🏞｜general-with-images

[smoky oak]

I have not been able to figure out if it upscales the original input, or the output, but I do know that things are way softer when you turn it off

[stark vine]

this is objectively worse without canny

[smoky oak]

is it?

[smoky oak]

damn

it was better for me

what denoise?

[stark vine]

well u know what, it can just be luck of the draw

could be just random

[smoky oak]

wait, what is your denoise

[stark vine]

.55

[smoky oak]

that looks like artifacts from too highd enoise

[stark vine]

trolol

[smoky oak]

oh thats wayyyy too high bro

[stark vine]

no no

I have a method

[smoky oak]

no wonder its making things all weird

[stark vine]

ima showcase it tomorrow

[smoky oak]

alright man, if you say so

[stark vine]

at least I know controlnet just doesnt work

so thats one less variable to deal with

[static tusk]

.55 will guarantee to have unintended noise

[stark vine]

the censored images that I showed earlier

those were done with .80 lmfao.

[smoky oak]

they have a process, so I will just let them do their thing

we seem to be parting ways now in our methods, which is good :>

[stark vine]

Im just clever 😏

nah jk

idk how viable it is

[smoky oak]

we have two people venturing into different things

[stark vine]

I just know for a fact there has to be photoshop involved

[static tusk]

and the image youre working on now isnt perfect is it?

[stark vine]

u just cant click a button and have details in the image at high res

no but thats cuz I just did it quickly

tomorrow ima try something big

I might have to get blender involved for poses too

which sucks cuz I dont want to have to learn blender lol

[smoky oak]

there are easy online tools, and also a 3D poser you can download into SD

its called posex

[static tusk]

The highest denoise I use is .35 and Ive upscaled hundreds perfectly.

[stark vine]

Oh yeh I tried the web browser one but I want more control

[stark vine]

for landscapes yes

but low denoise = mannequin skin

[smoky oak]

fair enough

[stark vine]

which is my biggest pet peeve with 1.5

[smoky oak]

I haven't had that problem, personally

[static tusk]

no, all subject matter. people, cars, everything

[stark vine]

well I havent seen an example yet but then again I might just have a picky eye

[smoky oak]

Pure fire and I have a lot more experience with this, but I have little experience with people, so who knows

[stark vine]

yeh im the portrait freak lol

[smoky oak]

I know I have done 1000's of test gens

[stark vine]

so I can tell when something is upscaled and when something is legit detail

I can always add skin textures in photoshop but its time consuming

[smoky oak]

I mean, I am a professional portrait photographer lol. or well, professional as in it is my only real source of income

but I have been doing it for years, if that counts lol

[static tusk]

This image I posted was half res of the original upscale. #1072010753462251521 message

[smoky oak]

6 now, maybe 7

[smoky oak]

that does look quite nice

[stark vine]

cars dont have tiny details like faces do

[smoky oak]

did you ever get to see my absurdly high res upscale I did?

@static tusk

sorry for the ping, meant to reply

[static tusk]

yeah saw that one

[smoky oak]

15360x6548 with basically pixel level detail 😩

was a lot of work, and it was not perfect, but it WAS worth it lol

[stark vine]

r0fl

after even 4k you cant even see that amount of detail

but hey, its ur time lol

[smoky oak]

I upscaled it that high just to see, and also cause it is so high res, I can get several different high res views out of the same image for different devices

[stark vine]

that makes sense

[smoky oak]

These 3 images (and way more) are all from the same extreme res upscale

[smoky oak]

and they are still 1440p+ each

[smoky oak]

granted, there are seam issues (the part that wasn't perfect and I learned from lol)

@stark vineOMG-

I followed what fire said with the lady gaga portrait, and the one that got rid of what i was saying (like he said) looks better-

like, considerably

I did not mean to send that, but it shall stay lol

ok, but cutting down the negatives did make it worse

original prompt

quality only

cut down negatives

trying no prompt

MY TAB CRASHED

ok, in order

Full positive/negative, cut pos/full neg, cut pos/neg, no pos/full neg, no pos/neg @stark vine

Cut pos full neg seems to be the best-

all identical settings, outside of prompt

[stark vine]

Wait what?

Are those still with controlnet?

I'm on my phone now I'll have to check them tomorrow

[smoky oak]

No, I turned off control net

I wanted to see if what pure fire was saying was true, and it seems as though it is

I've tested it on a few different things

[hidden walrus]

ds-image

prompt

/imaginerobot

[static tusk]

discord bots are offline

[twin falcon]

Apple

[glossy herald]

nah, that was a girl

but you were looking for the bot maybe

[crystal delta]

@twin falcon

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[twin falcon]

/lora blonde hair

/prompt Apple

[glossy herald]

Hey, I just linked you the FAQ saying we don't have a bot doing that currently

sorry but that won't work with any / you could try right now

[dense tapir]

No matter what I try in the neg I just can't get rid of that fog/haze

[twin falcon]

Sad , but thanks for helping

[static tusk]

tried volumetric lighting as neg?

[outer spear]

What is the server icon?

Is it a silicon wafer

or what

[glossy herald]

https://www.reddit.com/r/StableDiffusion/comments/11z7rrg/instructnerf2nerf

Quite impressive

Not sure at all about what the server icon is lol

[static tusk]

its an omegalul 😆

[wispy nest]

What exactly is a nerf studio?

or NERFs at this point

[smoky oak]

Oh nice! NERFS are cool

[smoky oak]

It's a research paper that allows you to create 3D photos like the example above, but it only requires a couple photos. I believe they are so advanced now that you only require one

[wispy nest]

hm, what can you do with them?

It's not an actual 3d model , I'm guessing I can use anywhere else?

[gray citrus]

Since i reinstalled my A1111 on a new computer (using a 4080) a lot of my images come out desaturated. Any ideas why? This is using ReV Animated 1.1

[static tusk]

desat gens are always VAE related. check you have the right one for the model, is my guess.

[gray citrus]

yes, that seems to be the case. I copied over the vae from InvokeAI and now it looks a lot better. But i didn't have these problems with past A1111 installs. Now i often get images that are either desaturated or overexposed

[static tusk]

I havent experienced that, so not really sure why youre getting overexposed as well.

[static tusk]

BTW, I enjoy watching your videos. Keep up the good work

[gray citrus]

thank you 🙂 feel free to write me if you see some new, cool tricks or things i should cover

[wispy nest]

Nice to see you there

[white pivot]

Best movies in the world

[dense tapir]

You know, hmmmm. Let me try that.

[dense tapir]

20 ton kitty on the prowl, now that had me laughing.

[dense tapir]

second is with volumetric lighting as a neg.

[dense tapir]

@static tuskI thought it might be seed related, but apparently not.

[random rivet]

a high quality warrior woman at night with rain, feelings

@random rivet

[dense tapir]

no image bots on this server @random rivet

[crystal delta]

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[stoic pivot]

My first Embedding based on someone else style, I used Brad Colbow style, you can almost see it is like his style but not even close to the original quality, if I self can say I think my embedding sucks.

[glossy herald]

this is your FIRST embedding. and it managed to capture the style. how dare you say it sucks ? it was a great learning experience, and it works mostly ! well done 🙂

[stoic pivot]

It not my first, it is my second one, the first I made was on my own style.

[stoic pivot]

I will not publish it, I would not feel good doing so, I hope I can use it to just get better outlines on my work.

[glossy herald]

usually, using a lower learning rate helps me capture more details, at the cost of the speed of the training

so maybe adding some steps to your last version on a 10 to 100 times lower LR could help bring some of the details in ?

[stoic pivot]

I am so sorry for this. Father, forgive him, for he have not known what he do", it is a combination of two styles and that scare me 😄

[upper pelican]

what model are you using for this? looks very real

[stark vine]

lmfao

yeh I just tried this and the good news is that it works, cut pos full negs

the only tiny little itty bitty pwoblem is that it no longer looks like my girl l0l.

bruh we have optimized this tool so much lmao

it will have to do until we get a 1024 I guess, which probably isn't anytime soon

dog I just thought of something crazy

so crazy idk if I should share my idea

here's the result that I got with this technique which is impressive

but still bereft of fine detail unlike the censored ones I posted earlier, although it is hella consistent

U know what fuck it. Ima share my idea cuz I aint a lil bih. and if someone snatches the idea before me then that's even better cuz it'll save me a lot of work

You know how this extension chops up the image into squares and processes them individually right?

What if instead of using a single prompt for the entire image, we throw in clip interrogator into each square and let it decide what the image is?

So we introduce a universal prompt which would be "hi-res, masterpiece, detailed, etc" and then mix it with the clip interrogator prompt for each square

So if it's just doing an eye and eyebrow the clip interrogator would do just for that square "photo of green eye and eyebrow, hi-res, masterpiece, detailed" etc

I'm gonna start working on this project and if anyone knows how to integrate clip interrogator into a prompt ur welcome to jump in. https://github.com/SenTheNastyDynast?tab=repositories

[smoky oak]

Welcome to the problem of high denoise. You're purposely setting it to change more than it keeps by using a denoise over .3

[dry crow]

@wispy nest Imagedata:

steampunk cat, anime style, copper and metal, intricate details, gears, clockwork, 4k, masterpiece, highres, highly detailed,
Negative prompt: comic, blurry, deformed hands, deformed face, deformed fingers, multiple limbs, ugly, bad anatomy, bad hands, ugly hands, deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limbs, ugly, poorly drawn hands, (grayscale), weird colors, censored, deformed glasses, lowres, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username,
Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 2948040081, Size: 512x768, Model: Dreamshaper_332BakedVaeClipFix, Model hash: 637d5dcb91, Denoising strength: 0.3, Hires upscale: 2, Hires steps: 10, Hires upscaler: R-ESRGAN 4x+ Anime6B

[smoky oak]

@dense tapir Nvidia just dropped a new GPU you may be interested in

Not sure of the price, but it seems awesome

Ultra low power RTX A GPU, with 20GB VRAM

[wispy nest]

@dry crow i guess im good?

[smoky oak]

It's only 70 watts and delivers 3070ti performance

[dense tapir]

Wonder what it is?

[smoky oak]

Nevermind, it's $1250 ._.

It's the Nvidia RTX 4000 SFF

[dry crow]

its not the same but yea works 😄

[dense tapir]

Oh, I just saw that mentioned

small form factor

[wispy nest]

alr yeah i got 331 baked and some diff things but it seems like everything is fine

[dense tapir]

it isn't gaming though

which intrigues me

[smoky oak]

It is a workstation card, so the price does make more sense, considering it has Quadro drivers and is focused on computation

[dense tapir]

Precisely

[smoky oak]

It does have 20GB ECC VRAM which is pretty nice

[dense tapir]

I used to use solidworks and on a geforce it was iffy and somethings locked out but stick a quadro in it and joy

[smoky oak]

But at the same time... That is very close to used 3090

[dense tapir]

See, tbh, I would buy that before a 4080

[smoky oak]

I wonder how many tensor cores it has, just a sec

It is ada lovelace

[dense tapir]

not even counting 4 gigs more so I wonder if this means a 4080 price reduction?

[smoky oak]

Not likely as this card is massively less powerful. It has much less CUDA, cause it's focused on ray rendering and AI workloads over cuda

[dense tapir]

thing is AI loves CUDA so I am confused

[smoky oak]

Oh, it is standard GDDR6 VRAm

That is a very interesting choice...

[dense tapir]

I can see why though

way more stable than 6x and unknown about the upcoming changes to 7

cheaper too

[smoky oak]

It has 192 4th gen tensor cores, which bodes very well

[dense tapir]

tc is king

Even on a 4090, or A100 the TC waits on the ram to hand it more data 50% of the time cause it is just that fast.

[smoky oak]

Ok, so it is pretty beefy. It has 6.2k Cuda cores as well. It's just not competing with cuda on a 4080. But the drivers it comes with probably help it massively

I think this card is actually very interesting as a release

[dense tapir]

more ram, ecc, costs about the same, and uses far less energy. hmmm

[smoky oak]

It's an exceptionally low power card that seems to be pretty damn strong

It's so low power it doesn't even need a power connector

[dense tapir]

Oh, wow like the A2000

[smoky oak]

Yeah, it runs off of PCIE only

[dense tapir]

75w then

[smoky oak]

Single slot too

70w in this case, yeah

[dense tapir]

yeah, A2000 size

[smoky oak]

Very very interesting release

[dense tapir]

75w is max on the pci-e header so they are playing it safe. no o/c though

might can O/C it, as I do mine, where you limit the power to it but up the clocks

I am 85% power and +160 clk/ram

WHY, can't I get rid of this fog?

it plagues me

[smoky oak]

Ohhhhh, I see where it cuts the magic lmao

[dense tapir]

where?

[smoky oak]

Current estimates are showing it's similar to a 3050 in cuda lmao

[dense tapir]

that is actually fine for any non gamer

that card is actually pretty sweet

[smoky oak]

Or sorry, I read it erong

Yeah, it's 3070~

[dense tapir]

even better

[smoky oak]

Their graph is stupid

Lower perf is higher lol

[dense tapir]

backwards

A 3070 in performance then with ecc mem, 20 gigs, and only 70w. Interesting

I know on the A2000 I saw a vid review with games and you could see it wanted to do more but still stood up for what it was in gaming it wasn't even made for.

limiting factor was the cap of 1100mhz I believe it was

I highly suspect some weirdness is coming like the same weird stuff on the 3k cards

when a 3060 had more ram than a 3070 kind of weirdness

I see something bad

160 bit bus with 20gb ram making it 320gb/s

if that were 192 it would have been good 256+ great

[grizzled sage]

I've been thinking about systems for multiprompting multiple images. I think there's a lot of potential for prompting different sections of the image differently. there is ones that let you paint each prompt in. There are ones that do a batch of images all at once, blending the seams together, and that can be combined with different prompts on each image in the batch.

I think a big thing here is the UX. Some way to feed multiple prompts in needs to be standardized. Multiprompting would do well for so many purposes outside of upscaling too. Node based workflows i think will shine best here

[wispy nest]

yea...we already have comfyUI which can do multiple prompts on different image sections, there's an implementation on different prompts for different gens on TPU , probably should be something for GPU on normal SD too, maybe someone will make it less janky

[hardy tulip]

this is why you should always edit the colors manually

[grizzled sage]

any digital artist taking this gig seriously shouldn't just fire off a prompt/seed and call it a day

[hard wagon]

thanks for the feature, @scenic atlas! 🙂
https://stabledigest.substack.com/p/stable-digest-3

[hardy tulip]

post-processing is near necessary

[dense tapir]

Yes

Even a simple burn tool in PS makes it just pop

[whole apex]

What model is this?

[sterile kiln]

parameters: RAW Photo, Montpellier, France, cityscape, historic buildings, narrow streets, outdoor cafes, sunny day, trending on ArtStation, trending on CGSociety, Intricate, High Detail, Sharp focus, dramatic, realistic photography

Negative Prompt: cartoon, 3d, ((disfigured)), ((bad art)), ((deformed)),((extra limbs)),((close up)),((b&w)), wierd colors, blurry, (((duplicate))), ((morbid)), ((mutilated)), out of frame, extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), Photoshop, video game, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, 3d render, freckles, watermark, hat, gun firing

Steps: 50,
Sampler: Euler a,
CFG scale: 13,
Seed: 1403634710,
Size: 910x512,
Model hash: c35782bad8,
Denoising strength: 0.2, Tiled Diffusion tile width: 96, Tiled Diffusion tile height: 96, Tiled Diffusion overlap: 48, Hires upscale: 2.4, Hires steps: 60, Hires upscaler: ESRGAN 4x

[whole apex]

Oh thanks~

[austere pilot]

https://colab.research.google.com/drive/1gYiwvvmAhjg1nDNjN8kv6L9eOsbP-9w-?usp=sharing#scrollTo=13zvMZH-tpsX

anyone that wants to run this img through this google collab? I don't have collab pro and my credits are up

[dry crow]

Do you have the vae file for that model ?

[hardy tulip]

wdym vae?

[dry crow]

Anime models need a vae.pt file for color and Detail correction

[hardy tulip]

uuh

i dont have

didn't even know that existed

[grizzled sage]

a lot of models come with vae's baked into them too. so by default they just look better and people use that instead

[dry crow]

Why do i cant post the image

Or do you see it ?

[hardy tulip]

i see

[grizzled sage]

some of the anything v3 remerges do that

[dry crow]

Oh ok

[grizzled sage]

i see. discords just puking

[dry crow]

Ahhh

Yea thats what vaes do

For Anime models you need one. Most are based on the AnythingV3 vae

Not Anime models mostly have the vae intgrated

[wispy nest]

Where is the best place to find these models? I have the websui setup

[smoky oak]

Civitai is what I use

I think it's what most people use

[dry crow]

@hardy tulip whats your model? I maybe can link you a vae to try

[hardy tulip]

anythingv4.5

but i dont think i liked how they turn out

i think i prefer adjusting manually

[wispy nest]

what are vaes?

[wispy nest]

thank you very much.

[dry crow]

You could adjust it after that too, but the image Quality overall gets better with a vae

[smoky oak]

No problem, best of luck

[dry crow]

Here is it for v4.5:
https://huggingface.co/andite/anything-v4.0/blob/main/anything-v4.0.vae.pt
Has to be renamed to match the 4.5 model

[smoky oak]

A VAE is basically like a color processing algorithm that rebalances the image

[dry crow]

Vae files are mostly needed for anime models for color and Detail correction

[grizzled sage]

i think the need for the custom vae comes from merging novelai's model that used theirs

well it stems from that

[dry crow]

Thats right

Every merge needs a vae too if the merge model has needed one

[grizzled sage]

over time people got new vaes to fix the over merging of anything v3 into models. it was sort of a gene correction lol

[wispy nest]

Wow, thanks!

[smoky oak]

No problem :>

They can make a monumental difference

Huge difference without and with

[grizzled sage]

one of the pixel models has a custom vae and other scripts tied to it too, so the outputs are diffused into pixel perfect alignments

https://github.com/devilismyfriend/PXL8 here's his script side of things

[smoky oak]

@wispy nest

[wispy nest]

hello!

alright so I kinda get the deal, its like telling an artists to use these reference images

[smoky oak]

Kinda, yeah

You can see some examples of LoRA's I have made

Original character

[wispy nest]

damn, thats sick

[smoky oak]

And this is the LoRA

[wispy nest]

so, I just drop the .safetensors file in the models/Lora directory, and use the trigger word?

[wispy nest]

wait you drew this 😲

[smoky oak]

No

That's the character I trained on

[wispy nest]

Oh, still, amazing result

[smoky oak]

I also am working on a LoRA for the Na'vi from avatar

[wispy nest]

I keep getting a "couldnt find" error

[smoky oak]

You have to add it to the prompt, just a sec

These are some of my early results

[wispy nest]

the future is incredible 🤩

[smoky oak]

@wispy nest so you wanna find this in a1111

[wispy nest]

a1111?

[smoky oak]

You click that and it will open all of your additional networks. Then you can find the tab for LoRA, And if you click on it it will insert it into the prompt

[wispy nest]

ah I see them!

[grizzled sage]

yours is red. mine's magenta

[wispy nest]

thank you

[smoky oak]

It should insert it like LoRA:NAME:1

The value after the color is the strength. 1 is 100 percent, .9 is 90, .8 is 80, so on

Some of them require high values, some really low, it's all experimentation

Oh, also made a LoRA for toothless lol

[wispy nest]

Oh my goodness, this is INCREDIBLE

[smoky oak]

Articles like this make me laugh, cause these people clearly know nothing about AI generations if they think that hands have been an issue up until very recently

[sterile kiln]

And I did a LoRA for Jacques Chirac, alias Supermenteur

[smoky oak]

More examples from my Na'vi LoRA early tests

It's been paused for a while now after somebody here stole my thunder by directly copying what I am doing

All I have to say is, if you're gonna copy my ideas, at least make sure they look better than my results lol

[dense tapir]

Did you see Diablo IV beta is doing what Amazon Open world mmo did to the 3090? Bricking some of them.

[smoky oak]

Ono

3090's can never get a break lmao

[dense tapir]

3090 and 4090 stay away until fixed

[sterile kiln]

@smoky oak how many images and what settings did you use to do good LoRAs like this?

[dense tapir]

Be my luck for a 2k card and BAM

[smoky oak]

For that early test, it was a 10 image data set, and I'm not comfortable with sharing my settings as of now, cause I have a feeling some people here are gonna take them and try to steal credit

Not looking at anybody in particular... For now

ALSO

Just wanted to say, not AI, but this is insanely impressive to me lmao

[dense tapir]

They will, 100%

[smoky oak]

Art of an Eye made from only emojis lol

[dense tapir]

damn

[smoky oak]

Shockingly high fidelity haha

If you get the screen for another way from your face, it just looks like a really stylistic and neat painting

[sterile kiln]

🥴 , I was just asking you for your LoRA generation settings, not your model (lmao).
I have my own inspirations, I don't need anyone x)

[smoky oak]

What do you mean generation settings?

@sterile kiln

[sterile kiln]

your LoRA train settings, sry

[smoky oak]

Yeah, those are what I'm not comfortable giving out right now, I'm sorry

Giving out my settings means that people who want to undermine me can do it at a higher quality. At least so far the person that likes mimicking my stuff isn't too great at it

A cheap imitation, as most would say lol

[stark vine]

Bro, you need to chill with worrying about people copying you

The AI community it's all about freedom openness and generosity

[smoky oak]

Do not tell me how to feel. Thanks.

[stark vine]

Being a selfish grump makes you no better than midjourney Devs

[wispy nest]

spiderman in china

[smoky oak]

I have given more back to this community than most people here have combined. I can be selfish when I feel like it. And for me that means not sacrificing my own personal enjoyment to cheap imitations

[crystal delta]

@wispy nest

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

[glossy herald]

hey, not again guys, please

[cursive lance]

mi art is spooky

[smoky oak]

Sorry, I did the think a general statement like that was a problem

[glossy herald]

no problem, I'm just saying, let's not follow that same path again x)

[smoky oak]

Understandable. I've been working really well with them, and we even did some collaborative efforts last night. However if they keep telling me what to do/not do, I'll have to block them

Just for my own sanity

Anyways, moving in from that

[cursive lance]

making something simple is difficult xd

[smoky oak]

Those have charm :>

I like it :>

@glossy herald

Love that meme lol

[smoky oak]

I am actually kinda obcessed with these spindly little guys

Please, feel free to share more haha

Just heard it's national puppy day on the radio. Pupper gens inbound from me

Some fluffer puppers

@glossy herald not sure if you are a cat only day, but apparently it's national puppy day lol

[glossy herald]

this is my puppy ❤️

[smoky oak]

Dawwww

Uno moment

Mi puppi

He's a little goof

[glossy herald]

I can't say anything but whoooo too x)

[smoky oak]

Haha

[sterile kiln]

Bro, finnally i don't care your stuff.
Stop (or start?) smoke weed

[smoky oak]

What?

[sterile kiln]

Save your settings, i'm not interest

[smoky oak]

Alright then. It's not that big of a deal. Best of luck to you

[sterile kiln]

and I wish you to have a real reflection on opensource, github, the people who made all these models. Because your talk about your settings.... There are 6 sliders max and 2 or 3 settings, you didn't invent nuclear fission either.

[smoky oak]

Bro, it's really not that serious. Please back off. If I don't want to give out my stuff, then I don't have to. I give out a ton of stuff, but I don't owe anybody here anything

[sterile kiln]

Everyone doesn't care about your settings, it was just a small element of comparison, the problem is your big pretentious mentality, down 10 floors

[smoky oak]

I do not understand why you are attacking me for exercising caution with giving out details

[glossy herald]

it's not really a problem to want to keep some settings to ourselves, no obligation to share all either.

no need for going personal at them, they aren't the first one here that do want to keep some of that garden secret

[smoky oak]

It's not even keeping it secret. I am still working on the tutorial to do it, and I'm working on a huge project that relies on these settings. Of which somebody has blatantly tried to steal the idea of. I'm trying to keep my settings out of the hands of people who are looking to maliciously undermine people.

If anything, that sentiment is about as open source as it gets. Betterment of the people

[sterile kiln]

Bref

The earth won't stop spinning tonight. Good evening

[glossy herald]

have a good evening too Seb 🙂

[smoky oak]

Good night

@glossy herald you got games on your phone? Lol

He's so nosseeeyyy

[glossy herald]

hehe

mine doesn't want to let me touch the phone when I'm with him, at all

[smoky oak]

My little goober is a sniff-freak

His nose leads him everywhere. EVERYWHERE

[onyx anvil]

Hi there everyone, just dropping by to say hi and that I just now started experimenting with Stable Diffusion. Wooow I'm slowly becoming a nerd

[smoky oak]

It's a very fast plunge into nerdage haha

Kiss your social life goodbye! It is all text and images from here on :p

[glossy herald]

welcome around here !

[onyx anvil]

Thanks a lot!! I now have a vision of creating some kind of abstract outcome including my both grandparents in a weird context, hopefully they will frame it and put it by the bedside 🙂 Also hopefully it is doable.. I'm thinking of them both riding upsized insects as vikings in the sunset or something...
Can Stable Diffusion get inputs from more than one image to better learn how to produce something creative to one new image (with crazy prompts)?

[smoky oak]

I know there are methods of blending songs, however I don't know them personally

Songs? Bruh lol

*images

[glossy herald]

well well well. the answer is yes, Stable diffusion can get multiple input image and learn from them.
but,and it's a quite big but, no pun
it's a whole other subject. You use SD to make picture, or you train SD to be able to make new pictures, 2 different tools and skilsets to learn.
Trying to first master making picture with less constraint would be best, that's my tip

[smoky oak]

I got that monkey music brain lol

[glossy herald]

I wrote an "intro to training stable diffusion" if you want, but tbh I don't want to scare you, this is a big tech to tame

[onyx anvil]

Understood! Here we goooooo (I'll be back in like 3 months) 😉
They will for sure have 6 or 7 fingers, that I know for sure because all I created until now is very fingery

[glossy herald]

you can already try some extensions like controlnet, that solves that finger problem in some cases 😉

[onyx anvil]

I'd love it! I will try to not get scared but this is a bumpy road in front of me :p

[smoky oak]

Oh thos make me feel unwell 😅

[glossy herald]

here you go mate 🙂 https://github.com/Guizmus/sd-training-intro

[smoky oak]

Omg that hand

It's shaped like

F
E

[onyx anvil]

Hahaha it really is!

[glossy herald]

is it cat-dj time yet ?

[onyx anvil]

Thanks a BUNCH Guizmus I'll get into that, slow and steady

[glossy herald]

I think it is

don't hesitate to ask if you need

one last non-dj cat

cause he's cool

DJ cat time now

and dj cat only comes in 1920x1080

[onyx anvil]

Ooooh can I join in? Beginner style

[glossy herald]

of course

here are my settings

upper body, furry catgirl dj cat, colorful clothes , quirky, vibrant appearance, playful accessories, creative behavior, imaginative, spontaneous, dj headphones, mixing platines, music club, 1970's night club, indie theaters, people dancing on the dance floor background, underground danceclub, highest quality, fur texture, intricate details, (cinematic lighting), RAW photo, 8k

Negative prompt: airbrush, photoshop, plastic doll, plastic skin, easynegative, monochrome, (low quality, worst quality:1.4), illustration, cg, 3d, render, anime, asian, korean, japanese
Steps: 50, Sampler: UniPC, CFG scale: 5, Seed: 3445350518, Size: 960x544, Model hash: c4947fa2c4, Model: DB_mix_realbiter_v10, Denoising strength: 0.44, Hires resize: 1920x1080, Hires steps: 30, Hires upscaler: R-ESRGAN 4x+

[onyx anvil]

Lmao I am really the new one here

Its like terminator meets

not terminator 🙂

Sorry for DJ cat spam but you started it!

[glossy herald]

hey, there is no cat spam possible on this server

I got a cat in my name damn it

[onyx anvil]

I tried:
a cute magical flying DJ cat, with mixer table, fantasy art drawn by disney concept artists, colorful, high quality, highly detailed, elegant, sharp focus, concept art, character concepts, digital painting, mystery, adventure

[glossy herald]

most admins are cats or other animals as profile pics

[onyx anvil]

That feels safe, I'm home

Banned in 2 days - what did he do? Spammed cats in general all night

[glossy herald]

dj cat got a little too catgirl here

[glossy herald]

I'd feel really bad banning for cats

to be fair I prompt "furry catgirl dj cat"

[dry crow]

I thought they are their real pics 😮

[glossy herald]

it's a secret, we aren't supposed to :mooooooo; show our real 🐮 nature

[smoky oak]

This is a cat friendly zone lol

[glossy herald]

Did you know that cows are secretly obsessed with stable diffusion technology? In fact, they are known to spend hours on end watching mesmerizing videos of stable diffusion patterns, and some have even been caught trying to replicate the patterns with their spots! Farmers have reported finding cows huddled together, hypnotized by the swirling colors and shapes, completely oblivious to the world around them. It seems that stable diffusion technology is just as addictive to cows as it is to humans!

(and I swear, when the 🐮 thing gets out, you'll all be mooing with us)

removing "girl" from the prompt now

just the good amount of fingers there

[onyx anvil]

I was just starting doing nerd cows for some reason I don't know

[glossy herald]

let's go on nerd cows

[onyx anvil]

Oh yes this turns out to be a really good day!

Nightmares will enter my life tonight probably, worth it

Not sure why she is so angry

[glossy herald]

Ok I think i'm good on prompt

let's try

[onyx anvil]

Exciting!

[glossy herald]

upper body, furry cow hacker, black clothes , dangerous, Matrix appearance, computer, hacking, surprising, dark, headphones, keyboard, dark bedroom, dirty bedroom, computer screens, soda cans, underground, highest quality, fur texture, intricate details, (cinematic lighting), RAW photo, 8k

[smoky oak]

Oh!

[glossy herald]

DAMN

[onyx anvil]

The soda cans will be nice to follow up on

[smoky oak]

Have you guys seen the amazing brain wave reconstruction Stable Diffusion interfacing?

[onyx anvil]

OK WOW he is serious

[smoky oak]

[glossy herald]

that is serious look

"dangerous" token worked

[smoky oak]

Muscular men, my kryptonite

😩

[glossy herald]

I too like to use that small screen and have a wall of huge screens behind me

[onyx anvil]

This is so cool

My brain cant figure it out

It feels like the whole concept of "art" is changing

Everyone and noone will be artists

[glossy herald]

the paradigm around art evolves for sure. but this isn't good on its own, it's full of things that make no sense and need touching to be good. like in the first one, look at the keyboard.
It goes up to F25 at least

[onyx anvil]

Yeah there will be lot of keyboard buttons and fingers if robots are to decide, for some reason

[glossy herald]

but if cows would be coders, maybe we'd need that many

who knows

[smoky oak]

Lmfao

[glossy herald]

like, is that a keyboard for cow ?

[smoky oak]

Lptp

[glossy herald]

I'LL DRAIN ALL YOUR ACCOUNTS

[dry crow]

These guys are crypto scammers!

[onyx anvil]

Im doing a bunch of cows with your prompt now, one will need a really good chiropractor... (any prompts on chiropractors?)

[smoky oak]

Shrimp cow

🦐

Lmfao

[onyx anvil]

Hahahahaha

[smoky oak]

I have a furry dedicated model, let's see what I can do

[onyx anvil]

Is he using the small laptop screen? The bigger one is just for show maybe

[glossy herald]

last hacker cow from me

let's go to beautiful backgrounds for my laptop

[onyx anvil]

Is that the new theme?

[glossy herald]

woman in automn light is my theme :p

[onyx anvil]

I'm on it! Prepare to change background!

[steep zodiac]

The first ever DMCA take-down just occurred for some models on CivitAI: https://i.imgur.com/qXPhatt.png

[glossy herald]

I saw that, seems like a proportionate response. I like the disclosure at least

[onyx anvil]

I should stop doing them in 512X512, not the best wallpaper format

[glossy herald]

it's quite a lot harder to do 1920x1080

[dry crow]

Try 768x512 and later upscale

[onyx anvil]

Oh it is!

[steep zodiac]

Anti-AI artists are also apparently setting up malicious bots to target the Stable Horde and LAION (Anyone know the username of the person running Stable Horde?): https://cdn.discordapp.com/attachments/1006139459860975716/1088501275898941520/image.png

@winged dragon

Saw this on the LAION Discord

[onyx anvil]

Oh ok here we gooooo

[glossy herald]

man, don't, do little things ^^ I use extensions and 24GB VRAM and still take more than 30 seconds per picture there, but without some things you'll just end up with duplications in there

[onyx anvil]

Understood, thanks

(in before connecting to the server room at work going ham with 300+GB RAM and destroying the whole electricity grid for my city)

[glossy herald]

it's not ram though

it's VRAM

[onyx anvil]

Oh right for the graphical card

[steep zodiac]

300 of VRAM?

[onyx anvil]

Of course!

My bad I'm not the hacker cow

[glossy herald]

yeah, I have 128GB ram and can't use it

[onyx anvil]

I'm more the DJ cat

[glossy herald]

x)

the largest good one I did, my current background

ok, the girl is technically in the water

[onyx anvil]

My first and maybe only try for 1920x1080 had a lot of leaves :p

[glossy herald]

we see the autumn come out there

[onyx anvil]

I'm having so much fun from all this

For how long have you been into this?

And for how many sleepless nights have you all been creating DJ animals?

[glossy herald]

hum, I've been around since public beta

so around summer last year

I mostly spent time making pictures, then doing animations, then training models

[smoky oak]

@glossy herald using a furry model

[glossy herald]

time for bed here, that's a good night cat to you people, hope to catch you all soon !

[onyx anvil]

Thanks for your time and creations tonight! Night night

Ok wow my recent hacker cow just forgot the hacking part completely and decided to become some kind of end game boss in dark souls instead lmao

[winged dragon]

I posted it there 😄

[steep zodiac]

oh lol, oops

I may be a tiny bit tired today lol

[tired nymph]

hello, could someone do pixel art fox for me? i wanted to see what basic it can do

like could midjourney create something like this?

[smoky oak]

This is not a commission server, and it is certainly not a midjourney server, but best of luck@tired nymph

[uneven halo]

cute little girl

[stoic pivot]

I made an embedding where the negative prompt "blurry" may struggle 😄

It was based on images taken with high ISO Ilford film.

[smoky oak]

@glossy herald he's so handsim

With them big ol peepers

The floor flopper sunlight fun time

[stoic pivot]

Good image, AI often struggle with lights going through thin skin as in the ears of a cat.

[smoky oak]

Lol

[dense tapir]

Well, no nude, or anything bad and the same prompt I used on SD1.5 that I showed here gets me this on the new Bing AI

Far too censorious like MJ

@smoky oak

[smoky oak]

So excited for the era of hyper cencorship

😐

[stoic pivot]

What do you post?

[dense tapir]

Nothing, I was trying to gen an image

It didn't like my prompt YET I posted the image in here many times with various outcomes. Not for me and they can go pound sand.

I did report it though

Oh, I looked at their content policy and it passed every last one.

[smoky oak]

@dense tapir did you see how bad adobe firefly is? Lmfao

[dense tapir]

I saw it on a vid but never tried it. I gotta say about the Bing one it is a MJ knock off though it did do my other prompt more to the vision in my head. Still, censorship like this I want no part of it.

[smoky oak]

Firefly is horrible lmfao

[stoic pivot]

I still do not understand why an app like photoshop need AI, the only place I can understand the need for IA is with "Content aware fill".
Pleas misunderstand me correctly, I think AI is good, but not in a Editor, that should be in a own app.

[dense tapir]

Well, like it or not it is here

even if you think it is not needed they will tell you why you need it

[smoky oak]

I have seen results from people who didn't cherry pick, and oh my god firefly is asssss lmfao

It's giving dalle mini levels of quality. I swear lol

[dense tapir]

oh, ugh

[smoky oak]

Like broooo

WHAT IS THISSSS

Vs SD fix of it

[dense tapir]

Looks like it was pasted on

[smoky oak]

And even that SD gen is the very good lol

[dense tapir]

isn't firefly still beta though?

[smoky oak]

Yeah but Jesus it's bad lmao

These are the images that adobe themselves are putting out

Like bro lmao

Dalle mini called, and it wants it's artifacts back lmao

It's giving SD v1

Artifacts, artifacts, and more artifacts

[dense tapir]

This, right here, ruins it for me

[stoic pivot]

Is that not during the test?

[dense tapir]

Not sure, but if not then lol

[smoky oak]

I'm pretty sure that is specifically put there because the product isn't fully released yet

So Adobe is just uploading these pictures to their Twitter, and they don't want anybody to like download them and use them

Kind of ironic

I just find it funny that one of the most hostile and problematic companies in the world is preaching ethics about their data set lol

[stoic pivot]

I did hear something that they had used images from own database that was under CC0 and therefore have to follow the same license during the test.

[dense tapir]

Of course. Hypocrisy reigns supreme.

[smoky oak]

Like sure, you guys may not be affecting artists through your data set, but instead you're going to hike the prices of your products that artists rely on, which is a thousand times more detrimental to their well-being

[dense tapir]

@stoic pivot If that is the case then I would suspect their dataset is tainted and legally would be allowed ever

[smoky oak]

AI image generation does not hold a candle to the amount of damage that Adobe causes to their own community by abusing their power to price gouge them

[dense tapir]

CC

[stoic pivot]

CC = Creative Commons, CC0 "No Rights Reserved", but I think I said wrong and that it would be the "CC BY-SA" Adobe used.
Just now not 100 % sure, it was some told me.

[dense tapir]

Yes, I know and yes, to all of that

CC showed just how ruthless Adobe was, imo

[stoic pivot]

There was a time when Adobe was good, but few remember that day.

[dense tapir]

btw, on bing AI this is not allowed

[smoky oak]

It surely hasn't been within the last decade or two lol

[brazen yacht]

I am trying something here, inpainting is dope af

oh lol, img to big to embed ig

[ripe cedar]

Win11 was installed on the PC I got and I hate it so I'm going to sail the high seas to downgrade to 10 again

[solemn topaz]

web-ui's masking works like a charm but I get low quality renders with a custom mask. what am I doing wrong?

[ripe cedar]

Spent a bunch of time installing stuff and taking out unneeded stuff, then it reinstalled the default trash it came with. Also kept deleting a program even though I disabled the security

[steep zodiac]

Fun is not allowed with any of the big corporation AIs

[dense tapir]

Yeah, I know.

[dense tapir]

ironic that you will end up having to pay for the service too

[static tusk]

your mask is too precise. needs to be more loose, similar to this. give it more room to work with

[dense tapir]

I forget but in 1.5 how do you push the subject away from the camera?

I tried long shot, medium shot, extreme long shot

[static tusk]

Almost never works for me either

[dense tapir]

Nothing I do moves them only changes them.

[smoky oak]

Starting to regularly see AI generated advertisements

[dense tapir]

Same

[smoky oak]

Specifically from this game "bloodline"

[stoic pivot]

If you want it to look that the girl is in the background, say what she wearing on the feet.

[smoky oak]

IDEK know what the game is. That's legit the whole ad lmao

[dense tapir]

Oh? Let me try that

[smoky oak]

If you specify body parts, it usually helps

If I want a cowboy shot I usually specify upper torso, and hips

Then it typically backs the subject up at least far enough to show those items

[static tusk]

much easier these days to use openpose CN

[dense tapir]

Funny cowboy shot is a wd1.4 tagf

[stoic pivot]

When the AI cut of the legs you can say "wering socks" and the Ai try to make socks by showing the feet too

[smoky oak]

That's how I do it too

[dense tapir]

the pose stuff is just not for me as I am waiting when the AI can do it, and it will eventually do it I just hope not on some damn paid service as SAI sleeps.

trying "wearing greaves" now

wearing greaves

wearing boots

in pink socks

[smoky oak]

Is his SD 2.x?

*this

[dense tapir]

cowboy shot (1.5)

maybe fullbody shot?

[static tusk]

looking at old gens ive used ((distant shot)). worked sometimes

[smoky oak]

I've used (wideshot:1.4)

And (full body:1.4)

[stoic pivot]

full body portrait.

[dense tapir]

I have to say cowboy shot is interesting

even changed from female to male

full body portrait

(wideshot:1.4)

(full body:1.4)

[stoic pivot]

Just look so you do not have any conflicting prompts. If I write something like "female" that the AI do that, but if I add "detailed face" that it zoom in on the face so more details are visible.

[dense tapir]

(full body:1.6)

1.6 seems to be it

yep, 1.6

Thanks @smoky oak

[smoky oak]

No worries :>

Glad it worked!

I have to manhandle the AI to give me what I want frequently lol

[dense tapir]

I forgot just how bad it was in 1.5 with varying shots

[stoic pivot]

Prompt - "silly fast rainbow girl run on a street sneakers and hairband , digital art, motion lines", then I added my "by BradC" embedding when testing a new model called CarDosAnimated.

I did read it as CarDosAnime first and thought it should be good doing cartoons, but the background is always realistic.

[dense tapir]

which is better batch size or count?

[stoic pivot]

It going forward, looking good.

I often like less details and bright colors, so I am not good judge dark fantasy.

[dense tapir]

See, I want more details and duller colours

[solemn topaz]

are my colors correct tho? it should pure black and pure white right?

[dense tapir]

good gracious I went to 50 steps and

Dunno why 1.5 likes to eat the camera

[static tusk]

yep, you want the white to be what is generated, just black and white

[pastel mango]

haaaaihooooo

[static tusk]

are you using Chempunk TI? It sort of looks like it, not sure. I found that embed to be very difficult to control

[pastel mango]

I'm liking this style I captured

[stoic pivot]

When you add one to many words in your prompt.

[pastel mango]

I feel that this is what God looks like

[stoic pivot]

Maybe I shall add "logic" in the negative ptompt.

No comment

[dense tapir]

The hands on this one 😦

[pastel mango]

heh

[flat pike]

Gaming Grandma Series 🙂 👵

[stoic pivot]

Last for today, it feel like I am still addicted to sleep and have to do it at least once per day.

[dense tapir]

Interesting

1.5 is fighting me

o.O

[dense tapir]

Nope, no tis with these

[honest abyss]

/hazme una portada para la guia de un taller de live streaming en alta resolución con mucho detalle y de aspecto moderno, minimalista y profesional

hazme una portada para la guia de un taller de live streaming en alta resolución con mucho detalle y de aspecto moderno, minimalista y profesional

[smoky oak]

@honest abyss lamentablemente no soporta español

:<

lo siento si mi español es rudo

[pastel mango]

Je ne peux pas commencer à vous dire dessas línguas hablar de estos idiomas

I don't think thta made a whole lot of sense

which gives it a bigger point

[static tusk]

If the same clocks were to be applied to the RTX 4060 non-Ti SKU, one could expect around 15 TFLOPs so 5.5 times less than RTX 4090.
5.5x slower damn!!

[coarse skiff]

Hello, I have a question, I want to make a AI learn to help with my art style, but I don't want to share publicy so I can use my art in the way I want. Is that possible with Stable Diffusion?

[static tusk]

of course. train whatever you like and use it on your pc. no need to publish online

[coarse skiff]

How can I do that?, but the machine will learn locally or the style gets memorized on the internet?

[static tusk]

you need a local installation of SD on your PC, then the extensions to be able to train. This all depends on how good your PCs GPU is

[coarse skiff]

So I need a powerfull pc?

[static tusk]

a good graphics card is important, the rest not so much

[coarse skiff]

Thank you very much! I´ll search for tutorials then. Is there´s a good Youtube video with an explanation for training locally?

[static tusk]

there are some but I dont know they are good. Just search on youtube for training, there will be a few

[coarse skiff]

Thanks!

[pastel mango]

hai

[glossy herald]

The Acowcalypse is cowming

[flat pike]

https://prompthero.com/BalrogDx my promthero gallery, feel free to copy handsome ideas

[glossy herald]

Don't mind if I do, those seem top tier, wel done

[flat pike]

thanks, if u need some more information about a pic, just write, but there open and readable with all needed infomationein SD (via png info)

[glossy herald]

adapting your fox cat hybrid right now

I need a good cat for #1023999442338201721 :p

[flat pike]

have a better version ready, im working on it right now

and more cat with better fur

wait a sec i upload it

[glossy herald]

woops I went overboard and OOM'd x)

[glossy herald]

I think I found the one I needed

trying to highres fix it rn

[flat pike]

ok, so i will check ur submission if its posted 🙂 great that u like my pics

[glossy herald]

always loving good prompts tbh

you don't have the same prompt style that I have at all, and it's refreshing to learn

[thick badge]

https://cdn.discordapp.com/attachments/1084340573097443358/1088759736264048742/The_future_landscape_unfolds_before_the_plaza_a_breathtaki_4652913c-7df9-437a-b223-09500d0f706d.png

[flat pike]

nice one, good lightning

[green plover]

Unfinished but verified functional AI generated texture map

[nocturne wagon]

生成能量的海报

[tough monolith]

seems my draw have some noise

Are there any good models for the background

[sinful mirage]

As a stable diffusion main for months, I have to admit the new dalle has a ton of detail, raw output

[oblique nimbus]

the program https://huggingface.co/spaces/lnyan/stablediffusion-infinity

the prompt: One complete, centered, dark black jade green, Italian muscle car, designed in Italy, in 1975, coolest car ever, cinematic, long shot, 4K, HD, depth of field, long shot, bokeh.

the neg: signature, autograph, watermark, out of frame, duplicate, people,

the result

but why the black and white background?

[dense tapir]

because it looks cool, emphasis on the car for your eyes, etc...

I have noticed "cinematic" loves b&w

[kindred jungle]

/bull dog akita

[oblique nimbus]

ok so that's it

kinda looks like a challenger

[wispy nest]

Hi. I have a pretty crappy mobile picture of two musicians which I would like to let SD do it's magic on. I don't have an exact vision of the results, but maybe something cartoon-ish. How would I do that in SD webui?

[oblique nimbus]

I know there is something I can add to my prompt to get rid of the extra merged car but what? prompt: (One car) complete, centered, ((dark black jade)) green, Italian muscle car, designed in Italy, in 1975, coolest car ever, cinematic, long shot, 4K, HD, depth of field, long shot, bokeh.

[dense tapir]

https://cdn.discordapp.com/attachments/1004159122335354970/1087593925516202094/tmpd1em0cjb.png

@oblique nimbus

[pseudo aspen]

Help. I am getting results of this kind all the time while trying to use Latent couple + Composable LoRA. SD completely ignores my prompt and loras and does random things. Anyone has some experience with this?

[agile jewel]

that kind looks like a washing machine using pants

kinda*

[pseudo aspen]

I KNOW. And it had to be a photo of two people...

The extension I downloaded had to make it but instead it completely fd up my loras and is doing nothing

[agile jewel]

good luck with that

i have no idea how to solve

[dense tapir]

I downloaded an extension that did the same thing and also fucked up my venv as I had to completely start over

[pseudo aspen]

Damn

[dense tapir]

I had no choice as nothing else worked

[agile jewel]

what is the best sampling method for architeture and horror images

[pseudo aspen]

Welp, I was trying to do it without the extension but SD gave me an amalgamate of the both lora trained people

[dense tapir]

my suggestion is to rip out the offending extension and delete the venv to start over

[dense tapir]

Depends

If you have a powerful video card DDIM 50-100 steps is actually the best.

[agile jewel]

ok

[pseudo aspen]

Bruh. I just realised something. Seems like that something happened with sharing LoRAs between my LoRA training account and my main SD one so the LoRA isn't actually here

[dense tapir]

Well, no extension can do that and survive for long as the crowd would report it

[agile jewel]

you are right

[dense tapir]

My 1060 I do 20 steps but a more modern I would be doing 50-100 almost always.

[dense tapir]

Is that 50 or 100?

With DDIM the more steps the sharper it gets

[agile jewel]

66

[agile jewel]

that is really cool

[dense tapir]

Cool.

[agile jewel]

nice

[pastel bluff]

New phone who dis?

[dense tapir]

Really hard to not get leather based armour it seems. A lot of it seems BDSM related.

Not that one but a few I can't even show

[agile jewel]

lol

[dense tapir]

DDIM 100 took my card 3:52 to gen.

[glossy herald]

Nah, you didn't release 3 new models while I was away, no way

Are these good people ? Enjoying ?

Can't test right now but so hyped !

[wispy nest]

Quick question - does gen and gen with highres fix produce same image if prompt and seed are the same?
I mean, can I just re-gen something I previously generated without highres fix, just to add more details with highres fix or it's not how it works? Will images be the same or not?

[glossy herald]

It will depend on how much denoising you put in the highres

[wispy nest]

makes sense

[glossy herald]

High-res is supposed to create more details

But it can change composition too, especially if you use latent upscale imo

[wispy nest]

eh, so best way is to keep just doing gens with higher resolutions then still

highres fix takes longer to gen then just normal gen at higher res

[glossy herald]

Well, for pow, i ran a lot without high-res, but still in quite big res, and looked for a good seed, then ran high-res on it while trying multiple denoising

And then ran the same with lots of steps

[wispy nest]

yea that might be an option too, but if it's not as consistent idk what's the point, lower denoise will probably do less details then normal gen at higher res too

[glossy herald]

0.45 is my best value for it

[wispy nest]

hmmmm...steps might be the way at low denoise too with DPM++ , right

[glossy herald]

It very rarely details the composition, and still let's a lots of details in

[wispy nest]

aight, ty

[glossy herald]

Using unipc mostly on my side but yeah, it converges on the same thing anyway

[stoic pivot]

Prompt was "super cool unicorn" and then I added my new Embedding "CraiyonUnicorn", I start to get the hang on doing very bad and silly Embeddings 🙂

[stoic pivot]

Some time I think AI was just made to confuse me... SD v1,5, prompt "art GigaChad CraiyonUnicorn".

[late remnant]

This artist uses stable diffusion to render their drawings. When I try to do this using controlnets canny and depth, SD just butchers my drawing. Maybe someone could point me in a direction of how to achieve this process.

here is another example

[wispy nest]

decided to upscale a bit old gen
probably will do more attempts on same theme at some point ^^

[sterile kiln]

Good idea, have 10 Gigs of old gen

[wispy nest]

lol

[smoky oak]

I keep finding more and more gens from firefly, and it's really striking me just how bad it is lil

Firefly should be renamed dumpster fire

[wispy nest]

looks like my attempts at 2.1 model...

[smoky oak]

I find out how everybody seems to be praising the most corrupt company in the creative space for practicing "good ethics" for training on their own shit. Like... They screw artists over every single day. They are probably the most unethical company in the scene

[crystal bloom]

probably painted the details in photoshop

[late remnant]

I believe first image is painted in photoshop, second is image to image SD

[crystal bloom]

then in that case maybe used image to image with low denoising strength. Check this art walkthrough maybe can be helpful. https://www.youtube.com/watch?v=R7ywLyOoBuE

[late remnant]

kk, thanks ill check it out in a bit

[brazen yacht]

Theres a new Dalle?

Oh, chat moved on

[onyx anvil]

Wow so many different nice results! 😮

[glossy herald]

COWs !

well, it's cat day currently

[onyx anvil]

I thought it was always cat days in here!

[glossy herald]

but i'm already eating all my bandwidth uploading models

it is always cat day

but model cat day is something else

[onyx anvil]

I forgot to sleep last night because of cows, cats, hackers etc

[glossy herald]

here is the showcase

[onyx anvil]

Someones been busy! 😄

The clockwork one was amazing

[glossy herald]

#1088926673644290058 if you want to check

[onyx anvil]

Also the B/W kittenish

Sure will

[smoky oak]

what shpeed?

[glossy herald]

I have a lot, it was HF that was slow as hell

it's online now

https://huggingface.co/Guizmus/SDArt_Catstravaganza

but I have symetrical 8MB per second

[smoky oak]

ah, not too bad at least

also, I am still not properly understanding what this weekly challenge is I think

[dense tapir]

I tell it "wearing full body armor"

[glossy herald]

cats

[smoky oak]

but you guys take the submissions and train a model on them? or am I misunderstanding

[glossy herald]

me guy, but yes

[dense tapir]

DDIM wouldn't do it but Euler_a did

[smoky oak]

ok, so that is how it works

and then everybody was tagged, so calling their token does what exactly? I just wanna make sure I understand this all properly 😅

[glossy herald]

the model is just me trying to make something fun around, it is not even mentionned in the #⭐｜pow-info

[smoky oak]

ah, alright

I thought it was done every week

[glossy herald]

I'm about to go to bed so I cut everything out already
but here are one or 2 examples

that's 2 original ones from the submissions

that's some variations with just 1 token

[glossy herald]

you could also use 2 tokens at once, and it would "import" both inspirations

[smoky oak]

I'll bother you some time about this, cause I don't think I am asking right

ohhh, wait

so its trained on them as inspiration, ok

that what I was confused on

[glossy herald]

each token is linked to only 1 pic

so yeah, it's a weak token, and acts as inpiration mostly

the token "SDArt" is different

it's trained on all of those, so it makes some kind of "mean" art

[smoky oak]

ok, ok, I get it now

I was really confused, that cleared it up

[glossy herald]

it's a strange way to tackle training, I'll give you that

[smoky oak]

its like a multi style inspired model

[glossy herald]

a little yeah

[smoky oak]

ok, that makes more sense

[glossy herald]

and people like to play around their submission a lot it seems

[smoky oak]

i thought it was trying to teach the AI how to make cats, and I was really confused lol

[glossy herald]

nah, cats are quite already known

[smoky oak]

yeah, that is where my confusion came from haha

[glossy herald]

but it's not just cat we got yeah, artistic cats instead

[smoky oak]

yeah, I get it now, awesome

[smoky oak]

lol

[glossy herald]

well, civitAI is down, the model will have to stay on HF for tonight

[smoky oak]

damn ._.

civit AI likes to be unstable diffusion

[glossy herald]

have a good evening !

[smoky oak]

you as well :>

[glossy herald]

or over part of the day !

[onyx anvil]

Are you also doing this to test things out? :p I just waited 2 hours and I'm not even sure why really lmao

[stoic pivot]

I think I forgot "Ultra realistic", "8k" and "Trending on ArtStation" and that's the reason it didn't turn out as I wanted.

[smoky oak]

lmfaooo hahaha

who was that one dude that made the most useless prompt ever

let me see if I can find it

"A stunningly beautiful girl. Cinematic in nature, this hyper-detailed scene is filled with insane details and beautifully color graded using Unreal Engine. The use of DOF, Super-Resolution, Megapixel, Cinematic Lightning, Anti-Aliasing, FKAA, TXAA, RTX, SSAO, Post Processing, Post Production, Tone Mapping, CGI, VFX, and SFX has created an insanely detailed and intricate world. The hyper maximalist approach and hyper realistic Volumetric and Photorealistic rendering bring out the ultra photoreal and ultra-detailed aspects of the scene. With 8K and super detailed visuals, this scene bursts with full color and Volumetric lightning, using HDR to create a realistic and breathtaking environment. Powered by Unreal Engine and rendered in 16K with sharp focus, the intricate details of this scene are truly mesmerizing."

That prompt lol

I admire the dedication to the craft haha

thats some passion, if I do say so muself

[dense tapir]

That prompt gave me that, lol

[compact meteor]

Black guy wears a black coat And brown pants, sleeps on the floor, and cardboard paper is underneath him

[vestal garden]

Black guy wears a black coat And brown pants, sleeps on the floor, and cardboard paper is underneath him

Bigfoot drinking beer and wearing hawaiian shirt

[smoky oak]

@dense tapirI am playing with fire

I edited the config to let me go to batch 16 hehehe

the higher the batch, the more time efficient it is so... fuck it

it works!

woooo

that means I can do a 1600 image generation in one grid lol

[dense tapir]

Oh, snap