🏞｜general-with-images

[lone halo]

it processes it slower but it uses less vram and etc

thats sick

u could also inpaint in animatediff

np

yeah the video i showed u

is inpainting a subject

and animating them only

using alpha mask

it makes the model focus on the subject and not the backround which increases the quality

[late sorrel]

luke is about to have a heart attack with all this greatness that he is discovering with comfy :3

[lone halo]

ngl i was the same when i started

[late sorrel]

im still discovering greatness as we speak

[lone halo]

u could try vid2vid with sd 1.5 models first before trying hotshot

[late sorrel]

civitai has some motion models too

[lone halo]

yeah im sure it will

[lone halo]

noneed, you could just use the animatediff v3 model, it works great !

[late sorrel]

time to make some money 🙂

yea i didnt check all the animate stuff, its not my main thing, but i saw great stuff

[lone halo]

hotshot XL is for sdxl

animatediff v3 is for 1.5

[late sorrel]

wasnt there another one for xl?

[lone halo]

there is animatediff xl

but hotshot is like the best

[late sorrel]

yea listen to minjun cause im no expert with animation stuff :3

[lone halo]

😭

agreed luke

also u need to know what controlnet models are

[late sorrel]

controlnet is awesome sauce

[lone halo]

they basically detect the position of the subject to help the model generate images easier

[late sorrel]

densepose 😮

[lone halo]

u can try enigmatics workflow on youtube first

it uses sd 1.5

it uses ipadapters aswell and he explains it pretty well too

ill link

[late sorrel]

yea let him do the simple stuff first, otherwise his head gonna explode :3

[lone halo]

https://www.youtube.com/watch?v=P4IdHKHrb48&t=863s

try a 8 second video at 20 frames per second

at first

u can use what ever model u want

[late sorrel]

street fighter? 😮

[lone halo]

and u could upload a reference image as to what style ur looking for

[lone halo]

lol

literally anything

[late sorrel]

lol

[lone halo]

ipadapters are kinda goat nowadays

[late sorrel]

omg Mateo is a god with ipadapters

[lone halo]

also accidentally generated this the other days

[crisp stream]

three words... Deforum

[lone halo]

not me remmebering how i used to bother him so much for help LOL

[late sorrel]

one day people will be able to make their own anime 🙂

[lone halo]

some ppl are doing that rn

yeah u could also use this node called rave for fun

its almost like editing a video

u could turn a car moving into a rhino running

or a car driving into a warship moving in the ocean

[late sorrel]

can you link the rave thing, im curious

[lone halo]

yeah sure

ill link a workflow

[late sorrel]

well i wanted to see like the project page or whatever, just to see what it's about :3

[lone halo]

yeah sure

https://rave-video.github.io/

[late sorrel]

thx ❤️

[lone halo]

rave uses noise shuffling

its an algorithm that generates natural consistant images

[late sorrel]

what pc reqs for it?

[lone halo]

i mean

it depends on what ur trying to do really

i could share a light workflow

[late sorrel]

sure

[crisp stream]

Deforum

[lone halo]

https://openart.ai/workflows/koalanation/total-video-transformation-with-rave-and-animatediff/OmIGg4ipjVoEm89QtGpk

not to flex or anything

im the first person to post a workflow using rave in civitai

[lone halo]

not me wanting to live in that world

good job moofi

[late sorrel]

cool stuff 🙂 il check tomorrow tho, cause late and tired

[lone halo]

night lodis

[late sorrel]

yea gn

[crisp stream]

thank you 🙂

thanks 🙂

Deforum is available for Comfy btw

locally

while these were done during beta-testing the Deforum Discord bot

[lone halo]

never knew that

since when?

[crisp stream]

don't know exactly because they announced it in early february and I finally discovered it having been released already in early April. So not that long ago

[lone halo]

so theres a node for that?

[crisp stream]

several:

[lone halo]

thanks m8

helpful

[crisp stream]

pleasure

[lone halo]

ima try to build up a vid2vid because ive seen some people use it on the webui

[crisp stream]

yeah sure, haven't come around doing it yet 🙂 Btw: https://github.com/XmYx/deforum-comfy-nodes

And these are the example workflows https://github.com/XmYx/deforum-comfy-nodes/tree/main/examples

[wild sorrel]

What is this ui thing? I mean tabs and eevrything

[crisp stream]

The workflow I sent is already modified, yet not fully worked over. Best to read the documentation anyway

[lone halo]

its a website where u can find workflows and etc

[lone halo]

thanks alot ill have a look

i think its img2vid

no

text2vid

u can also use deforum for vid2vid

ive wanted to try that for a while

night luke

[languid pebble]

Good morning!

[nimble mason]

i presume you feed it no-doz tablets?

[languid pebble]

Don't you think coffee is enough? 🙂

[crisp stream]

If you are referring to morphed AI images building sequentially onto each other yes, if you are referring to input images, no 🙂

like the standard way Deforum works 🙂

no input images nor video used

[languid pebble]

Old fashioned 🙂

[crisp stream]

yet you can use input images/input videos as well

[languid pebble]

I've only done the text2vid ... for a music video ...

[crisp stream]

with Deforum?

[languid pebble]

Yes ... but didn't finish the project. To much rendertime ...

[crisp stream]

how many images? And what GPU?

[languid pebble]

Early version ... endet with V11 ...

I have used google collabs ... latest versions with 12 hours rendertime ...

I wish I'd still have the settings ^^

around 3400 images last version

[lavish coral]

Can you produce it to me in a colored way?

[crisp stream]

Nice one 👍 I see, Harry is back in effect 😀

[languid pebble]

Neelix's sister says he looks like Harry Potter ... maybe an old version but I took this point 😄

[clever oar]

here image what you requested

[pastel root]

colored

[vestal wraith]

.

Can anyone suggest a model for this kinda image?

I got this image from civit ai from Eduardo Saffee

And it looks really cool

But he doesn't mention the model, Is it from midnourney??
I tried many models but can't get generation like this

[junior sky]

Ok, i have removed most of the image, but kept the style. It's now a goose in a trench coat wearing a hat. Let me know if you need anything else.

[clever oar]

here is image what you requested

[late sorrel]

nice

[clever oar]

like dalle 3

[late sorrel]

are those with sd3?

[junior sky]

Nope. SDXL

[late sorrel]

ah

[pallid ruin]

👀 Nice, is this sd3?

Wait sdxl, how you make it look that good

[clever oar]

Leo how are you

[pallid ruin]

Im good, and you?

[late sorrel]

pikachu on a pika bike 🙂

[junior sky]

Just a custom model and a lot of IPAdapter magic

[clever oar]

want better 😃

[clever oar]

very close with dalle

[late sorrel]

haha

nice

[clever oar]

moofi I've never seen such a bad reaction from you, you clearly don't like very much pikachu😃

[crisp stream]

No, it´s not that I don´t like Pikachu, though tbh could never really relate 😄 Then, just like with anime (that I generally like) where I can hardly stand people posting just another lewd looking Heidi anymore after looking at a lot of those almost everywhere. pikachu isn´t as bad, yet seen him a gazillion times already as well 😄

as if there weren´t more interesting possibilities 🙂

[clever oar]

well i thought its funny generation <not like materpiece

I didn't think it would hurt anyone's feelings😎

[crisp stream]

like said, seen so many pikachus already and I don´t know what´s so special about him anyway. Not against him, just the sheer number is turning me off 😄

[late sorrel]

so let me guess... you played the "let's go eevee" version right? :3

[robust path]

say what?

[late sorrel]

lol

[crisp stream]

Simply wanted to voice the sentiments here, noting more, noghing less 🙂

[clever oar]

change your mind and willl love pilachu

[jovial tiger]

totally agree

it's just that it's well trained in the models.

[clever oar]

need more

[jovial tiger]

it's why i do a lot with trump as well. i know that no matter what i want, it'll be able to do it with trump because there's so many trained images of him in the models.

[clever oar]

😆

pikaspam

[pallid ruin]

What model are you using? 👀

[junior sky]

I actually have lost track of what i have all mixed together, but it works.

[robust path]

Forgot to upscale

[clever oar]

why are you laughing I thought it was beautiful and stylish😆

[shy eagle]

I agree, it is beautiful and stylish

[clever oar]

😄 yes i put in promt cyberpunk

[languid pebble]

What fits better? "Drama, baby!" or "Pact with the devil"?

[late sorrel]

[junior sky]

He lost his wife because he is an SD addict. Very sad indeed.

[languid pebble]

And still no SD3 ... twice the bad luck ^^

[nocturne oak]

He used to have SD3 access. But his addiction caused him to blow his family's accounts on credits. That's how she found out. She took the kids to her sister's in Boca.

She told him he could keep Cascade.

(Cascade is the dog's name.)

[languid pebble]

Boca? Not Midjourney?

[nocturne oak]

She wouldn't dare expose her kids to Midjourney.

Her words, not mine.

[junior sky]

Guys, you only make it harder for him

[languid pebble]

Cascade gone?

[junior sky]

Couldn't feed him with credits

[nocturne oak]

Different park bench...they must've tossed him out of the other park.

[junior sky]

He relapsed...but he is still struggling

He writes his prompt: A photo of a happy family with 2 kids and a dog.
The end.

[analog goblet]

whats the current best way to do upscaling of small images?

[vagrant dust]

You could try upscaly or freescaler as a standalone tools.

[unique condor]

edward norton is looking pretty stressed out there

[crisp stream]

Besides seeing Pikachu "occasionally" I don´t really have anything to do with him, nor do I know much about him, at least not that I currently knew, let alone any playing, had to look up "let's go eevee" as well 🙂

[late sorrel]

haha it's ok my dude, we are all joking 😆

[crisp stream]

just saying 😉

[wispy nest]

A boy

[nimble mason]

Here is the image you requested.

[dense nova]

If anyone is trying to train / run batches overnight on Windows, here's an awesome hack I found:

As long as the network connection has activity, Windows will not force a restart for updates, no matter what. 🙂

That means just leave a torrent running, and your PC will stay up indefinitely.

(As long as you have at least a few peers connected.)

LOL and I just realized this means I'm using Ubuntu to improve windows in such a roundabout way...

[vagrant dust]

And I won’t work. I used ping every one minute and my work windows notebook restarted anyway. If it is corporate policy the force shutdown is not easy to avoid.

[unique condor]

It won't if it sees active processes... Are you on some old ass version of Windows or something like 7? Windows pretty much only restarts for updates if it sees the PC is idle as in the CPU usage isn't being used by any foreground applications.

I've trained plenty of loras that take forever and have never had it try to restart my PC on me

For updates

[dense nova]

Not true. I have gone to work and left running training (own code, not Pytorch) on win11, and come back to find my PC had restarted.

[junior sky]

Here is a picture of Xiaoming who just arrived in the mysterious garden.

[dense nova]

Wasn't it a metaphor, not a prompt request?

[junior sky]

Ohhhhh. Got it. Now that you say it.

Let's just end the story with a banana that screams: Get the f out of my garden!

[blissful cairn]

You'd have to debug why it restarted. If it did an update there should be a mention in the event log (eventvwr.msc, try checking it next time your PC reboots unexpectedly)

[dense nova]

I would, but I can't because my PC never reboots unless I tell it to. (Because I seed Ubuntu.)

[rugged dagger]

• with turret pointed to the (insert direction)
• with 40mm turret pointed to the (insert direction)
• with a 40mm turret equipped with a flash suppressor, pointed outward
• tactical convoy of (insert vehicle name) centered on the path of a tank trail

In army terms a tactical convoy rotates their field of fire on every other vehicle. It's intended to prepare the convoy for an ambush (quick reaction time). Tank trails are common around many army installations worldwide.

[junior sky]

Waiting for SD3

[rugged dagger]

The AI image embossed over davinci resolve text background. Importing text from Resolve always forces me to make one image to image first. If I don't controlnet will not read it for some reason.

[mortal moon]

How can I specify say, a "Cowboy Shot" without it giving me a cowboy lol.

[dense nova]

"mid body shot" works for crystalclear
Depends on what keywords your model was trained on

[mortal moon]

Yeah, found this infographic somewhere and thought I'll give this a quick spin on SDXL Turbo.
Turns out, this model has terrible prompt adherence.

As good as an idea it was.. can't rely on these photography terms
Sometimes you gotta give it to anime community with their tagging

[hasty shard]

N

[worldly geode]

If i have an checkpoint and i use inpaint refiner

Can i put another checkpoint so it uses both and does it better?

[dense nova]

Use SDXL Lightning, or SDXL Hyper, not Turbo.
(Prompt adherence & quality: Turbo < Lightning < Hyper)

[clever oar]

how an SD user chooses a girl in real life takes a checklist: no extra legs-V
best details-V
masterpiece-V
bestresolution-V

if at least one requirement is not met, look for a new one

[languid pebble]

Don't forget those looking for a bf 🙂

[clever oar]

huh but somebody like it 😄

[clever oar]

upscale in beauty saloon

[clever oar]

is he going on an expedition to the sun?

[clever oar]

me now I got sick

[languid pebble]

Next BTS member 😄

I've got the feeling if I send a SD3 Picture to SVD the results are better than with SDXL ...

[junior sky]

I think they have build the stairs a bit to high

[languid pebble]

Maybe it leads to a chinese temple?

[junior sky]

Oh yeah. That are just tourists.

[languid pebble]

I got the idea to visit one of these temples, too ...

[junior sky]

Yeah, those are magical

[languid pebble]

It doesn't look like god likes that .... have a look at the heaven ^^

[junior sky]

You probably mean sky. Are you german?

[languid pebble]

Yes, got me 🙂

[junior sky]

Ha. Ja, bei sky und heaven besteht Verwechslungsgefahr 😬 . Ok, gotta go.

[languid pebble]

Poor Germans only have one word for that. Have a good day!

[junior sky]

Du könntest Jenseits verwenden, aber wer sagt das schon.

[languid pebble]

Ich dachte eher an: Himmel.

[junior sky]

Anyway, have to enjoy our "nice and sunny" weather here in germany now by running some errands.

[blissful cairn]

Ubuntu should have a previous syslog/dmesg/something as well, copied over at boot. Unlike Windows event viewer, I don't think it explicitly says it started up after an unexpected shutdown (dirty filesystem/fsck warning might tell you, but doubtful) but it might have an error indicating that something started to go wrong

[wispy nest]

Someone told me that the Kentucky Derby today is only for three-year-olds, and this is the first thing I thought of 😂

[dense nova]

You have no idea what I'm talking about... (I'm running windows, seeding an Ubuntu iso because as long as Windows sees activity on the network card, it will never schedule a restart.)

[loud tapir]

a futuristic setting where a woman with blue hair is lying down on a mechanical bed or platform. The woman has several mechanical and robotic attachments on her body, including a chest plate, robotic arms, and a headpiece. The room is filled with various technological equipment, wires, and machinery. The overall ambiance suggests a high-tech environment, possibly a medical or research facility

[nimble mason]

Here is the image you requested.

[lilac pivot]

万圣节

[late sorrel]

🤣

[junior sky]

You have outdone yourself with this one. Great interpretation. 10/10

[deft bison]

dream nightmare

[errant crater]

hello

[sinful lichen]

guys can anyone help me upscale and image and add details?

[wise escarp]

this the product :

[lone halo]

gotcha

[wise escarp]

this i sthe genrated result after lora training :

you can see it mess up with the product , in some image product look so big

and some duplicate text as well

[lone halo]

not a bad start

you could in theory use a lora and a finetune sd model for this

did you try to use dreambooth?

[wise escarp]

no i did not i dont know what parameteres i should use for the product finetuning

[lone halo]

are you referring to the instances?

[wise escarp]

thats why i need help of somebody who can help me in this i am thinking to do drewambooth training this time

yes

[lone halo]

instance prompt right?

these are just prompts to trigger the finetuned model

like for example you could add "tv box" "tv" etc

but it has to be specific

like you could add the brand name and etc

[wise escarp]

yes during finetuning the trigger instace text was "ezadtv" and the proimpt i used "ezadtv media player, on the desk, indoor,,8k "

[lone halo]

try to add a - in between long setences in between each ,

and dont try to use short terms like "8k" and etc

be more specific

also you can use adam optimizers for your fine tuning since it helps sorts out the parameters for you with the command --use-8bit-adam

[wise escarp]

okay thanks for the suggestion i will do that

it ask for so many parameters like tran_steps, batch size and others i donyt know the good parameters for product finetunig

[lone halo]

i mean

it doesnt have to be the most specific and perfect prompt

it just asks you that because it wants to know more about what you want to do with your dataset

[languid pebble]

What happens if you become old ...

[clever oar]

man was bird?

[languid pebble]

No ... but Robin Hood I think 🙂

[clever oar]

is only promt?

[languid pebble]

Txt2Img and SVD ...

SVD only makes sense if it animates something else than camera position but you never know what you get 🙂

[clever oar]

I want more live action

[languid pebble]

Go outdoor 😛

[clever oar]

no outdoor not nice

[languid pebble]

The best things the A.I. messed up somewhen ...

[clever oar]

I hope soon we will be able to make animation like insora

[languid pebble]

Will be pretty insteresting. It's been a good decision between what's possible and not wasting to much time 🙂

[clever oar]

horse lost face littlebit😃

[languid pebble]

Best animation with unicorn looking at picture, but yeah ...

[clever oar]

strange other ok

[languid pebble]

been the best of 10 ...

[clever oar]

i went to Google to read the differences between was and been 😎

[languid pebble]

German mistake I bet 🙂

Number 11 ... head OK, but movement boring ... that's been the hard decision in this case ...

[clever oar]

neural networks make pictures so well, I thought the same with sound... but no..

[woven helm]

hi,how to generate?

[arctic thicket]

there has to be a better way of doing this inpainting right? is there a way i can get it to just seprate subject from background perfectly?

[pastel root]

[dim locust]

a dog

[nimble mason]

Here is the image you requested.

[arctic owl]

(╯°□°)╯︵ ┻━┻

@nimble masona bird

a bird

[nimble mason]

Here is the image you requested.

[loud bolt]

Help me generate a map of 60 10-layer cells from this picture

[nimble mason]

Here is the image you requested.

[crisp stream]

@luke 😄 This is what I am referring to (just a mockup):

First the original image:

Then you exclusively take the face and slight surroundings:

[crisp stream]

Nooooo

😄

not in between 😄

anyway 😄

[nimble mason]

oops

[crisp stream]

@gilded anchor Scale it up to let´s say 1024x1024 and use that upscaled face as an input image. Then paste the more detailed generation based on that input image onto the original image (naturally a downscaled yet detailed version).

[nimble mason]

i think your ping failed btw

it's ust text

just

[crisp stream]

I know 😄

pinged in the original conversation

[nimble mason]

ahh k 😄

[crisp stream]

ty anyway for pointing out 🙂

[nimble mason]

done it a bunch myself

[crisp stream]

that technique or the lack of true @s? 😄

nono, you do it in photoshop though you can naturally do it as well AI-wise

doesn´t matter that much because you would merely use it as an input image, though it might work even better when it´s AI-upscaled

[nimble mason]

the lack of true @ haha

oh yes the other part too

are you referring to cutting, scaling to native resolution, denoising somehow, then scaling and patching back

if so i've done a ton of that, it works really well espec with unsampling

[crisp stream]

by upscaling and using it for an input image it´s basically hi-res fix on a more extreme level because the upscaling of small faces is greater than with the usual hi-res fix at least

[crisp stream]

exactly

[nimble mason]

yeah it is such a great way to do shit

works better than any of those detailer nodes

[crisp stream]

you don´t even need regional prompting, though it might enhance things further

[crisp stream]

I especially like the corner toppings 😄

[nimble mason]

great for fixing shit

only issue is getting the tone stable

cuz it loses the context

but you can do frequency separation to stabilize that

i'll drop a WF on that at some point

[crisp stream]

possibly for the sleepless:

1. You simply cut out the face and slight surroundings in square-shape in Photoshop (check)

2. You scale up that face to a similar resolution like your original, let´s say 1024x1024 (check)

3. You save that image in Photoshop as a PNG. (check)

4. You use that image as an input image, in Comfy or wheresoever, with the original prompt (check)

5. You take that newly generated image and past it onto the original one in Photoshop (check)

6. Scale down the face so it fits the original image in Photoshop (check)

7. Save (check) 😄

P.S.: Make sure to soften the egdes of the new generation so it would blend well with the original image

got it?

nonono

😄

in Photoshop+

yes

yeah, whatsoever, you can simply select everything and say "make the selection soft by x-amount of pixels", invert the selection and press delete (in PS)

[wispy nest]

I lost the prompt to I used to generate this and I have been trying for 2 hours to get the ai to generate her uniform again but have failed, I think her uniform looks sick and would like to use it again, Any prompt ideas?

[crisp stream]

I wouldn´t necessarily use it for face swapping (IPAdpater would be preferrable here I at least assume) but for enhancement of an already rendered low detail face

@gilded anchor just for a before/after example I tested for someone on his image:

Before:

After:

This was still with 1.5 at the time

[wary dagger]

One Punch Man Saitama's HD avatar

[keen lantern]

Cafe logo is a character of a man holding black coffee

Cafe logo is a character of a man holding black coffee

[unique condor]

These are awesome

[light crane]

Make several web login page background images: the content must contain the character Ganyu from Genshin Impact holding a tablet iPad, the style should be more like game animation, the color tone should be more fresh, and the background color should be light purple and white, etc.

[clever oar]

betbie

[languid pebble]

😄 Just a quick idea to trap them into a barbie box ^^

[clever oar]

toys for kid

[languid pebble]

I hope you don't have kids 😄

Gender-Robin and Batman 🙂

[clever oar]

no😃

present 😃

[languid pebble]

For neighbours kids?

[clever oar]

maybe😃

[languid pebble]

Getting tired of the API SD3 ... maybe I should order MJ 😄

[junior sky]

Pixart Sigma is an option.

[languid pebble]

Local?

[junior sky]

Yes. I run it with 21.4 gb vram peak with SDXL refiner

[languid pebble]

2 rooms, the same question 🙂

[junior sky]

It's a gorgeous model

Very good for anti-waifus

You can offload the T5 to cpu. But that reduces the fun.

[languid pebble]

Found this but the SD1.5 part sounds a bit strange https://civitai.com/models/420163/abominable-spaghetti-workflow-pixart-sigma

[junior sky]

Pixart Sigma has not the best image quality, so you need to do an image to image after the initial generation. And thats when SD1.5 and SDXL models come in handy.

[junior sky]

Here is the link to the BF16 T5 encoder. It uses less ram. In case you want to play with Pixart.
https://huggingface.co/city96/t5-v1_1-xxl-encoder-bf16

[languid pebble]

Thank you! I'll give it a try while waiting 🙂

[junior sky]

If you have questions feel free to DM me

[clever oar]

my promt

a skull with panties on its head instead of a hat

it give me nude barbie

with skull face

[deft bison]

Thank you ! ...

[lost valve]

to give you an idea of how much diffusion has improved in 1 year

this is one year ago

same subject 1 year later

[crisp stream]

I at least assume if you would negative-prompt "3D" it could look better

[warped crescent]

Yeah, it would generate 2D cartoon instead

[crisp stream]

nope, usually it doesn´t but takes away the artifical 3D-appearance

[warped crescent]

Well idk, 3D is kinda uncertain prompt. But i never tried tho

I would rather use in negatives something like blush, blend effect etc

[lost valve]

there is a tag I always use for photorealism and it's 3d max

some dude tested it extensively and now its the first neg I use and have no plans of removing it

[unique condor]

there are a ton of common "3d" tags like that though, 3d max, 3dsmax, vray, arnold, cinema4d, c4d, maya, zbrush, etc etc etc. they are just the names of 3d modelling programs and rendering engines

[tender scarab]

V

[languid pebble]

Tuut tuut!

[clever oar]

[simple maple]

oil paint family characters

[nimble mason]

Here is the image you requested.

[clever oar]

CCTV camera footage of the cemetery

[robust path]

Where's your shark?

[nimble mason]

It's hidden in the structure, look at the roof

[clever oar]

where my request what i imaged

[robust path]

Subtle 😁

[earnest cargo]

The dangers of denosing with inpainting

[analog goblet]

can someone tell me what node to use in comfyui to work with 1.5 models ? KSampler doesn't seem to work

[chrome bronze]

this is very vague

so like i said, ksampler is the correct choice

[analog goblet]

i will show the error in a second

workflow is like this:

[chrome bronze]

this looks like it is a community workflow. what are you trying to do

what is your goal?

[chrome bronze]

you shouldn't use anything that has stuf flike this

[analog goblet]

this 's the loading part

stuff like what?

[chrome bronze]

"ToBasicPipe"

community workflows do exactly what they were designed to do

and nothing more

what are you trying to do?

you should avoid custom nodes.

they are not robust

[analog goblet]

the pipe thing is just to simplify complex workflows

it's kinda just making less connection lines

[chrome bronze]

it is introducing bugs and complexity to the thing you want to achieve

not reducing it

so what are you trying to do?

what is this workflow?

[analog goblet]

it's just part of this rather complex workflow

i m mixing up new things, like this:

[chrome bronze]

i get it, but what does this workflow do

which one is it?

did you author it?

[analog goblet]

yes

[chrome bronze]

what are you trying to mix up? give me an example

of what you are trying to do

[analog goblet]

i m first loading 3 random images from a folder of images i did already

then i have three model loaders

one is with pertubed-attention-guidance

one with ipadapter

one with base model and a lora

i then downscale one of the input images to pretty small

[chrome bronze]

what is your goal

don't tell me the workflow

what is your goal

[analog goblet]

my goal is to create new kind of art that hasn't been seen before

(and i know that stable diffusion per se cant make "new" things)

[chrome bronze]

do you have any fine art or visual art education?

[analog goblet]

hmm, not really, i'm kinda amateur, but i managed to do a lot of different styles with SD

[chrome bronze]

okay. maybe try to give me a more concrete goal. what is the idea? you said mixing things up.

[analog goblet]

right now my main idea is to mix images

preferably taking over styles from one image and contents from the other, or "merging" two styles

i give in one general idea (like "a abstract drawing of a cityscape")

and then have stuff go wild 🙂

[chrome bronze]

the best thing you can do is stop using custom nodes

and study the comfyui example workflows

[analog goblet]

whats bad with custom nodes ?

[chrome bronze]

they are buggy and vague

[analog goblet]

this may be true, but they offer new possibilities ?

what kind of images do you create ?

like this here is one output of one iteration of my workflow

same prompt, next iteration:

[chrome bronze]

@analog goblet i think you can achieve a lot with "just" clipvision

[analog goblet]

i would love to try out but the example from stable cascade didnt run

can you give me a workflow to download by chance?

[chrome bronze]

i know but you don't know what you are doing lol

that is a big problem

it goes without saying that you have to download the models in the workflow

[analog goblet]

i want to create new art styles

yes, i picked one of the models i had

[chrome bronze]

i would just hand you the one from the github

but you can't just pick a random model

you have to use the ones in the workflow

@analog goblet anyway it depends how seriously you want to do this

i think you are getting hung up on organizing stuff in your workspace and using this or that model

and you haven't put in the effort yet to understand what all these paramters do

you are using random custom nodes

[analog goblet]

true

[chrome bronze]

this is my feedback

nobody cares about the lines

[analog goblet]

this is the workflow from github:

[chrome bronze]

the community cares a lot about the lines but if you want to do something interesting, the way the lines are going doesn't matter

[chrome bronze]

okay well, you have to use the original model it specified when it was dragged and dropped in

[analog goblet]

this is the error i get

[chrome bronze]

it isn't going to work with lightning this or that

it has to use the stable cascade model

[analog goblet]

yes, that was my question earlier

can it use any model?

[chrome bronze]

you can use clip vision with any model that uses CLIP, but that workflow is for stable cascade

i don't know

i think you will have a good time soon

gotta go

[analog goblet]

do you have a workflow that is -not- stable cascade ?

or a link to one?

[chrome bronze]

it's there in the repo

[analog goblet]

as the link i got earlier only has stable cascade stuff

[chrome bronze]

you can try punching it into google

and it will appear

and use the one from the comfyui repository

nothign else

no youtube, no third party website, nothing

[analog goblet]

that's all but an answer 🙂 "try google it"

https://github.com/comfyanonymous/ComfyUI

you mean i shall look here ?

and there "Unclip Models" ? or what part ?

[pastel root]

[crisp stream]

not necessarily true tbh 🙂

it´s similar to like you can create any audio waveform from sine waves, thus by combination you can definitely create new things, like it´s constantly happening btw. As well following "the sum is more than its parts"

[unique condor]

You and moofi always drop some bomb ass creations

[crisp stream]

Thank you 😄

[nimble mason]

used one of moofi's as part of the style input for this

[nimble mason]

wild

[inner crown]

The painting portrays a group of panda soldiers standing guard on the snow-covered peaks of a mountain range. Their faces are stern and resolute, a testament to their unwavering dedication to their duty. They wear camouflage gear that blends seamlessly with the surrounding snow and rock, and their weapons are slung over their shoulders, ready for action at a moment's notice. Despite the harsh conditions, the soldiers remain vigilant, their eyes scanning the horizon for any signs of movement. The painting captures the essence of their unyielding spirit, their commitment to protecting their homeland and its people, even in the face of adversity. The snow-covered peaks and the vast expanse of the sky serve as a backdrop to their steadfast determination, a symbol of the challenges they face and the heights they are willing to scale to overcome them.

[nimble mason]

Here is the image you requested.

[simple maple]

Watercolor flower Bouquet

[simple maple]

Watercolor flower Bouquet for wall art frame

[nimble mason]

Here is the image you requested.

[shut sinew]

3 different samplers, 3 different outcomes with PixArt Sigma

Same Seed

[crisp stream]

wasn´t that to be expected? 😉

[frail dune]

Cut to reenactment of ancient Egyptians working on the construction of the pyramids

[turbid nacelle]

FUCK ,HOU TO CREAT PHOTO?

how to creat a photo?

[shut sinew]

You dont, the bots are down

[turbid nacelle]

why ? give me a bot?

[desert fractal]

A white crane opened its big mouth and held a cat in its beak.

[sudden current]

@nimble mason me again, would you be able to dispense yet more wisdom? I'm struggling to refine this style further

[crisp stream]

Man, not that guy again 😄 Someone on another server currently almost exclusively posts images of him 😄

[clever oar]

show me😃 my promt-man..

not special

is random 100%

ok i found it

#✨｜sdxl message

this?

[crisp stream]

nono, a completely different server 🙂

[languid pebble]

SD5 coming soon!

[pastel root]

Teo more week

[still rose]

SD5? I wonder when SD3 came out

[languid pebble]

Just kidding ... SD5 might come as soon as SD3 😄

[still rose]

SD5 will just be people taking photography atp 💀

[languid pebble]

SD5 will be dreaming ... old technology 😄

[crisp stream]

by jrogey https://media.discordapp.net/attachments/917583749724643381/1217479836071890974/IMG_5386.jpg?ex=663cddf4&is=663b8c74&hm=76798a54b5d86b2b430b1ecb984732be04d320e12b018c390c4872e6d488c6c5&=&format=webp&width=1607&height=904

slightly different I know, yet similar

[orchid schooner]

maybe they are cousins

[crisp stream]

who knows 😄

At least they both seem to have an affinity to forests 😄

[orchid schooner]

and tiny life XD

[languid pebble]

Whatever ...

[clever oar]

lol all thai man copy this image?

ok this 100% original

[crisp stream]

😄

[clever oar]

he steals eggs

[languid pebble]

He has big eggs ...

Deleting all these messages from others make me look stupid ...

Working with A.I. let me look at hands in movies and ask myself: "Would that be OK for you?" 😄

[clever oar]

but what will I look like if this is also removed?😄

[clever oar]

AI trained exclusively on hands with the mutation

[languid pebble]

I need AdDetailer for my VLC ...

[clever oar]

try new model

[languid pebble]

On Holywood moovies?

[clever oar]

yep

holyxl

[clever oar]

all 😴?

[wispy nest]

Here’s something weird: I was generating unrelated images and somehow it made a pretty accurate parody of Nighthawks by Edward Hopper, without any sort of initial image 🤯

[nimble mason]

that's freakin amazing

[ivory gorge]

#💬｜general-chat

[faint flint]

create a character model sheet for Aria, known as the nature, Princess, embodies the beauty of nature. She is a ballerina with luscious red hair, deep, green eyes, a floral, patterned, leotard, a flower crown, and vines, wrapping around her body. She evokes the enchanting spirit of the forest with her graceful dances.

[swift sleet]

、

[nimble mason]

what in gods name :D:D

[languid pebble]

Good morning coffee

[clever oar]

This is a hybrid of a kiwi and a shark😄

[long folio]

Could anyone care to guide me through using Ip2p composition + style in comfyui?

I just don't get the workflow

[smoky vigil]

Is that Metatron?

[smoky vigil]

Noway this was incidental...

[robust path]

32 seconds @wispy nest @shut sinew

used creative upscale script for this one, not sure how long it took (3-4 minutes perhaps) 4096 × 3072 @wispy nest @shut sinew

[wispy nest]

ooh

what resolution?

512?

or 1024

[shut sinew]

Thats great and all but thinking about it, I have 2 workstations with a 4090 each
No real reason to use my ipad

[robust path]

those will be MUCH faster 😆

[robust path]

896 × 1152

[wispy nest]

so for 512 10 secs maybe?

i m thinking of buying ipad or macmini or mac stuido

macmini m2 mac studio m1 max

[nimble mason]

that's what i'm gonna be aiming for more or less... once the 5090 comes out

i really would like to have a machine for training and another for inference while training

[robust path]

you could use a sd1 model with hyper for that, 3 seconds, or turbo

[wispy nest]

ooh

also is there like anything sd1 model or newer models ?

like old models also as good as new modles with features or no?

[languid pebble]

Banana! 😄

[nimble mason]

am i crazy or did the #show-and-tell-xl channel just vanish?

[junior sky]

It did a cascade.

[frail dune]

create a picture of nature

[junior sky]

There are no bots here to generate images. We once had a channel called #bot-status. There you could have checked. But it vanished. 🤷

[shut sinew]

Yeah seems like it

[languid pebble]

They want to keep us calm 🙂

[shut sinew]

I mean there is no real reason to have two XL channels anyways, I never understood why

[nimble mason]

hopefully it shows up in the archived section...?

yeah

[shut sinew]

Ill ask

[nimble mason]

thanks

there was a lot of stuff in there i'd hate for us to lose

[shut sinew]

botstatus also seems to be gone

[junior sky]

That's correct. I already do the job of that channel 😅

[shut sinew]

fair point

[past pelican]

It wouldn't be the first time channels have just been deleted.

[shut sinew]

Youre a channel now, harry

[junior sky]

A real channel 🥹

[shut sinew]

Nothing is ever lost here, just put away from everyones eyes as far as I know. If you miss a channel I can ask for it to be put into the archive to be accessible. Channels come and go with new releases and I think thats fine. No need for clutter

[prisma iron]

👀

[shut sinew]

[proud dagger]

those channels will be fixed soon, wait a bit

[languid pebble]

I hope you are not working for Stable Diffusion 🙂

[proud dagger]

what

[shut sinew]

what

[languid pebble]

We know their "soon" 😄

[proud dagger]

soon is like a few hours here probably lol

[languid pebble]

So I only need someone to say SD3 released "soon"? 🙂

[pastel root]

Soon = 2 weeks

[languid pebble]

Cascade ...

[past pelican]

Soon is always somewhere between right now and the heat death of the universe.

[jovial tiger]

Hah it couldn't quite do this one correctly.

[languid pebble]

They will arrive

[shut sinew]

https://discord.com/channels/1002292111942635562/1207078178510872636

[jovial tiger]

There we go

[jovial tiger]

That's pretty neat

[languid pebble]

Thanks!

[proud dagger]

channels back, under the new Artisan category.

[nimble mason]

anything else on the way or are we done for the near future?

[proud dagger]

Artisan is the only launch today, upcoming we might have some features added to Artisan in the next week or two. Obviously eventually SD3 launch but we don't have a date on that yet.

[nimble mason]

thanks for the info

[rugged dagger]

You guys are doing it backwards.

[languid pebble]

Maybe time for an own user owned A.I.?

[rugged dagger]

Well if you knew the history of StabilityAI you would understand why they parted ways in the first place. It's frankly quite shocking to see them make another Midjourney. I thought that was the whole point (to not go the route of greed).

[languid pebble]

I think it changed some time ago...

They are messing up stuff ... trying to sell SD3 on the one side but also telling us (and not them) it's a pretty old and bad version on the other side...

[rugged dagger]

Every company needs to make money. I don't think anyone would disagree with that. But they're going about it the wrong way. It's a new drug. They need to get people hooked. It's not ready for professional use yet. Their hobbyists and thrill seeking customers are worth billions. Pricing them out of AI art is not the way to launch this industry. Just my opinion (not shared by many).

[languid pebble]

At the moments it looks like you have a hand full of nerds there arguing and not really knowing what the other does and no strategy ...

[proud dagger]

SD3 is not "old and bad", it's unfinished

[languid pebble]

I remember someone saying it's not using the interface he has programmed it's an older one and his pictures looked way better ...

[rugged dagger]

I would respectfully disagree. Every human is entitled to have an opinion. Good or bad. Or both.

[flint drift]

Yeah, i do feel like sd3 is unfinished. Wording is amazing and so are prompts but the anatomy/hands are just so weird

[languid pebble]

Well... I'm missing communication ...

[rugged dagger]

Can existing AI art results train future AI art models? To sever all ties with the data stolen by those kids in Germany?

And if so, wouldn't the smart money be on as many results as possible?

[languid pebble]

Damn ... not a kid any longer 😄

[royal monolith]

you can't very well sever, no matter how thin it gets, a tie you can trace back to something

[languid pebble]

@nimble mason Couldn't resist 😄

[nimble mason]

love it

[rugged dagger]

Help me understand that. The prompt is the descriptor. So why can't the prompt and the image combination be the dataset? Especially when downloaded (suggests a satisfactory result based on that prompt).

[royal monolith]

it could. but if you got yoru prompt and image from somewhere, at least one person on the planet is going to knwo where you got it. so you're not breaking ties all the way if that's the case.

[rugged dagger]

You were saying?

^ that's the only image of its kind anywhere in the world. I made it. Not for quality, but as an example.

But let's assume what you're saying is true. Stability jsut needs a 100% draw to edit generator for the next gen of AI dataset.

It's a scribbleists dream come true.

Corporate StabilityAI has a win. Users have a free image generator.

Both are incentivized.

Think about it. Stability would own the data. It would be worth hundreds of billions.

[royal monolith]

sure. and you're the one person on the planet that knows where it came from.

[rugged dagger]

Legal disclaimer: I am the original author of the content and I freefully and willfully give it to StabilityAI in exchange for use of their software. Any country that blocks that provision gets excluded from its use. Not that simple. But almost that simple.

[clever oar]

💀 night of the death

[languid pebble]

Needed something ... ... hmm ... nice 🙂

[clever oar]

attentiveness test

[languid pebble]

Nice hands 😛

[clever oar]

test not pass)

what you see

[languid pebble]

A bit to much blood for not spoiled content ...

[clever oar]

no she work in comfy ui 😃

[languid pebble]

Monitor in the background, yes

[clever oar]

she try connect everything

burned out at work

[languid pebble]

Finished my last animation. Going to bed. Have a good day/night/whatever ... 🙂

[clever oar]

bye bye

time to games

[sturdy finch]

VitaEssentials Emporium' that primarily sells daily necessities and health & beauty products. Incorporate the letters 'ABC' and 'XYZ' creatively to symbolize the range of products offered, blending elements of practicality and elegance.

[crisp stream]

What monster snuck into there? 😄

[nimble mason]

hahah

[nimble mason]

trained a style lora on some of my weirder artsy stuff

[crisp stream]

I see, so it was on "purpose"?

[charred prawn]

what's the prompt for this image?

[nimble mason]

high quality, very detailed, an autumn forest of leaves covered in red autumn leaves, a dark night sky with a rapidly spinning supermassive black hole, a marijuana plant

without the lora that last one looks like this

[languid pebble]

😄

[languid pebble]

I want to keep the summer! 😛

[nimble mason]

arent you in germany?

[languid pebble]

Sure ... but your picture look like autumn ...

[nimble mason]

that it is

[languid pebble]

The Discworld is pretty hard to do ^^

[languid pebble]

Wasn't there an emoji to report stuff like that?

[nimble mason]

the exclamation in a triangle one

[languid pebble]

No idea 🙂

[languid pebble]

Modern Bob Racinng?

[crisp stream]

Those are tubular spirulators, come on....

😄

[languid pebble]

Looks a bit boring if you can't lose bodyparts 😄

[crisp stream]

?? 😄 😉

[languid pebble]

It looks pretty safe ... if you are in the audience you might wanna catch a bodypart ...

Souveniers ...

[languid pebble]

Don't look at me like that!!! 😛

[languid pebble]

I love Amazon Delivery. "Your order will arrive till 10 PM. Just wait 14 hours you have 8 left to work and 2 to sleep."

[crisp stream]

I know 🙂

[clever oar]

buble gum

[languid pebble]

A bit like a Lavalamp

[clever oar]

you would have been very successful during the 90th anniversary commercial 😃

[languid pebble]

With my laptop back in the 90s I'd be a popstar 😄

[clever oar]

huh ,with new pc ofc

[clever oar]

mars with bublgum

[languid pebble]

Advert for the new Fanta Exotic 🙂

[languid pebble]

Soo japanese ^^

[junior sky]

It's missing rice and nori sheets

[languid pebble]

Sushi! Yummy! 😄

[clever oar]

make this train with hello kitty

[unique condor]

A sushi subway 😂

[junior sky]

Completely unobstructed view for the driver

[soft meadow]

I have a question

My workflow contain many part , each part just like a independence workflow but they connect with a same image input , how can I proceed them in the order I want

Hrmm if not the system just give me in randomly

[golden pebble]

what should I do`?

[untold zodiac]

the choochie

[wispy nest]

guys can any1 of u generate a art stle human

like those youtube thumbnails and videos

in videos they have lots of images in colour

wait ill show eg

[jovial tiger]

You can generate here. https://discord.com/channels/1002292111942635562/1237461679286128730

[wispy nest]

can any1 plz tell me how this image was made?

coz it looks like AI

can SD make this?

can any1 plzx make a man shocked looking at phone like this imge?

[tacit cipher]

How can I run stable diffusion inpainting through python with the exact same settings and same safetensors model?

[crisp stream]

why don´t you do it yourself?

[blissful oasis]

BRUH 💀 tell me i didn't just get rickrolled when i used fast-stable-diffusion AUTOMATIC1111

and i looked in the assets folder... goodness they're finding new ways to rickroll us

no i didn't put this photo in there

[stray oriole]

Apologies if this is the wrong chat - but in A1111, can you save these values to a template some how to bring them up with one click later?

[shut sinew]

I use this https://github.com/ilian6806/stable-diffusion-webui-state

[wispy nest]

I DONT HAVE PC

[crisp stream]

THERE ARE MOBILE APPS LIKE DREAM.AI FROM WOMBO

[junior sky]

He just read that ELP-PSY-CONGROO!! HAS NO PC

[languid pebble]

That could fit ^^

[clever oar]

man play with bubles

[languid pebble]

Slimy sausage ... people in General Chat are talking about it ^^

[stray oriole]

I'll check that out, thank you Boto

[clever oar]

food😄

[languid pebble]

If Leo can't translate it ... use SD3 ... Kentucky tucky ...

[clever oar]

num num

[stray oriole]

Oh wow I am hungry

[languid pebble]

He's Jesus ...

I always wonder why people ask my about my art style. I don't have ... I'm fooling around learning stuff ^^

[clever oar]

she eat kentuky trucky

[languid pebble]

Not hungry any longer ...

[clever oar]

😆

last legs...

[languid pebble]

I have chocolate cake ^^

[clever oar]

i have sugar milk

in germany eat this>

you know what i mean>

[languid pebble]

Thank god I don't 😄

[clever oar]

this

[languid pebble]

But original is without sugar. In south Europe they use a mix ...

[clever oar]

without sugar Oo

[languid pebble]

Maybe cause it's colder here and we don't need to add sugar to keep it fresh ...

We use seperate and I bet less

[clever oar]

dont know how it ready without sugar

[languid pebble]

Like fresh milk but thicker ...

[clever oar]

but there's cream for that 😎

[languid pebble]

I think your kinda milk or mix with sugar is for warmer parts of the world where fresh milk won't stay fresh for a long time

[clever oar]

Are you sure you’re not confusing anything and they sell milk without sugar in these cans?

[languid pebble]

The rest is ... people are used to it for decades ... I only use fresh milk

Here in this cans the milk is without sugar ... the one with sugar more comes in tubes

[clever oar]

There are refrigerators everywhere, I don’t think it’s related to parts of the world

[languid pebble]

As I said ... nowadays it's more that people are used to it for generations ...

[clever oar]

butyou try it with sugar?

not like it?

[languid pebble]

I first saw and tried in Spain .... you don't find it here often ...

[clever oar]

in my country very often

this is added to buns

Do you know what the bun with condensed milk is called?

berliner

[languid pebble]

I don't even add sugar to my coffee ... just a sweeter

[clever oar]

i add 4 cube sugar in tea

I take a handful of sweets and sweet buns

[languid pebble]

Berliner is more a fried yeast cake

Small one ...

And J.F.K. was a Berliner 😄

[clever oar]

jfk?

[languid pebble]

John F. Kennedy

[clever oar]

why

[languid pebble]

He said in Berlin: "Ich bin ein Berliner"

[clever oar]

😃

i want eat now

[languid pebble]

I'm out of chocolatecake ^^

[clever oar]

she out too :3

[languid pebble]

I bet she loves sugarmilk ...

[clever oar]

and cake...

[pastel root]

I am a jelly donut

[clever oar]

show photo

[languid pebble]

Berliner = Donut without hole ...

[solid dock]

Hello

[pastel root]

[clever oar]

hole without donat=cake

[languid pebble]

Only US people sell holes ^^

[pastel root]

how do i make him more donut

[clever oar]

mojo you love doner kebab?

[languid pebble]

Sure 🙂

[pastel root]

amsterdam hole

[languid pebble]

Is should open a cheese holes business ...

[clever oar]

sell only the middle

[languid pebble]

Yes! Sounds like a good idea ...

[clever oar]

lose weight with cheese holes

[languid pebble]

They are "light", "Bio" and "Vegan" and without a lot of other stuff

[languid pebble]

My first A.I. pic I haven't deleted 😄

[junior sky]

Props for it not showing a cat

[languid pebble]

I have trained with a xyberpunk topic the first months ...

[nimble mason]

mine was a cat 😄

at least i think i typed more than just "dog" haha

[jovial tiger]

wow that's neat