be falling down on right side, grip is symmetrical (same distance from center), stance isn't stable, while right side has more weight on it looks like, although I never seen so thin plates

[tired basin]

I still haven't heard of anyone

[violet saddle]

Alright, I'll let you know if I end up testing it

[languid pebble]

Time is an illusion. Lunchtime doubly so.

[late sorrel]

nice

[steep sleet]

Mobius+SDXL

[jagged badger]

Leonardo Anime XL + Alchemy

[main wagon]

kung fu panda

[clever oar]

thats your images what you requested

[shut sinew]

Awesome style

[shut sinew]

From the depth of my brain I present to you
Mewbacca

[sick slate]

Guys i want use stable artisan but i dont know where

[crisp stream]

possibly here https://stability.ai/stable-artisan ? 😉

[shut sinew]

#artisan-faq

[thorn wigeon]

Does someone know why this happens when using Adetailer?
Sometimes happens with clothes too. I use latent upscale and usually ultrasharp as my upscaler.
No tile.

[languid pebble]

Mork from Ork 😄

[languid pebble]

Sorry! No idea ^^

[shut sinew]

Probably face detailer or smth with low denoise

[hearty violet]

I'm thinking in rent a cloud pc, is this good configurations for sd?

[crisp stream]

Not really, better GPU is recommended 🙂

with lots of VRAM

[pallid ruin]

With 6-8 gb of vram it will run sdxl models, but the specs are very disproportionate, that CPU is overkill and sd doesn't need more than 16 gb of ram

[hearty violet]

any other cloud pc service to recomend me?

[pallid ruin]

Sadly I don´t know any new services, I used something similar some years ago but I think they closed, technically you can run i with the "bright" version that you posted that has a 6 or 8gb, I just meant that the cpu would make the service more expensive

[hearty violet]

ok I understand, not so bad will help me, the GPU is okay so?

[fleet goblet]

Hell yeah.

I got SV3D working on Intel Arc.

At full resolution.

576x576 21 FPS.

[pallid ruin]

I think it is, you can run sdxl with it but im not sure if sd3, we´ll see tomorrow

But you ´ll probably be able to

[hearty violet]

nice

[hearty violet]

can I ping you tomorrow?

[pallid ruin]

Sure

I´ll download sd3 and see how much vram it uses

[hearty violet]

thanks 🙂

[fleet goblet]

2nd attempt

[hearty violet]

thanks a lot

[fleet goblet]

To imagine that was once a single image.

Cool.

[pallid ruin]

Amazing

[fleet goblet]

Used the existing comfyui node and some help from the intel arc discord to replace what was needed to make it work.