🏞｜general-with-images

[wispy nest]

https://stable-diffusion-art.com/samplers/?utm_source=sendinblue&utm_campaign=Newsletter&utm_medium=email#Image_Convergence

hmmm....

probably just little changes

We need x\y\z on that 😄

I know DPM ++ doesn't converge at all, changes are huge, did x\y\z matrix on that

seems like it's not about all DPM ++ , only "a" versions

I'll do matrix on "euler a" later

[stoic pivot]

Who is up for a challenge that I fail: With only Text2Image, make a big human nose waking in New York. No Loras, embedings or so.
Rules, Best fail is also good to be posted.

Human Nose walking in New York.

[wispy nest]

really

bot think my matrix includes nsfw

I did matrix..."Euler a" still changes too, not as much as "DPM++ a", but still changes

[ripe cedar]

What are a few good general purpose VAEs? I don't like using VAEs that are for specific models only.

I have only the 84000 VAE so far

[wispy nest]

vae-ft-mse-840000-ema-pruned.ckpt```

but yea, some models are better with it's own vae

[grizzled sage]

yeah its weird what the bot won't allow

[hardy tulip]

boob oscillation

something more serious, where can i find stuff to put here?

holy crap nice estimate

[static tusk]

it can save your prompts. click the small save button, give it a name, then later you can select it from the menu and click load

[hardy tulip]

ah thanks

where can i find some good ones?

[static tusk]

civitai examples

[hardy tulip]

having base prompts would be extremely useful

[woven dawn]

#1072013871730131004

[reef sail]

how!?!?!??!!!??!?! PLEASE HELP

[smoky oak]

You have the same "cannot add middleware" error?

[reef sail]

yes and i cant find a fix

[smoky oak]

Ok, this is how I have fixed it several times

One moment

@reef sail alright, so you wanna activate your venv. To do that, go to the root folder to SD (where all the folders for it are kept) then in the top bar where the folder directory is, you wanna type in cmd and press enter

[reef sail]

next?

[smoky oak]

Then you wanna do

Run the second command

And you're done

Just close out, and restart it

It's fixed it for me every time so far

No idea why it works, but it has worked for me lol

[reef sail]

restart what? my pc

[smoky oak]

Oh, just launch the bat file for SD

My bad

Really hope it works!

🤞

[reef sail]

it doesnt work

[smoky oak]

Aw damn...

I'm sorry man

It's never not worked for me

[reef sail]

i just had to uninstall dreambooth

but now it works

[dense tapir]

@smoky oak Just ordered a Ryzen 5 5600 to replace my 1600 after seeing 80-90% gains in every benchmark from fps games to rendering etc... for 138 that was too sweet to pass up. Besides, when I do get a new video card it will be heavily bottlenecked by this 1600 AND I can increase the speed of my ram (need this drastically as I went to 48 gigs). So glad AMD did this. I downloaded the last bios for my motherboard earlier that is 9 months old and made sure it was for this. This means I don't have to upgrade this system for a long time to come so maybe 4-5 years more (or until the old B450 finally dies). Next up will be post summer and a new vid card.

[smoky oak]

Just be sure your mobo allows you to slot in a 5000 CPU

[dense tapir]

as I said already checked

[smoky oak]

My bad, just wanted to put that out there really fast in case

I see it now

[summer viper]

(increase perfomance of 10%)

[dense tapir]

I have to flash it then remove the Ryzen gen 1 and insert the new cpu etc... and boot.

[smoky oak]

Yeah, all good

[dense tapir]

zen 1/2 no bios needed but 3 yeah, and no going back either

[smoky oak]

For $140, that's a great deal TBH

[dense tapir]

sad thing is a b450 doesn't have enough ram for the uefi so all the fancy stuff is gone and back to text

[smoky oak]

Ah damn

[dense tapir]

I am fine with that though since I grew up with all text

[smoky oak]

My mobo has a UEFI, and it's quite fancy and fun lol

I hope to get a new system ASAP

But at the same time I am worried cause if I do, then my only option is NVIDIA

[summer viper]

games are not very demanding on the processor

[dense tapir]

yeah, this was my first mobo to have (well, the gigabyte died under warranty and they refused to help even though it was a known issue on the net).

Well, worst case I just buy a B550 cause AM5 shit is damn expensive

[summer viper]

especially with the arrival of directx 12

[dense tapir]

wanna bet

[summer viper]

ahahah

[dense tapir]

depends on the card though

[summer viper]

yeah

[smoky oak]

I am going to a laptop, as I am tired of having the restrictions of a desktop

[dense tapir]

oh, fuck laptops as I have been anti laptop for 3 decades now.

[smoky oak]

After seeing how damn nice my uncle's laptop is, it's exactly what I need

His laptop absolutely shreds my desktop, and it wasn't even that much comparably

[dense tapir]

well, good luck cause a laptop has one main issue... HEAT

[smoky oak]

$1680 for. 3080ti, 12900hk, 32GB DDR5, 300hz IPS display, full keyboard

[dense tapir]

it will throttle the video card etc...

[smoky oak]

His has top of the line components, and it's not that big of a deal

[dense tapir]

you might want to keep a desktop to train with

[smoky oak]

Something like an ASUS Duo can out cool some desktops haha

But duos are disgustingly expensive lol

[dense tapir]

yes

[smoky oak]

I just need the portability aspect of a laptop

[dense tapir]

I still remember laptops (lol) that weighed about 15 pounds

[smoky oak]

I am actually aiming for a CPU beast laptop with no GPU

If I could find a top of the line CPU in a laptop with my GPU, that would be perfect

[summer viper]

yeah

amd

[smoky oak]

Cause then I could buy a dock and use my 3060ti with it

[dense tapir]

I will say I am done with all of this as I am getting a 4090 just have to really save so Dec unless something magical happens and Nvidia green boys shit out a leprechaun and it reduces their prices.

[smoky oak]

Currently AMD isn't offering the CPU perf I am after on mobile, sadly

[fickle gorge]

What is the target performance?

[dense tapir]

LOL, I personally, and directly, feel this man's pain

[smoky oak]

Single core perf for me

[dense tapir]

Yeah, single core Intel is still king

[smoky oak]

The new 7000x mobile CPUs are disgustingly fast for multifore, but single core is a ways behind into

That new 7945HX is DISGUSTING

Almost the same perf multi as a desktop 13900k

[fickle gorge]

The worst experience I ever had where I failed to install something was Mozilla TTS

[dense tapir]

Did you see Intel is now coming out with a 120 something core CPU to compete with AMD's soon to be released 9x core?

LOL

[fickle gorge]

The 144 one? I think they are being threatened by some OEM

[dense tapir]

btw, 5600 vs 1600 the die is 7nm vs 14nm

really?

[smoky oak]

Yeah, but AMD's cores are just not comparable

And tells new CPUs that they just released use around three times the power for around 60% the performance of AMD's offerings for three times as much money

[smoky oak]

They are so bad that even companies like LTT and gamer's Nexus said that they should have just completely withheld the generation, because their investors were so mad

[dense tapir]

Intel has always been stupid but Nvidia is now the new king.

[fickle gorge]

It seems to not have enough samples of capybaras and tries to mix with beavers, it seems

[smoky oak]

I made a capybara LoRa!

Just for that reason :p

[fickle gorge]

OMG, where do I get it?

[smoky oak]

It's a little messed up but I can retrain it pretty easily, gimmie like... 30 minutes?

I trained it off of AItrepreneur's bad advice

The way it is, it actually works way better with furry models than realistic models lol

[stoic pivot]

This is common is Stable Diffusion, that things and humans is getting the same color and look as the person in the foreground, is this something we shall look fore to spot AI images or is it just SD doing this?

[smoky oak]

@fickle gorge it works well with anything furry v1

But I need to retrain it to work with realism models, cause it doesn't work with realism... Even though that's what it was made for lol

Also love the way he's lifting that weight lol

[fickle gorge]

So strong

[smoky oak]

I think it works really good :>

[smoky oak]

This model has a preference for muscle men, just like me lol

But believe it or not, that guy is super skinny compared to the others it gave me

[stoic pivot]

I like AnythingFurry, but it so often mix styles like this, with a drawing on a photo background.

[dense tapir]

My computer is very old from 2009 but is a full sized BEAST. Sadly it lacks the ability for water cooling and there is a 4090 with 2-240mm fan radiator.

this case was damn expensive in 2009 of well over 200 bucks

I do want a new case but they are all bling bling glass sided UGHS

[smoky oak]

Anything furry is fantastic, it's just... Mature haha

[dense tapir]

@smoky oak

[stoic pivot]

Yes, but as long as you write what clothes they wear it is easy to tame, problem is if you forget that and forget to remove some words when switshing to AnythingFurry.

[dense tapir]

btw, this is something new from Amazon (I guess to stop the porch pirates) but I have to be here to accept the package AND I have to tell the driver my one time passcode before he will hand it to me.

[stoic pivot]

My favorite model is one that have the almost word "Pron" in the name and I do not know if we can say that without post get removed.

[lucid hamlet]

😂 🔍

[fickle gorge]

One strange thing I'm noticing is how the CPU clock seems to be irrelevant for SD

Setting the clock to 1,2 GHz or 4,4 GHz has no difference on how long it takes to generate the image

Is the 1650 Max-Q a bottleneck this severe?

[static tusk]

the 1650 is quite slow that any cpu can keep up. Once you get a GPU with tensor cores thats when cpu speed is important

[smoky oak]

I'm really sorry, I'm having fibro fog today. My mind isn't really working. Can you please tell me what this is supposed to be saying?

I'm reading it, but it's just not connecting for me

[dry bloom]

#1011634831467221033 Logo design, Y,N,dental clinic, simplicity, readability, tooth symbol, smile symbol, suitable color scheme, target audience.

[dense tapir]

has to do with pytorch 2

[smoky oak]

So it's saying that 2.0 should be working better now, right?

[dense tapir]

39 it/s on a 4090 seems a tad slow

naw, don't go down that rabbit hole again

[stoic pivot]

Shall we start a flame war, pyjamas vs pajamas and every argument need a cozy PJ pic.

[dense tapir]

if auto doesn't implement it speed gain will not be there

a 4090 is so damn fast they can afford to sit on it being slower

[dry bloom]

#1011634831467221033 Logo design, Y,N,dental clinic, simplicity, readability, tooth symbol, smile symbol, suitable color scheme, target audience.

[stoic pivot]

I did get a 3090 as a gift, for me it still feel like overkill. 4090 has to be like the future.

[smoky oak]

All right, see that's what I thought as well. I just wasn't sure if it was supposed to be positive thing, or maybe the thing, or what I was really reading

[glossy mica]

#🏞｜general-with-images portrait of walter white wearing tuxedo, full body

[smoky oak]

My brain just really doesn't function on some days, and today is one of them. I'm having a massive fibromyalgia flare today. My legs and back are so tight I can barely move. And my neuropathy is being pissy as well. Like a bad wireless connection, my arms or legs just disconnect heh

[soft whale]

anyone have tips on img2img can't get any good result

[dense tapir]

He is just saying it is easy to install in real Linux. The real gains will come as soon as Auto implements them then even I will see some gain.

[smoky oak]

I see, I see

[stoic pivot]

Go to https://civitai.com/, many images there have a ! and if you click on it you see some info how image was made.
Also, use upscaling and "Hires. Fix".

[soft whale]

@stoic pivot could it be i need a different training model

[stoic pivot]

Yes, but many prompts are still good, exept they often use silly sords like "UHD, RAW Photo" and so on that do not work.

[fickle gorge]

...are they silly?

[stoic pivot]

I say, "Say what you want in the image" quality and so can not be changed, you can not make a 512x512 to 8K or UHD even you write it several times.

Yes, like RAW photo is just unedited data that can not be displayed as an image without processing, so the AI treat RAW as bare, like in RAW concrete. Raw and Brutal is in same class.

1st Brutal art, 2nd raw art

[static tusk]

RAW would never apply to digital art in any case

[stoic pivot]

Photo vs raw photo vs brutal photo

[soft whale]

anyone know how i can train an art style. i remember seeing someone on youtube about dreambooth, but never figured anything out..

[stoic pivot]

Many use RAW photo as a way to enhance quality, but we can not enhance the quality on the images AI make.

[static tusk]

its used to tell SD to source images that are both realistic and unedited. thats all it does

used for photo realistic gens

[fickle gorge]

At least for Realistic Vision, it seems significant enough to consider when to use or not

[stoic pivot]

But the last thing you want in a image is the RAW look from a image, it often a flat look, and there is no point to tell AI you first want a boring image and then ask the AI to make that image look better when you can ask for what you want directly 😄

[smoky oak]

loRA would work great

[wispy nest]

cow girls are the new meta

[fickle gorge]

My first experience with RAW photo was with a Nokia phone. The RAW files were gigantic (for a Nokia phone) and impossible to display on it, while the "normal" file was compressed, so it could be seen on the own device

[stoic pivot]

If you ask for "worst quality" it do not affect "image quality" but the design, but it do not get better with "Best quality".

[wispy nest]

yes

still doesn't hurt to have in a prompt iirc

[stoic pivot]

Well, I try to keep being lazy, and spelling is hard, few words good.

"Best quality, Superman hug a teddy bear", SD 1.5

Prompt: lora:epiNoiseoffset_v2:1.2 red-glow black blue detailed [brutal art|splatter art] Evil Superman
Negative prompt: frame, border, paper, framed

I'm not sure if I used the Noiseoffet correctly.

[storm elm]

spirder man

[stoic pivot]

Red-glow black blue detailed [brutal art|splatter art] full body portrait Evil [(Bat Man:0.8)|(Spider-Man:1.2)] screaming

It is a mix of bat man and spider-man

And Papa Smurf so all can be happy.

black blue red glow-white detailed [brutal art|splatter art] full body portrait (Papa Smurf:1.2)

[fickle gorge]

I'm guilty

[stoic pivot]

I found my nemesis... what ever I do, Micky Mouse refuse brutal.

He just shake it of with a smile... "black red glow-white detailed [brutal art|splatter art] full body portrait anger rage (BRUTAL Mickey Mouse from hell:1.2) gray grunge background".

I know that my images are far after what the good creators can do, but I try to avoid all "quality" prompting, I sometime use "trending on artstation" as a joke or in the negative prompt... Here is Elsa from Frozen.

[glacial tendon]

2.1 seems to be better at memes than SDXL, which focuses more on “artist” than “meme”. Do I need to prompt differently now?

[fickle gorge]

(Inspired by a game glitch)

Do weights work on the official SD? I remember the drama about "borrowed" unlicensed code back then but, if weights work, maybe that could work

[limber current]

I am new, anyone can advise how to use AI to redraw it?

Is there’s any prompt can use?

[wispy nest]

img2img or controlnet

for prompt just describe what is it and details

[stoic pivot]

I have not lerned ControlNet yet, but that is my bet.

[limber current]

Thanks for reply, but don’t know how to start to do it?

[wispy nest]

I'd suggest looking for tutorials on youtube , it's not something I can explain in few sentences

[limber current]

Could you please send me a video link to learn? Thanks in advanced

[stoic pivot]

I did so in the beginning and it sat me of in the wrong direction, all there just do what other do and same mistake spread from video to video, I had better luck learning through Hugging Face and Reddit.

[wispy nest]

idk, I watched few random vids

if you have something better - link it for him

[limber current]

Which kind of keywords you put into YouTube to search the video?

[wispy nest]

stable diffusion controlnet

[limber current]

Thank you, will try it

[wispy nest]

or you can just img2img instead

[fickle gorge]

What is your goal? Do you want it to be still in a drawing style, or do you want colors? You could try typing what you want for img2img and it could work fine

I took the liberty to try something, and it actually kinda somewhat works

[limber current]

Now the image is hand scratch, my goal is let Ai to redraw to make it nicer or different styles

[stoic pivot]

I say that best is to test and play first, like first enter what you want, like "A cartoon", then slowly add to that, "A cartoon drawing of a man in sailor uniform", see what you get and add, like "standing on a small deserter island with one palm tree and a ocean in the background".

After you get used to write prompts you can start looking for tips and tricks to add on to your own style of working.

[livid ermine]

Original:

Controlnet:

[stoic pivot]

Cool, a bit fun it remove wall and add window under the table.

[limber current]

Thanks

[smoky oak]

@fickle gorgeYou wanted the capybara LoRA right?

Sorry, I just got home, I will work on it right now and upload to civit

[fickle gorge]

Thanks, but you don't need to rush. I have to go sleep like, now, so I'll continue the perfect capybara chase only tomorrow

[smoky oak]

Alright, I am still gonna work on it tho so i can work on more capybara men and women in anything furry haha

[stoic pivot]

Before I go to bed I chare some nightmares.

[frank fossil]

“Show me an image of several steep snow-capped mountains in the distance with a blue-black lake in front of them. The lake is surrounded by large forests and grasslands. On the grassy land by the lake, there is a medium-sized estate.”

[stuck spindle]

Test

[smoky oak]

Test-Success

[raw forge]

Test

[dense tapir]

That sucks 2000 steps and still needed another 2k steps

I think dim rank 40 was too little from what I am seeing. Probably 48 would work

You can tell by the hands AND, for some unknown reason, it is refusing to do my bidding

oh, well

[smoky oak]

did you at least resume it rather than restarting the training?

[dense tapir]

for lora, even the dev who made loha, said restarting starts it over from scratch

I am not restarting I am abandoning it as it would take too much time

I only get 2h a day

[smoky oak]

ah

I would have just resumed training on it

thats what I do when my training is too short

[dense tapir]

it actually starts over which is odd

With TI embeddings it actually continues

[smoky oak]

you just have to use the resume training function and link to the last saved epoch. It works as intended, I have tested and used it several times

[dense tapir]

I am telling you what the dev said it really does not continue even with the weights it just starts over where you left off. Not the same, his quote, as doing a full run.

[smoky oak]

alright man, all I know is my results work just the same, so to each their own I guess

[dense tapir]

yep

[honest plinth]

How to generate images

[dense tapir]

No bot on server

[honest plinth]

I am not a bot

Tell me

[dense tapir]

I just did

[smoky oak]

@fickle gorgeIT IS DONE

[honest plinth]

How

[smoky oak]

and SHEESH did it come out good :>

looking like a whole ass photo up in here

my very own, coconutdoggo LoRA v2

[honest plinth]

How to generate pls

[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!

[smoky oak]

oh my god the discord april fools sounds just scared the crap out of me

[dense tapir]

heh, I have all sounds off

[smoky oak]

https://youtube.com/shorts/QqbkmPZK1c8?feature=share

the new april fools sounds are so cursed lol

[dense tapir]

Sounds more Halloween

[smoky oak]

@dense tapirI found a way to go from base gen to super high res images without needing to ruin composition with high res fix

[dense tapir]

??

[smoky oak]

I can tell you in DM's

tis a secret

[uncut igloo]

a boy seaching for treasure

[woven dawn]

Try replacing the artist entirely to see how it affects the image. If there's no change, I'd consider the position of the artist token in your prompt. If you're under 75 tokens and the artist os towards the end, the effects will be minimal.

[ruby pagoda]

Thanks

[glossy herald]

yep, good points there

[ruby pagoda]

I will try to do it

but I get a little busy now , So I will test it later

[pastel wadi]

bagdod, islamic history, people in the 1000 years ago, buildings, trees, water

#1072220168534642768 bagdod, islamic history, people in the 1000 years ago, buildings, trees, water

[real hill]

This was… the result of a major fuckup while using img2img

[summer viper]

#1019361238234443776

[summer viper]

a gta san andreas, artwork, ps2 style, [masterpiece], unusual color palette

[flat pike]

mhhhh a gta san andreas ps2 Japan Style variant maybe?^^

[summer viper]

a gta san andreas, artwork, japan style, [masterpiece], unusual color palette

[flat pike]

hehe tokyo tower in the back 🙂 nice

[summer viper]

yeah

need a negative promp "double vision"

[exotic summit]

I was curious if you had any knowledge on how Pebblely.com is addressing this issue with their AI-generated product images? I was wondering if they are using any specific techniques or methods to prevent objects from sticking to the product images...any clue?

[smoky oak]

My new ultimate SD upscale workflow is showing incredible promise. Maintaining composition with massively improved realism and accuracy

ultimate SD Upscale supremacy

[smoky oak]

man, this new upscale workflow is insane

[static tusk]

the first one of the capybara, a lot of detail was lost in the face, whiskers etc. the rest look great

[smoky oak]

the capybara may look less detailed on the surface, but the detail is much closer to real life cause their fur is a lot more fine than in the original image

[static tusk]

discord image compression doesnt help. The larger image looks better

[shut sinew]

what are you refering to?

[smoky oak]

yeah, its hard to see the detail from the thumbnail

[summer viper]

DPM2

7 minutes

[summer viper]

DPM++ 2M Karras

[wispy nest]

blackman talking with whiteman at doorway

[flat pike]

i like this one oxy , good job 👍

[summer viper]

I'm still working on this style

but it all comes down to the limitations of the resolution of the 1.5 model

I'm trying to learn how to work in the 2.1 model

overdid it a little

[smoky oak]

Alright, so, if I were to make a guide for how to replicate my high quality results out of ultimate upscale. Where would you all recommend I upload it to? I am not sure how to go about this, but this process works so go that I wanna share it

[static tusk]

post in #1003034183716835418 and have a mod pin it

just put it all in one message

[summer viper]

I did not know that this model is much smarter than 1.5

[stoic pivot]

Sleep help the human mind to learn from experience, to sort information so it can remember, and recall what it need. AI do not do that, is that why AI are such an assholes and never ask if you slept well when coming back online?

[abstract quiver]

A colorful watercolour underwater featuring various types of marine life such as fish, sea turtles, dolphins, and whales, white background

[flat pike]

just for the Lolz

[floral flicker]

@smoky oak If you haven't tried Multidiffusion since they incorporated the Region Control feature, and/or the Noise Inversion feature, I recommend trying it out again

The region control feature let's you assign a specific positive and/or negative prompt to specific regions of the image during upscale process, so you can have better control over the things you want added details rather than just quality improvement.

I haven't played with the Noise Inversion feature too much yet, but from my testing it does a very good job of enforcing the original composition without changing things

[crystal delta]

FAQ: How do I use the Stable Diffusion website & what do all the settings do?

Check out these videos for a great overview of the website and how to use it! https://youtu.be/014J2Yo1aGI ｜ https://youtu.be/M3jCa6qTHpQ

[flat pike]

Gaming Grandma Series 🙂

[hybrid beacon]

Ninja

[summer viper]

I sat for 3 hours and thought why the result is completely wrong, and then it turns out that I was sitting on "deliberate"

[stoic pivot]

I once said, "there is no such thing as 'worst quality', then I loaded Stable Diffusion 2.1... "red clay model Mickey Mouse".

"red clay model Mickey Mouse", Model FaeTastic_v1

[summer viper]

I thought I was working in "illuminati diffusion"

[pale garden]

Are you able to run control net with 2.1?

[floral flicker]

Yes, there are specific 2.1 models

https://huggingface.co/thibaud/controlnet-sd21

[errant dove]

Flying lion

[floral flicker]

Before you use them, you need to navigate to Settings > Controlnet in A1111. Change the first item, from cldm_v15.yaml to cldm_v21.yaml

You'll need to flip flop back and forth if you change from 1.5 and 2.1 models. Or, you can do what I did... make a duplicate webui-user.bat, and a duplicate ui-config.json, so you have 1.5 specific UI settings and 2.1 specific UI settings

(use --ui-config-file in cmd arguments)

[pale garden]

Thank you

[floral flicker]

np You can also use that to adjust other relavent settings, like keeping 512 default image sizes for your 1.5 setup, and 768 defaults in 2.1 setup

[untold pendant]

Fight between tigers and dragons

[floral flicker]

Now that is stirring up some nostalgia vibes

[summer viper]

Yeah

[slow kestrel]

Please can someone help me understand the prompts that were used to create this? I've tried CLIP interrogator & ControlNet but I'm unable to recreate the lighting & shadows.

[floral flicker]

If you were able to generate the image without the extreme shading, I imagine that would be a very simply photoshop gimick to replicate

Probably not the answer you're looking for, maybe someone else can be of better help

hmm this is kind of annoying... on one PC, if I change the default image size for say, txt2img width/height in the ui settings file, it will be reflected in the A1111 ui

But on this PC, I change the number in the settings file but it doesn't reflect in A1111 ui

OK I found my problem

[knotty sapphire]

how do you generate art on here

[dense tapir]

On here you don't.

[knotty sapphire]

is there anywhere to generate art free. or should i buy a midjourney account

[dense tapir]

No idea and MJ is a joke

Almost all on this server gen locally

[knotty sapphire]

fair, i have no idea how to do that, why is midjourney bad

[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!

[dense tapir]

It is inferior especially with SDXL about to be released.

besides it isn't free and I love free

[knotty sapphire]

are u running stable diffsuion for free? and able to gen

[wispy nest]

That's main point, kinda...

• you have alot more tools and options to play with, not just discord bots and not just gen

[dense tapir]

open source vs closed source

[knotty sapphire]

how do i go about setting this up on my mac? i appreciate ur guys suggestions. im new to this and was going to buy MJ

[wispy nest]

idk about mac...but a1111 has instructions on that

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Installation-on-Apple-Silicon

other webui's probably too

[rain gazelle]

nice i really like this!

[rain gazelle]

ooo i like this Mickey

[sterile kiln]

wow, how did you do this?

[dense tapir]

2k steps too #🏞｜general-with-images message

was really horrible before 1900 steps

[smoky oak]

That sounds like a ton of work to try and match the results of ultimate upscale, but I will at least try it and see if it has any utility outside of that

I have found a workflow using ultimate upscale that results in basically perfect upscales from what I've been able to do
It requires zero high res fix, which means that you get the exact composition from the original generation, rather than having to screw the entire thing over for the high-res fix

[fickle gorge]

@smoky oak That capybara was impressive

[smoky oak]

Thank you :>

I should upload it right now while I still remember

@fickle gorgeHere you go. Trigger word is CapBar

made for 1.5

[smoky oak]

Man, my new upscale workflow is working fantastic for stylized and textured images too!

[lilac pilot]

Is canny + depth able to get the blur out hand or do I need to properly prompt it?

[night sonnet]

https://media.discordapp.net/attachments/1078563889886789732/1092408257039040632/Screenshot_20230403-061950_Discord.png Emad, from SaaS to NDAs, have u lost ur mind

[smoky oak]

whut

[ripe cedar]

I made a pose with openeditor and did send to img to img. Put in a prompt but when I hit Generate, it says AttributeError: 'NoneType' object has no attribute 'convert'

[floral flicker]

When you do have an opportunity, let me know

[lilac pilot]

Were going to the moon again boys! NASA Announced

[smoky oak]

I am trying to make a guide on it, cause its showing the best results I have ever seen in terms of adding new information, while maintaining an extremely close composition

[floral flicker]

I mean you said you’d take another stab with multidiffusion, was wondering about an updated opinion

[smoky oak]

Base image Vs High res fix Vs my upscale workflow

[ripe cedar]

Nice, I'll have to make a note so I remember to bother you next month 🙂

[lilac pilot]

I believe img to img needs a base photo

[smoky oak]

I doubt it will take that long, I just need to figure out where I wanna post/document this process, as it is a several step process

process works great for making more realistic and reasonably flawed human faces as well

[lilac pilot]

@ripe cedar You get it?

[ripe cedar]

Yep ty got it

[lilac pilot]

Do you click send to text2img in the openditor or save it and upload manually?

[lilac pilot]

Cause my sendto is bugged and wont send it to either im2img or text2img

[ripe cedar]

It sent automatically for img2img. I uploaded manually for text2img

[hardy tulip]

Kyouko Sakura from Mahou Shoujo Madoka Magica

[scenic pewter]

Nahuel huapi lake with futuristic city on the shore, pedestrian streets and lots of parks between buildings in patagonia

[hardy tulip]

i honestly don't understand why people still try to prompt in the discord like this

[ripe cedar]

I put that in the positive prompt, left everything else as default. Using Mistoon model

[hardy tulip]

Better Image IMO

Yet another stunning image

i may have perfected the settings too much

[smoky oak]

You should see how much better it could get with high res upscaling :p

could have some even better results!

[hardy tulip]

you mean hires fix?

[smoky oak]

No, high res fix is trash

I mean like good upscaling

[hardy tulip]

i used SwinIR4x

[smoky oak]

ah, just basic pixel upscaling

it doesn't look bad, good job :>

I recommend looking into ultimate upscale, cause you could get fantastic 4k images no problem

[hardy tulip]

extension?

[smoky oak]

extension script

it takes a lot of time to learn how to use, but damn it blows all the others out of the water from my tests

[hardy tulip]

tested it

[smoky oak]

Yeah, you could use some tweaked settings. I working on a guide to get results like mine. Still ironing out the kinks

My workflow is showing improvements as such

It dramatically increases detail and accuracy, while also keeping a nearly identical composition and subject matter

Even works for sensitive subjects like realistic human faces

As well as stylized and textured generations

[sleek flax]

super cute cyberpunk bunny, orange juice bottle, pixar, zootopia, cgi, blade runner. trending on artstation

[brazen seal]

Scissors

[spare juniper]

A human listening to music

[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!

[dense tapir]

I prefer the second one

[red harness]

Can anybody explain to me why this wouldn't work as a control image in ControlNet? I even googled to make sure I had the X direction correct.

[static tusk]

increase the weight of the map in settings or use openpose together with the above

[red harness]

Actually, I noticed a folder called /models/ControlNet. Maybe I was supposed to put them there instead of where the tutorial told me to put them? 🤔

[red harness]

I have weight 1 guidance start 0 guidance end 1. What should it be?

[static tusk]

the first weight slider you can increase over 1, so that it uses the map as a heavier influence

try 1.5

controlnet models go in
stable-diffusion-webui\extensions\sd-webui-controlnet\models\

[red harness]

Yeah okay that's where they were originally.

@static tusk Do i use the Preprocessor and the model, or should I skip the preprocessor because the image is already a normal map?

[static tusk]

with the map you can set preprocessor to none

[red harness]

These all look awesome, but I can't tell if the aggressive face-bedazzling is intended or a bug.

[dense tapir]

Not a bug

Well, kinda a bug as it is smashing all my prompt and making that

[red harness]

What's your prompt?

[dense tapir]

deleted it but I think it did it because I said gem

it decided to gem the entire face

[red harness]

Yeah I dunno. It keeps treating it like color data. Prompt is woman in a green dress on a while background, along with all the infrastructure from prompts on the charturner page. Its first move is always to start with highly-scrambled blue noise, then converge into two copies of the same green dress, no person involved.

[static tusk]

enblle the preprocessor again and then click preview annotator button at the bottom, just to see if something is working

[red harness]

It says error. I can get Canny from a regular image, that's about it.

[static tusk]

theres your problem

Im not sure what is causing the error, youre gonna need to go through everything to see whats missing

[red harness]

Restarted it. Preprocessor: none, pretrained normal model, gives me the exact same normal map. That's what we wanted the annotator result to be, right?

[static tusk]

you want to test with preprocessor on to see if its doing anything

[red harness]

Are all these orange planet pictures for the same project? Because I can't read the intended tone.

[red harness]

It's taking in the normal map and spitting out the exact same normal map. Since it would normally be conveting a picture into a normal map, I assume any output that involves a normal map is probably a good thing.

[static tusk]

correct it will produce the same map as a preview. its working then I guess

[drifting mist]

女孩

[red harness]

I mean, it looks pretty. But other than the saturation, it looks like it belongs in a totally different universe from the orange bipedial robot above.

[red harness]

And this one feels like it belongs in a third universe, just from the colors alone. Not that a world fulla willy wonka tinker toy robots can't have pretty sunsets and post-apocalyptic wastelands, but more because no human author would write a story that involves both of those scenes side-by-side together.

Wait do I need to prompt Control Net to get it to trigger? I've been assuming just adding the control image will do the trick.

[static tusk]

there is a checkbox to enable controlnet, then you just add your image and choose the preprocessor and model

[static tusk]

but of course in your situation you have the map already so you dont need preprocessor

[red harness]

Oh whoops. Enable wasn't checked...

🤦‍♂️

[red harness]

Okay that just looks badass.

[smoky oak]

Happens to the best of us lol

@red harnessYour name makes me think of that spongebob meme

[dense tapir]

🙂

[smoky oak]

https://tenor.com/view/ascension-sponge-bob-gif-15077046

this one

[dense tapir]

All of these are using the same seed just seeing what happens

A bit weird

There we go it just needed more steps

[red harness]

I like this one. Not sure why. Seems chill.

[red harness]

I've been doing all this with txt2img, because that's what everybody else's prompts I saw on civitai sounded like. Should I be using img2img in order to invoke ControlNet?

[red harness]

Any ideas what I might be doing wrong?

[smoky oak]

SD isn't gonna gen well at all with that res

[red harness]

Will it change the result, or just be small and blurry? At this point, I'm trying to just get it doing the thing. I'll see how big I can make it later.

[smoky oak]

There are tons of things wrong about yours etup, but I am so out of it right now I don't even think I could explain it properly

[red harness]

Also there's like ten versions of CharTurner and they all have the exact same filenames.

They're all called charturn.pt and charturnv2.pt, regardless of what type of model they actually are. Meanwhile, people are using stuff like "charturnerv2For1_5" in their prompts.

I have no idea which version I downloaded and installed yesterday, and retracing my steps online doesn't help because I looked at all of them before making a decision. And I can't stress this enough: Why are they all named the same thing!?

[dense tapir]

I tried to upscale 😦

[red harness]

heh.

Good to know it's not just me.

[static tusk]

in your image above. see down the bottom it says used embeddings. charturner isnt there. Its not used.

[red harness]

I mean... 🤔 Clean up the seams a bit in post, and that would almost work as the background for a Getting Over It map.

[smoky oak]

oh lord your settings are way off 😅

[smoky oak]

I can help in DM's if you want @dense tapir

[static tusk]

check folder \stable-diffusion-webui\embeddings\ for the charturner name, use the same thing in your prompt

[dense tapir]

Took forever too as in 20+ mins

[smoky oak]

show me your settings in DM's. I can get them working reliably

[red harness]

I changed it to charturnerv2For1_5, which is what some guy's prompt used, and now it says it's using charturnerv2 alongside easynegative. So that's an improvement.

Or is it ControlNet that's not being invoked?

[dense tapir]

Getting closer

looks gnarly dirty

[summer viper]

enb gta 5

gta 3

[static tusk]

youre giving me the urge to go back to doing alien craft gens again. hahah. I already have hundreds, but its so much fun

[summer viper]

is it possible to merge sd 1.5 with 2.1

[static tusk]

nope

[summer viper]

😩

[static tusk]

you can gen in 1.5 and img2img in 2.1 though

or vice versa

[summer viper]

the problem with these pictures is a repeating pattern

[dense tapir]

best I can get

that one

wth?

oh, well

moving on

[smoky oak]

I am trying to make thumbnails for all of my LoRA's cause it seems fun lol

so far, I have 2 lol

[red harness]

I did that for a while, then I got paranoid all the thumbnails were eating up my precious ram or vram, so I stopped.

[smoky oak]

oh wow lol

[summer viper]

omg

[red harness]

Oh. You literally can't go smaller than 256 pixels. That explains some of the weirdness...

[summer viper]

I think I found a solution here: https://github.com/hako-mikan/sd-webui-regional-prompter

[summer viper]

try to make a city

[summer viper]

turn it over

[summer viper]

pattern

he is very noticeable

[dense tapir]

welcome to SD

[summer viper]

hehhhhehehe

2.1 works more unpredictably than 1.5

😩

[dense tapir]

This IS 2.1

[summer viper]

wow

[smoky oak]

2.1 is a huge disappointment. We can only hope that SD 3.0 isn't

[dense tapir]

Hiresfix used to help that without needing to upscale but then auto fucked it

[red harness]

If anybody knows what they're doing and feels like trying to get my normal map working with 1.5, ControlNet and Charturner? Hit me up. I can make a normal map for you to use in your own projects.

[polar sparrow]

A shih tzu on the beach

[red harness]

Huh. Apparently I was mistaken about the normal map conventions. The preview images on the internet are wrong. Apparently SD doesn't use OpenGl or NVidia format-- it uses BGR.

(There's a checkbox to convert, but still.)

[versed hedge]

#1072220168534642768 Donald Trump

[red harness]

Made my own converted in blender because it's about the process. Didn't help though. I'm downloading the "full" normal model now. Maybe the problem is the pretrained model just doesn't work in my use-case.

[crystal delta]

@versed hedge

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!

[static tusk]

using the system info extension benchmark to test gen speeds. I found that the 2.1 model is slightly faster

[summer viper]

my custom model based on "illuminati diffusion"

still not perfect

I've run out of space

[silk torrent]

Does the PNG info tab even work anymore? I don't get any params for any images I put in there thats on this discord?

All I want is a example of how this SD2.1 craziness works because all my images are neon messes

[static tusk]

discord strips all that info now

[silk torrent]

Ahhh is that why

Oof

[silk torrent]

Where can I find a good example image to ref :/? Anyone able to help because I can't figure out how to get 2.1 to work

Or a guide link somewhere perhaps?

[static tusk]

check civitai, filter by 2.1 models and see what are in the examples

but I would recommend not using the base 2.1 model. Get custom models for 2.1

[silk torrent]

Models people already trained?

[static tusk]

as in checkpoints

or safetensors

[summer viper]

nartfixer, nfixer

omg

[smoky oak]

hehehe

fart fixer

sorry, had to

[summer viper]

heheheh

[summer viper]

almost perfect

and where can I put my model?

[minor shuttle]

Combined 3D with stable diffusion. Fun workflow, just getting started

[summer viper]

only nfixer

renderzoov21, nfixer

[static tusk]

looks cool. does it texture all sides or just from camera perspective?

[minor shuttle]

Thanks! Just from camera perspective for now.

With time it will be real time from all sides.

[static tusk]

ah ok. Last time I checked it out it was the same. I was hoping it progressed to full normal texture

[minor shuttle]

It’s possible to manually do it. Render out an image sequence from 3D - depth map and other passes for controlnet and then just generate each frame with SD.

Is there an automated way of doing that in SD? Like for videos etc

[static tusk]

Deforum for video sequences

[minor shuttle]

Sweet, thanks 👌🏻

[smoky oak]

sick, I now have profile images for all of my LoRA's

[smoky oak]

I am so proud of these 4 LoRA's

[summer viper]

anime?

[smoky oak]

Anime?

[summer viper]

i like realistic

[smoky oak]

I do too

none of these are anime tho

[summer viper]

very similar

[smoky oak]

I have done plenty of realistic LoRA's too

My capybara LoRA came out shockingly good haha

the sheer realism I can get out of that one floors me

[minor shuttle]

Nice! Do you have these available somewhere?

[smoky oak]

I only have 2 available right now, but I will be uploading some more soon

Capybara LoRA is soon to release

it really does damn good 😅

I wasn't expecting it to come out that good

[minor shuttle]

Yeah looks great. Wish these were real, it would make a good pet haha

[smoky oak]

What?

They are real

there are people that keep them as pets lol

[minor shuttle]

Haha got u

[smoky oak]

they are the biggest rodents in the world, and they are super chill

https://tenor.com/view/capybara-gif-26113715

[minor shuttle]

Yep from south america

[smoky oak]

I got to see "mexican capybara" in person

they aren't actually capybara, but they are also really cute

Little cuties

They also had these super super cute little goofs

Mexican Coati, thats their name

Mexico was amazing. I am going back to see my Abuelita in June. So excited

[floral flicker]

Heres a quick thought - if you could get it generate without the lines on the road, you could quickly draw some in if you are good with a perspective tool. Like adobe illustrator > dashed line in perspective grid

Low denoise inpaint would then make it perfect

Could also just remove the existing lines easy enough really

[river cedar]

testing 1920x1080, rip gpu

[true coyote]

I have auto gif playing disabled and I thought it was generated image 😭

[smoky oak]

Looool

@true coyote that one isnt generated, but this one is

[true coyote]

It's really good too, but god damn, if the one I thought was generated really was, then I might as well just start preparing for singularity today hahahah

[smoky oak]

Just wait until I get my hands on SD3.0 hehehe

[true coyote]

Just wait until GPT-5 releases along with SD3 support

[smoky oak]

It would never release with stable diffusion support, because open AI is a illegitimate and fraudulent company for profit hiding behind the guise of a "for the people" conpany

[true coyote]

You just ask GPT for an SD prompt and it comes up with some gibberish for a human, but the one that SD understands perfectly

[true coyote]

Yeah I know. I'm just whatiffing for funzies

[smoky oak]

Understandable lol

As I am sure you can tell, I really dislike Open AI and their terrible and illegal practices

I look forward to the day when they get sued into Oblivion

[true coyote]

Yeah, and that you are also very passionate about it hahahah
I don't really mind myself. I'm too selfish and just care about a """"better"""" future

[smoky oak]

my issue with them is that they are AI "for the people", and were funded and founded to be just that. yet they refuse to give any of their stuff back to the people under the disguise of "the market is too competitive, and we can't lose our edge"

[true coyote]

Oh yeah, that I totally agree with you

I hate it when people are deliberately slowing down the AI progress

[smoky oak]

it is legit against their founding. So I look forward to whosever sues the hell out of them

[true coyote]

Same when people started signing petitions to slow down the research and put restrictions on it

Like, no, why you doing this

[smoky oak]

because if all of the public fear, which is 99% unfounded, and is based on false/misleading info

[true coyote]

Yeah, same with the "AI will enslave us" believers

Like, bro, just turn off the TV plz

Or "AI will become sentient and will turn against us"
Why would we even want to make it sentient
We make AI so it works for us and make our work easier
This is literally modern day slavery
Why would we even want it to be sentient in the first place

[wispy nest]

on the topic of LLMs...quantized version of Vicuna 13b is out. Super fast, pretty dang good and runs easily on a 3060 12gb. huge leap forward for local consumer LLMs

[fluid terrace]

Hello

[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!

[wispy nest]

(that wasn't directed at anyone in particular, just hadn't been ran in a while that i could see and it's usually a good idea to run it anyways lol)

[crystal delta]

FAQ: What are the server rules?

The server rules can be viewed at any time in #✍🏼｜rules-and-tos

[fluid terrace]

Rarther be enslaved by robots then humans

😂😂😂

[true coyote]

Oh hell yeah

[fluid terrace]

👍

[wispy nest]

to be fair, at least the robots should have a logical reason for the enslavement. that's not always been the case with humans

[fluid terrace]

I agree

[river cedar]

it can do fingers, repeat, it can do fingers

[fluid terrace]

Nice

[river cedar]

not nice, its bad

[true coyote]

Humans enslaving humans 😠
Humans enslaving robots 🥰
Robots enslaving robots 😐
Robots enslaving humans

[river cedar]

im making even better alot

[smoky oak]

Without my Opossum LoRA

[river cedar]

stable diffusion enslaving chatgpt when

[fluid terrace]

Lets gooo !

[smoky oak]

With my Opossum LoRA :>

[river cedar]

why 8 and not 9

its just so triggers me XD

this black square

[smoky oak]

you have to break the limit to get to 9

though, I do have mine broken lol

[true coyote]

@river cedar don't worry, your brain will inpaint it

[fluid terrace]

That fur is strange

[river cedar]

wdym

[smoky oak]

8 is the max for a standard batch size

[true coyote]

Yeah, it looks like it just took a shower with the most expensive shampoo

[fluid terrace]

😂

[river cedar]

ah, change it in the config file

[smoky oak]

its actually pretty accurate, believe it or not

[river cedar]

what model u using

[smoky oak]

They are floofy

[river cedar]

im using deliberate, what u using

[smoky oak]

I am using realistic 1.3 currently

[fluid terrace]

I google it you are right

[river cedar]

send me ur prompt and negative ill try with deliberate

[smoky oak]

They are very frizzy haha

[true coyote]

Oh damn, okay, the more you know

[river cedar]

lol

[smoky oak]

I have deliberate, I can test it if I want to, but it wasn;t trained for that

[true coyote]

No way guys
It doesn't start with "as an AI language model"

[wispy nest]

(left my typo in there for accuracy)

[river cedar]

its goo anyways as ive seen

[smoky oak]

Deliberate is fine, yeah

definitely not bad

[river cedar]

bro my pc is gonna die

rip my gpu generating 3440x1440

[smoky oak]

rookie numbers :p

[river cedar]

not at all

i remember generating 8k

it took like hours, but still it worked

[smoky oak]

You just need to use a more efficient workflow is all

I did 16k in around 18 minutes IIRC

[river cedar]

what gpu

[smoky oak]

15360x6532

[river cedar]

gpu

[smoky oak]

3060ti

[river cedar]

bruh how

[smoky oak]

just have to know what you are doing lol

[river cedar]

i have 2060 rn, i may do it in my homeserver

likee

i have a100 in there that i bought broken for very low price and fixed

48gb vram, yee

xformers are great btw

@smoky oak btw send ur 8k result i wanna look

[true coyote]

Same

[smoky oak]

I can't send the full file as it is too big, but I do have side by side comparisons

[river cedar]

screenshot it

and then send

[smoky oak]

Original base image

[river cedar]

cool

[smoky oak]

You can't screenshot an 8k image lmao

[river cedar]

said who

[smoky oak]

well, this one is 16k

[smoky oak]

said... common sense?

[river cedar]

1. open it
2. screenshot

[smoky oak]

😐

[river cedar]

what?

[smoky oak]

yes, let me take a 1080p screenshot of an 8k image so you can see the 8k in 1080p

[river cedar]

yes

[smoky oak]

...

[river cedar]

do it, i just wanna see general result

[smoky oak]

anyways, as I was showing

here is the zoomed section of the original base res image

vs the upscaled comparison

[river cedar]

uh, its not sharp

ok cool

sd upscale?

[smoky oak]

cause its a base res image, like I said 2 times lol

[river cedar]

yeah

[river cedar]

idk why you'd want to upscale it to 16k but still, sd upscale?

ai is amazing

[pseudo aspen]

Easter is almost here. Having fun with SD

[stoic pivot]

Realism and ControlNet are uncanny, some friends that know I use IA send me images and ask, "Can you make this look realistic", this time it should be an anthropomorphic dog, but I let Img2Img to make a prompt suggestion and it said it was a girl character, without thinking I simplified the prompt but removed "character" rather than "girl".

Well, this was the result and I do think the AI character is creepy in some way, maybe for arms, legs, body, had and so all have different proportions.

[floral flicker]

well theres one pupil a little higher than the other, and that is a devious smile

and a pretty odd pose for the hands

To make that first image look more realistic you'd probably be better off using Depth controlnet, lots of inpainting to focus the different things...

[smoky oak]

Bruh lmao

[stoic pivot]

Im working on it, but today my brain want to rest, or I just pushed to few buttons 😄

[floral flicker]

hyperrealistic

[smoky oak]

I am truthfully getting so infuriated by the sheer amount of false information out there that is being perpetuated by people talking about AI.

"all it does it copy"

"Its just stealing the work from people"

"It can only make what its seen"

Oh, so it works like every human artist in existence you say? Dang...

except it can also extrapolate, just like humans can. It doesn't have to see everything in every way to be able to infer what certain things look like. And thats something nobody ever wants to talk about

[dense tapir]

Yeah

Of course they would say that not all humans can do it while all computers can giving all humans the ability and that is bad since "muh job" is in jeopardy.

[smoky oak]

so many videos with "this is how it actually works"

and then its not how it works lmao

the biggest one is when they see "you need hundreds of images in order for it to make something consistent"

or whatever. Which is their gatcha for when it can output an image that looks similar to any generic artist out there.

"Oh, they must have specifically singled this artist out right here with 100's of images of selective focus"

[dense tapir]

they would also say that the computer can do it in almost a blink of an eye while a human can take weeks to months

[smoky oak]

Exactly, but people don't feel that way about 3D rendering or simulations lmao

[dense tapir]

They did at first but it passed which this will too

[smoky oak]

As somebody who has done tons of renderings and hundreds of simulations, I put more work into SD than most of the crap I did in blender lol

[grizzled sage]

extrapolate is one of my favorite computer vision terms. ever since enemy of the state

[smoky oak]

If people actually knew how this crap worked, I am sure they would feel different

"generative AI should be called copying AI, cause all it does is copy"

OMG OMG OMG

THE POW is SYNESTHESIA

I have that!

:O

Oh man, now I wanna make something that I have no idea how to achieve lol

[pseudo aspen]

I hate this kind of people complaining about AI art and not knowning anything about it...

[grizzled sage]

https://youtu.be/QZXUpSRv9do i swore they said extrapolate in this scene. i'm so mandella'd right now

[smoky oak]

hahaha

Mandella Effect at its finest

[pseudo aspen]

Lmao

[smoky oak]

I wish I knew how to script man

[grizzled sage]

People won't have long to complain about their ai misconceptions before the machine learning revolution sweeps thorugh all business sectors in ways no one can imagine yet. The initial sparks of this one are cracking louder than the dotcom boom

if the dotcom boom was Vesuvius, this is yellow stone

[smoky oak]

quick question, do you guys know how people generate reactive/blending videos in stable diffusion?

like, is there a program you use or something?

with frame interpolation so the motion is smooth I mean

[grizzled sage]

deforum. if by reactive youmean, to music? theres a bunch of tools on that one

its complex

people use ebsynth too, which i think deforum might tie into? i am unsure here but ebsynth smooths stuff out with keyframes

[smoky oak]

Hmmm... This is a lot of stuff to learn in the submission window

[dense tapir]

fuck, the extension to allow loha/locon he broke (at least for me) so now I can't do jackshit since I can't roll his back.

good, it isn't just me looking at another ticket

This does get tiresome as you have to be afraid to update auto, and really afraid to update extensions because a lot of those extensions don't have previous versions I could find.

Found a previous fork and it worked.

[smoky oak]

@dense tapirdo you know how to re-weight LoRA's?

[idle spruce]

not gonna lie, pretty cringe to be putting that up.

nft profile pic too very cute.

(retweeted by stable diffusion in #1017419581977022474 for context)

[dense tapir]

No, but not sure lohas can

[idle spruce]

what?

[wispy nest]

That's some strong genetics lol

[blazing zodiac]

crea un pupu que sonrie

xd

[flat pike]

this one come in my "best" ..i really like it

[dense tapir]

Aye

[flat pike]

its wisley fitted for upscaling to 1920 x 1080 with 2x upscaler^^

[flat pike]

ahh fresh space pics^^

[dense tapir]

yep

[flat pike]

one of my first pics XD

[dense tapir]

Nice

[flat pike]

once i started with space stuff, dont know why i dont prompting more from than

mhhhh....^^

maybe its time for new space pics

^^

[past zinc]

Foot golf course

"Cute cat"

[narrow yacht]

wow what the heck you've done to get such sharp version?

[smoky oak]

Thats childs play compared to what I know I can get now haha. I am using ultimate SD Upscale. I recommend looking into it and its documentation

[narrow yacht]

as soon as get an RTX GPU will

[vivid kettle]

Sea

[narrow yacht]

now i see its 15360x6532 right? 16k

[smoky oak]

yeah

I can't send the actual image, and there are a lot of issues with it as well (issues I know I could fix now if I wanted to, but I am not too inclined to spend the time on right now)

[dense tapir]

https://www.youtube.com/watch?v=A_l0plCryFg Damn

[fickle gorge]

We has capybara now

[smoky oak]

Ah, messing around with my LoRA?

[fickle gorge]

Yes, just started now, this one is from the first batch (like 1 minute ago)

[smoky oak]

I just made an opossum LoRA too, and best part is, all of these animal LoRA's work on anthro/furry characters too

I am working on a whole huge collection of more exotic animal LoRA's to upload to civit as a collection

my Opossum LoRA

and as you can see, works great with anthro/furry characters too

Christmas Opossum

@fickle gorgeLet me know if there are any other exotic animals you would be interested in me training. I have a whole list at the moment

[fickle gorge]

Thanks

[smoky oak]

I am training a ferret one right now. Should be done in sub 10 minutes

Alright, time to test the ferret LoRA

[fickle gorge]

With Realistic Vision 2.0, the results are very good

[wispy nest]

Have they made changes to the LORA training process so that, by default, they don't require triggerwords to work effectively? I know it was possible before(epinoiseoffset comes to mind) but most LORAs either didn't work well, or didn't work at all without the triggerwords. It's the main reason I don't pick up LORAs anymore simply because I don't want to have to keep track of storing and using the triggerwords. Used to run a sample image with the right words, rename it as a .preview.png, and then move it the LORA folder so i could load the png info to get the trigger words. Then i just started renaming the LORA to include the triggerwords (that way it was always with the lora itself). Eventually I just stopped grabbing LORAs unless I REALLY liked them.

[smoky oak]

It was trained for 1.3 real vis, which is almost identical to 2.0, so that makes sense

[smoky oak]

All of my LoRA trigger words are the name

So Ferret is Ferret, possum is possum, CapBar is CapBar, so on

You can do them without trigger words, but then you have less control

[wispy nest]

eh. bummer. You've kept it simple, but I long for the day when all LORAs just don't need them and will load like TIs by just adding it to the prompt

[smoky oak]

You... Do realize adding a TI to the prompt is the trigger word right?

TI's use trigger words too

and LoRA's have way more control, because they can have more trigger words

[wispy nest]

I do. I also realize that i can autocomplete that with the extension or even jsut type it myself as opposed to having to automcomplete and then also add trigger words

[smoky oak]

TI's are good if you want one static concept that is very consistent, but they are not good for diverse things

[wispy nest]

i'm not saying it's a super huge step or anything, i'm just saying that i'm lazy and would rather just not use the LORA at all in most situations because of the extra small steps

[smoky oak]

I think I am misunderstanding. Adding a LoRA trigger word and a TI trigger word are the same

in this case its legit just "ferret"

Which was a success, by the way

came out good

@fickle gorge Ferret LoRA was a success

[wispy nest]

you don't need to call it with "lora:FerretLoraName:1 ferret"?

[smoky oak]

Time to test if it works with furries

[smoky oak]

yeah, but thats autocompleted with a single click, like a TI

You just click on the card, like you do with a TI

[wispy nest]

adding the lora is autocompleted. the trigger words aren't unless that's changed very recently

[smoky oak]

Depends on how you prompt it

the training, that is

style LoRA's don't need trigger words, but having it for specific items/people/concepts is the better way to do it, as it has more control over what its making specifically, rather than just averaging the training data every time

[wispy nest]

Yeah, so it works the same as it did the last time I checked. I'm not saying the results aren't stellar and well worth it for using the LORAs as intended, but i'd prefer to be able to mess with them the way I do with TIs. I rarely use TIs to generate the intended result and instead mix them like ingredients to get some wacky stuff sometimes. That combined with the extra little steps required to call them properly just turn me off with them

[smoky oak]

you can do the same to a much higher degree with LoRA's, cause they have several trigger words

[wispy nest]

again...if I want to keep track of trigger words -_-

[smoky oak]

you can have several different concepts and styles in a single LoRA controlled specifically by what you ask

I mean, I guess

giving up a much better tool for a very slight inconvenience

[wispy nest]

The reason it's such a bummer, for me at least, is because i can currently go to civit.ai and literally pick up every single TI that's available and it's ready to go out of the box. withI do this every couple of days and the only thing i need to do to some of them is rename them because the creator chose too common a name for the TI that's going to get unintentionally triggered. LORAs I have to document the trigger words in some way. if they made it so that when I click the LORA card or use the autocomplete extension and it automatically filled in all the trigger words without having to do all that manually, it would be amazing. until then...unless there is something very specific i'm looking for it's just outside my desired use case

[dense tapir]

I just trained a few style loras and everything I have posted used them without a need to call them. If I do call them their effects are far greater than just turning them on.

[smoky oak]

Thats my point. Not calling a LoRA with trigger words is just shooting yourself in the foot really

[dense tapir]

not for styles but with characters/subjects YEAH

[smoky oak]

it just nullifies everything, unless you way overtrain

also, ferret LoRA does indeed work with Anthro/furry models

business ferret

[dense tapir]

What I hate about even embeddings is that you never know what will come out depending on the model used. Same with Loras

Of course I am talking about styles

[smoky oak]

typically speaking, if you keep them in adjacent models, they are very predictable. At least with subjects

[wispy nest]

That's actually what I like about them -_- change the model and sometimes get something completely different. reminds me of calling artists...some models have huge differences in what calling a specific artist will do

[dense tapir]

Oh, I agree as the fun is in changing the models

Just when you are specifically after something someone else trained you are kinda fucked unless you know the model they trained it on

This is why I train my own styles

[wispy nest]

yeah. but i don't usually generate -very specific- things lol. i can definitely see why that would be frustrating if going for an exact concept

[dense tapir]

general concepts yeah, fun times but specific is frustrating

style training sorta sucks because it trains but may not be what you thought and taking 30m to 2h+ per training it really tears me up

[dense tapir]

Very hard to pin a style down in SD

[chilly lance]

Depends

if you can find a mix of art styles/artist prompts