🏞｜general-with-images

[feral pebble]

(i added lines because sd clearly was not understanding what was bone and what wasn't)

it would do weird stuff like that (and worse)

[hasty nova]

does it have to be photorealism if I do it?

Well darn then probably not. I could try with the in-the-works digital diffusion v2, but I am not a great prompter

[smoky oak]

Some vogue like model generating

[feral pebble]

@hasty nova i'm actually sending your image to my friend's son.. 🙂 What prompt did you use for it?

[hasty nova]

I can check

[feral pebble]

(not a photo of you.. not your image.. but the phoenix you made 🙂 ))

[hasty nova]

a dramatic cinematic action shot of a flaming phoenix, bird, feathers, flames, wings, particles, beautiful, digital art, magnificent
Negative prompt: low quality, messy, distorted, extra limbs, extra claws
Steps:15
Sampler: UniPC
CFG scale: 6.5
Seed: 1660526554
Size: 768x640
Model hash: d0d0902731
Model: digitaldiffusion_v1
Denoising strength: 0.6
Hires upscale: 1.8
Hires upscaler: R-ESRGAN 4x+

[feral pebble]

thanks a lot bud

[hasty nova]

I could try making it blue since it seems like you were trying for that

On the left is using v1 of my model, on the right is v2.

They are too normal bird like I thiunk

I am gonna remove bird from the prompt

Well here it is without bird

[hasty nova]

Not much different

Still pretty cool though

[rugged pollen]

H.R. Giger inspired

[hasty nova]

Thanks. I fought with hires fix for a while and eventually just generated on a higher res.

[brazen anvil]

respect the flow bro

[hasty nova]

the hands are not great here

but it's a fairly simple prompt

Interesting. A wider image made the person bend down.

[brazen anvil]

I'm going to try it c:

[hasty nova]

Thanks, please share what you make! I am about to release v2 hopefully tonight tho

With improved photorealism, as you can see in the images above

[hasty nova]

Though it can still do digital art

[smoky oak]

lord help me, for I have learned how to make cute men

😳

[smoky oak]

ah yes, the crissi crossi

[hasty nova]

otter you doing there

this one is perfect

I love him so much

[brazen anvil]

can you send me the config file? can't download it from civitai

[hasty nova]

Weird, here ya go.

But v2 is currently in the process of uploading to civit 😆

If you would rather have that one I would wait a few minutes, though there is no harm in having both I suppose

[brazen anvil]

haha ok im going to eat some tacos outside for dinner, I'll check it then when I get back

[hasty nova]

Oo sounds yummy, enjoy!

https://civitai.com/models/48858?modelVersionId=59420 TIS DONE!

[kind quartz]

i love funk art 🙂

[kind quartz]

contraweight to @dense tapir your dark images 😄

[dense tapir]

OUCH

I still have not seen a real dark image only a lower contrast one. Workable at least, though.

[kind quartz]

i mean dark by mood 🙂

[dense tapir]

Oh, I am more into dark foreboding art.

Is there a place for digital AI artists to post their work? "Real" artists have theirs but what about us or are we relegated to some remote corner of the internet (if it even exists) like some back alley in a remote red light district?

[static tusk]

deviant art

[dense tapir]

I thought they took an Anti AI art stance? Again, only one place though. Good to know if they are cool with AI art.

[static tusk]

they have a section dedicated for it

[wispy ether]

[dense tapir]

I still, for decades, hate that name. I am no deviant.

[sterile kiln]

Someone read something about SDXL models? When they will be available ?

[dense tapir]

soon, and they aren't just simple models you can plug into auto unless that changed.

Supposedly comes with new tools etc... but you need 8gb+ to use it. I think smaller models are due after.

[static tusk]

twodukes said they were using a modified version of sdxl for testing

[dense tapir]

aye

[static tusk]

so maybe not that hard to implement

[dense tapir]

all we can do is wait and see. I expect to not be able to run it and I am fine with that.

[random pendant]

is the bounty still up

what kind of model do you want to see wearing them. I dont get it

[wispy nest]

I want a model that can generate people wearing that sock.

[bold cave]

Sounds like a good excuse for me to deep dive into loras

Aight challenge accepted. Oh btw Stop Banning is my alt

I'll give it a shot when I get some free time

in the meantime enjoy my latest masterpiece lol

supposed to be billie eilish lmfao

Tonight Im uploading my master blend model to CivitAI. Ive been meaning to do so but it takes eons to upload 7gbs into a website unfortunately.

that looks nothing like billie

Ill share my model once it uploads and I guarantee it's better than the most popular ones out there

OH GOD MY EYES

LMFAO

kill it with fireeeeeeeee

[smoky oak]

Deep Floyd is out right now, which seems like an offshoot of SDXL

its able to do text, and they were referring to it as a full scale XL model, and said it will be available to run locally semi soon

@stone cipherI am finally at my favorite sort of mix between realism and slight artistic detail

I am quite happy with these results

[wispy ether]

@smoky oaki just made a actual lora now, how do i continue training THAT one instead of making a new one

[smoky oak]

Are you finally using kohya yet?

[wispy ether]

ye

[smoky oak]

alright, screenshot the UI please

[wispy ether]

i cant

its bugged lol

[smoky oak]

bruh

windows key, snipping tool

[wispy ether]

i dont wanna restart

yea

its bugged

it just loads

i can make phone pic

4%

[smoky oak]

alright, you can fix it. You wanna go into task manager and end task (file explorer)

I am not sure why it does that, but that's how I always fix it

[wispy ether]

but still

why tho

[smoky oak]

that will let you screenshot

so I can see what the options are

I know how to do it, but I don't remember what the option is

[wispy ether]

even the new lora is broken again

forward
lora_apply_weights(self)
File "C:\Users\Lukag\Desktop\stable-diffusion-webui-master\extensions-builtin\Lora\lora.py", line 273, in lora_apply_weights
self.weight += lora_calc_updown(lora, module, self.weight)
RuntimeError: The size of tensor a (768) must match the size of tensor b (1024) at non-singleton dimension 1

[smoky oak]

I have no idea lmao

[wispy ether]

do i have to verify the model?

[smoky oak]

no

Well I mean

I have no idea what you have done

[wispy ether]

ive only set the paths

[smoky oak]

again, never had a single issue with any of this

[wispy ether]

maybe its this

because error says tensor a and b

@smoky oakhow do i continue training 1 model once i created it

after quitting the gui

[smoky oak]

screenshot the main UI

[wispy ether]

why

[smoky oak]

bro, just do it

[wispy ether]

i cant

🗿

[smoky oak]

how the hell did you just send the image above? And I told you how to fix it

[wispy ether]

snipping tool

[smoky oak]

found it on myown

[wispy ether]

what

[smoky oak]

training parameters, advanced

[wispy ether]

what

[smoky oak]

resume from saved training state

link to the last version you trained, it will continue from there

[wispy ether]

shi takes forever

[smoky oak]

0 clue what any of that is, was never needed for me

[wispy ether]

the save state files

[smoky oak]

😶

[wispy ether]

after downloading

subprocess.CalledProcessError: Command '['C:\Users\Lukag\Desktop\kohya_ss-master\venv\Scripts\python.exe', 'train_network.py', '--enable_bucket', '--pretrained_model_name_or_path=runwayml/stable-diffusion-v1-5', '--train_data_dir=C:\Users\Lukag\Desktop\mdee\image', '--resolution=512,512', '--output_dir=C:\Users\Lukag\Desktop\mdee\model', '--logging_dir=C:\Users\Lukag\Desktop\mdee\log', '--network_alpha=1', '--save_model_as=safetensors', '--network_module=networks.lora', '--text_encoder_lr=5e-5', '--unet_lr=0.0001', '--network_dim=8', '--output_name=spongebob1', '--lr_scheduler_num_cycles=1', '--learning_rate=1e-5', '--lr_scheduler=cosine', '--lr_warmup_steps=14', '--train_batch_size=1', '--max_train_steps=140', '--save_every_n_epochs=1', '--mixed_precision=fp16', '--save_precision=fp16', '--cache_latents', '--optimizer_type=AdamW8bit', '--max_data_loader_n_workers=0', '--resume=C:\Users\Lukag\Desktop\mdee\model', '--bucket_reso_steps=64', '--save_state', '--xformers', '--bucket_no_upscale']' returned non-zero exit status 1.

🤷🏻‍♂️

@smoky oakcan you send me your config

and if it doesnt work then idk

[smoky oak]

Sorry man, i am kinda over helping with this anymore, and I am not looking to just hand over my settings at the moment

mmmm

[wispy ether]

i mean send the file

the .json

[smoky oak]

@wispy nest I am starting to really hone in on better realism >:D

[smoky oak]

I know

[wispy ether]

you said looking over your settings

not sending it

[smoky oak]

yours definitely has that home photograph feel, more vintage, but I think i am starting to finally cement my look in more modern DSLR type realism

I think we are both killing the realism scene right now :>

I saw that

👁️

[keen spade]

Hello

[wispy ether]

wtf

[smoky oak]

OH GOD I HATE THAT

[wispy nest]

Noice. Drop some firestarters. I see Pete is cooking well today too haha

[smoky oak]

I think we are both coming into separate but adjacent looks

[wispy nest]

I'm cleaning my closet drawers, not as exciting a morning haha

[smoky oak]

pea suh

[wispy nest]

I want some pizza with that peperoni

haha looks like a meat lovers dream pie 🍕

[tough monolith]

using lora!

[smoky oak]

modern style/quality, yeah

kinda

no, no

I mean the quality and processing of the images

[wispy ether]

is making a normal model just for spongebob worth it?

[smoky oak]

my custom model mix likes to give woman boob plates for armor lol

This is more of what I mean when I say modern realism

I also think it has to do with model differences as well

[wispy ether]

sd is so broken

even my spongebob model now generates random stuff

[wispy nest]

odd

[sterile kiln]

L'OREAL knight

https://tenor.com/view/cow-wig-loreal-hair-meme-funny-animals-blonde-gif-11349696

[smoky oak]

wait wait

[sterile kiln]

Hahaha, sad, i'm not a home to reply

[wispy nest]

lmfao!

[wispy ether]

i need better vae

i get weird fingers

[sterile kiln]

Probably just a better prompt

[wispy ether]

what prompt

[sterile kiln]

Your negative prompt is : EasyNegative, bad fingers, bad hands, extra fingers, deformed fingers

you just need to polish it

[wispy ether]

you said prompt

not negative prompt

[sterile kiln]

When people talk about prompt, it usually implies both

[wispy ether]

[smoky oak]

I consistently get really good hands with my mix model

just most generations don't show them off cause of the nature of portraits

[wispy ether]

check out the prompt

[sterile kiln]

Yeah, just a shit small prompt (sorry)

[wispy ether]

what

i said check out the prompt

bc its funny

[sterile kiln]

You have to polish it

[wispy ether]

,

this is not finger related

i just said check out the prompt

bc its funny

the prompt is funny

[sterile kiln]

among us, really funny

[wispy ether]

[sterile kiln]

👀

[wispy ether]

ok

[smoky oak]

getting a little too spicy there IMO

[wispy ether]

https://cdn.discordapp.com/attachments/942861254157029457/1102649232269267015/00255-3829226670.png

[smoky oak]

you know, I am starting to get annoyed with how these models default to white people for generations

and how the more descriptors you use, the more white they look

[wispy ether]

you can generate a black man

[smoky oak]

there's more than just black people

[wispy ether]

no

theres white and black

fym

[smoky oak]

what i am saying is how when you start adding generation details like "handsome" or "[retty" the models automatically assume you want them to look more white

[smoky oak]

what-

[wispy ether]

there isnt yellow

or green

https://tenor.com/view/hulk-gif-25068351

[smoky oak]

...

anyways

If I try to generate a person of an ethnic race, it so frequently just makes them look white, and its annoying

more specifically if I add descriptors like "pretty" or "handsome"

[sterile kiln]

in your poor narrow conscience, yes

[smoky oak]

their joke was white and black are the only "color" defined people

[wispy ether]

iron fist 😎

[smoky oak]

doesn't mean it was funny

[wispy ether]

no

have you ever seen any another color except white and black on this earth

[sterile kiln]

sorry but, they are bad, increase your prompts before posting here maybe?

[wispy ether]

i try

its controlnet

[sterile kiln]

DO it alone

[wispy ether]

I DO

FYM

bruh now he got a stone

[sterile kiln]

No, you just share every sh*t you done

[wispy ether]

yes

and

what does that have to do with prompting

[sterile kiln]

In this channel, you are supposed to post something worked out, polished, not the slightest shit you do.

[wispy ether]

and you could be more polite

briish

[sterile kiln]

Believe me I am. But you have been puffing us up for days, being haughty and pretentious, while people here are very cultured on all these issues. Try to find a Discord at your young age maybe

[wispy ether]

what

[wispy ether]

[wispy ether]

sd folder

[smoky oak]

rookie numbers

[hasty nova]

yall have over 100GB?

And here is my entire AI folder.

I'm efficient :)

[smoky oak]

here is all of my AI stuff

A few files

[wheat rock]

i got like 2 tb

on my pc, was cheap like 100$

isnt it standard to just buy like 1tb these days

[smoky oak]

I have 4.5TB, so, probably

5.5 if counting my HDD, but I don't have it in

I don't think I will ever use a HDD again

They are just too slow for anything I do man

[static tusk]

I see A1111 is now updated for torch 2.0

been using it for a while now. its good

[smoky oak]

does it have a speed benefit, or is it slower like before/ with vlad?

[static tusk]

the update is the same speed, but I didnt have any slowdown when I manually moved to 2.0

[smoky oak]

if its the same speed then... what was the point?

I remember when they said it would be like 30% faster, but it has never not been slower for me

and basically everybody else

[static tusk]

most people werent using 2.0

vlad is using something a bit different I think, but even then you were saying it had no benefit in some scenarios right?

the 30% is in reference to 40 series cards

using system info benchmark with extra steps and extensive options ticked

[smoky oak]

I have texted auto 2.0 and Vlad extensively, as well as comfy and even Volta ML, and they have always been slightly to much slower than normal Auto for me

Volta ML was an absolute nightmare, and it was also a monument waste of time

Vlad can be considerably faster, for some pretty big trade offs (massive caching processes that are annoying)

[static tusk]

did you test with SDP?

[smoky oak]

Yeah, that almost got to as fast as normal auto

[static tusk]

also with SDP you cannot have xformers installed or it wont be a proper test

[smoky oak]

And then that with torch memory and cuDNN, you can get about 15% faster than Auto, but for a pain in the ass compiling process

[static tusk]

for reference im using --api --opt-sdp-attention --opt-channelslast

[smoky oak]

All I know is I tested about every single combination in Vlad, and I could get faster, but for a huge trade off at other time

[static tusk]

with vram +1000

[smoky oak]

I'm not gonna risk updating to 2.0 for auto considering how bad it's been for me and many others up until noa

*now

[smoky oak]

Man I wish the arc a770 was more usable for AI @dense tapir

Cause I would sell my 3060ti to get one for the 16GB VRAM

I may have just severely damaged my PC -

It seems to be ok now, but at the time it was reallyyy messed up. I hope this isn't gonna lead to issues down the line. I can't afford a new PC right now and I am trying to move to a laptop as is

It seems ok I think from inspecting the inside and shaking it around a bit...

I have no idea what happened, but I do know that I cannot afford to replace this PC

[dense tapir]

I 100 percent agree. Battlemage then Celestial then Druid. My eyes are watching them closely. As a content creator they offer the least expensive av1 setup so a lot of creators swung over to them to use their offering as a second card.

My fear is that Intel had the worst lost ever so investors are reeling. That can't be good.

[dense tapir]

Interesting result

[dense tapir]

Damn :/

[spark stratus]

🫨

[woven dawn]

[spark stratus]

[woven dawn]

I see you found the emojis

[spark stratus]

Help me pablo

Jk

[ivory onyx]

#💬｜general-chat emoji, happy, smile

[brazen anvil]

hi all c:

[bold cave]

This lesbian kiss took blood sweat and tears

[hybrid geyser]

what's causing this eror in automatic A111

[wispy nest]

didn't google ban gradio from drive? or did they repeal that ban and i didn't notice?

[hybrid geyser]

no its working now 🙃

[wispy nest]

i mean you didn't get some kind of warning saying you are running illegal code?

[bold cave]

why would they do this

[hybrid geyser]

no working fine

[wispy nest]

they did it a while back stating that it wasn't being used for "interactive workflows" or something

[bold cave]

lmfaooo

google is officially the pussiest company alive

[dense tapir]

@marsh lily https://github.com/RadeonOpenCompute/ROCm/releases/tag/rocm-5.5.0

[wispy nest]

found it. banned back on the 21st. was only for free accounts though i think

[bold cave]

all I gotta say is. Im fookin glad to own a 3090. I aint nobody's bitch. I can run whatever I please without getting bent over by any companies

[dense tapir]

@marsh lily Apparently there is no 7900XT/XTX support with this. Seems a community support version needs to be released? I hope that is wrong, but that is a bad look coming from AMD.

[dense tapir]

I gave up on Colab about a month ago because that is precisely what I was. They gave 2h per day and everything took so long so to burn my time up. I was always rushed. Just not a good time for me with it.

You are welcome, and yeah, MI300 etc...

[wispy nest]

this turned out way creepier than it should be lol. i blame it on the eyes and lack of fingers on the one side

[neon crystal]

I’m in love with 2.1 👽. I’ve seen folks complain about it not being great for art and stuff but I don’t think that’s the case at all!

[bold cave]

I mean... if you like weird looking corny stock-image-looking pictures, then hey! Whatever floats your boat.

Oh btw. Here is the before and after of the lesbian kiss. Video TuT coming soon? 😏

No Ultimate Upscaler. Just good ol' fashioned ControlNet, Latent Coupler, Inpainting, Dreambooth, Photoshop Frankensteining and a lil' bit of elbow grease. 😏

[neon crystal]

Nope not at all. If you can articulate what you want or have some knowledge of art in general to describe properly, 2.X >>> 1.X. If you’re into generic fantasy stuff and have been dependent on using trending on artstation, rutkowski etc to do the job for you yeah it’s definitely not for you.

[bold cave]

https://tenor.com/view/memeblog-therock-rock-dwayne-gif-21928039

He can, as long as it's a stock photo lmfao

team 1.5 ftw

[neon crystal]

I use my own custom 1.5 models and stuff from civit.ai for photorealistic stuff yeah. I also use 1.3 for minimalist art stuff. What’s the point you’re trying to put forward though ?

[wispy nest]

that's kind of the opposite opinion of pretty much everyone who has used 2.1. most people agree that the model would be fine except for the text encoder not being great. you need a lot better negative prompting to get where you want

[bold cave]

1.3 lmfaoooooooooooo

[neon crystal]

Haven’t really explored photorealistic renders so yeah it’s possible you’re right. For art in general I found 2.1 to be better. Especially with following the prompts.

[wispy nest]

i didn't mean just for photorealistic renders. i rarely do those at all. i mean in general 2.1's text encoder isn't that great

the one thing i can say about 2.1 is that it follows textual inversion embeddings pretty dang well...but i actually enjoy seeing how TIs work with different 1.5 models

to be clear, i'm not saying you can't get great things out of 2.1, but it's a lot easier for a lot more people to use a 1.5 model

[neon crystal]

Cool got it. I only have a very high level understanding of the math behind stable diffusion and haven’t really grokked it but have worked with CLIP and Open CLIP and found Open CLIP to be better at encoding both text and images ( at least in the context of clothing and fashion )

Yep I’ve heard great things about textual inversion embeddings with 2.1 too. That’s what I’m gonna explore next

[chilly rune]

What the fuck is this?

[wispy nest]

looks like a botcheck

like the ones that say "Answer the following question with 7" and then asks like "What color is this car?"

[smoky oak]

This reads so narcissistic 😅

2.x is greatly lwt down by its step backwards text encoder unfortunately

I personally don't think more steps for the same output is a generally benefit

[neon crystal]

#okfreud

[smoky oak]

not saying 2.x isn't good, just its not as great as people who use it like to shill it say lol

[smoky oak]

like if these are the examples you are trying to use for how "photorealistic" it can be, them that's a pretty poor argument, IMO

[chilly rune]

I assumed it was some form of way to mess up the ai by making it so these rating things always come up at 3

[wispy nest]

Is there any other company that provides free GPU's ?

I thought all of them owned by google anyway?
Collab, Kaggle, paperspace?

We can blame them all day long , but at least they are giving free gpu's for some work...

[smoky oak]

Imagine thinking a company that is legit trying to give everybody access to cloud computing for the low low cost of literally free, is a POS because they had to put restrictions on people who were legit stealing power and resources from them in order to make their own money... Like... Google collab is a purely benevolent service that benefits everybody that uses it, and because some people decided to be jerks, now we all lost the ability to use collab

[dense tapir]

Entitlement.

[smoky oak]

Right.

the fact that google collab had a free tier that gives you more computational access than the resources used to just log you in is insane. At free, I expect nothing, and am delighted with everything from there on

[dense tapir]

I can't even train with it now because I ceased to use the notebook. I understand why they did it, and do not hold it against them I just know, deep down, they made it hell because of this until they had no other choice they pulled the plug.

[smoky oak]

but apparently people feel the need to be so entitled about getting to use free hardware that costs money, and using power that they also didn't pay for.

[wispy nest]

I'd still use it for non webui version of SD with basic functions, like model choice, vae, clip skip, cfg...
(But I don't think it exist)

[dense tapir]

Went from 12 to 8 to 4 to 2 and I figured up next would be 30-60m or nothing.

[smoky oak]

people abused their generosity until google couldn't afford to give handouts

[dense tapir]

notebooks are fine to use just not remote uis like gradio

[wispy nest]

I was still using it for 3-6hours , but I didn't use it often lately, maybe 1-2 times a week

and mostly at night

[dense tapir]

Oh, 2h is if you used it every day. I saved up some time and had it for 4 hours one day as I had a training that was over 3h

[wispy nest]

yep

[dense tapir]

took me two weeks of not using it to get 4h

[smoky oak]

like genuinely, imagine if any other paid service offered the capability that google collab does for their equivalence

like a free version of netflix where you get to watch everything you want with no charge for like 6 hours a day max

that would be INSANE

sure its not a perfect analogy, but some people are just so damn greedy lmao

[wispy nest]

technically spotify almost does that, with exception of quality

[smoky oak]

you know, thats actually fair, i can see that

they do still make money off adds, but adds are a good trade off for free, IMO

[dense tapir]

You know that free sandwich and a drink you give me every single day? I am pissed you no longer offer that to me unless I pay.

I demand my free sandwich and a drink like you used to give me. Damn you. Your service is crap and sucks.

[smoky oak]

the entitlement is insane lmao

[wispy nest]

Oh...ads, you're right, I forgot about them...
I don't have them , since well...spotify technically blocked in my country and It has no ads...

(But nope, I paid for it for a long time until I couldn't)

[smoky oak]

I no longer receive a daily charity handout, and now you are the problem! >:C

[smoky oak]

no I get you, no worries

I am trying to work in the music industry, and I have several friends who's entire livelihoods are based off of the money they get from their music (which is not majority from streaming)

[wispy nest]

Gpu's is a bit bigger deal to be paid with ads, I think 😦

[dense tapir]

Like that one lefty dude who sued his own mom because he no longer was allowed to stay for free at 26 or so. Never figured how someone could do that to their own blood and someone was being nice to you for all those post 18 years. Dude didn't even buy anything or do anything to upkeep anything. Just gimme.

[smoky oak]

oh goddddd

spotify pushed its new update to my phone and PC

god, it sucks sooo bad

actually so annoying

[wispy nest]

oh...I thought they cancelled it...

[smoky oak]

I don't know who the hell decided we need a tiktok style scroll for music video feed when we are looking for playlists

[wispy nest]

I have no update yet

[dense tapir]

Anyway, thanks to the services that popped up offering discord, and web based gens they were on the free Colab.

[smoky oak]

just updated mine so now my default playlist dropdown is all messy like this

[dense tapir]

Thanks to them we all suffer now.

[smoky oak]

ok, i will admit, this is kinda cool

list, and grid style playlist viewing is pretty dope

[wispy nest]

eh...idk, kinda messy from first view, but might get used to

[smoky oak]

not sure why the stranger things playlist got injected in there

looks like this with the side pop out

but can also be condensed and stuff

[wispy nest]

no...I'd still prefer when more info fit on left side 😦

[smoky oak]

they do still offer this view

[wispy nest]

That one is better, I'd take it

[smoky oak]

this redesign is much less bad when compared to that stupid tiktok one on mobile

actually infuriating to use

[wispy nest]

text+visual representation, while not being too big, looks good

[smoky oak]

I will say, this is actualyl kinda cool IMO, cause you can do single or multi filters, or use it as the spotify search through all of your stuff/spotify curated stuff

[smoky oak]

@dense tapir Looks like the 4060ti is gonna be a new low heh

actually disgusting

I mean, it literally can't be good

there is no way it can be, cause in order for it to be good, it needs to at bare minimum reach 4070 perf lmao

[dense tapir]

Yep

4050 will be 4gb for 499 and you will like it by God.

4060 will be back to its rightful place of 6gb

8gb ti version

welcome to 2017

[wispy nest]

sad but true

[smoky oak]

People are so annoying lmao

I mention that i am working on becoming an ML engineer, and suddenly everybody, their dog, and their mom are machine learning experts telling me why I will never make it, and how I need to know 20 coding languages

Like bro, one of my friends is already the head of his own department at a major company, and he is mentoring me actively. I take the word of somebody doing what I wanna do, over the word of the people who don't even understand what it is

[dense tapir]

It's the Internet.

It is what they do. :/

[past pelican]

Looking pretty nice!

[wispy nest]

yeah those are both incredible, love the depth on the first one

[wispy nest]

youtube changed design too when you start watching video

[smoky oak]

the glow effect?

[wispy nest]

eh,...it changed back lol, can't screenshot it anymore

basically - instead of comments below the video there was a list of other videos

and description on right side

[wispy nest]

But yea, I hate glow effect too, have to turn it off each time, it's distracting if I'm watching not full screen, it needs option to just turn it off globally 😦

[merry hornet]

Ur future girlfriend

[austere pilot]

[wispy nest]

been a while since i made a big critter, so here's a forest guardian (edited so it wasn't as washed out)

[smoky oak]

oh, I am pretty sure that only happens on premiered vdieos which had a live chat

side note

I absolutely HATE Civit AI and its trash search function

download a model and can never find it again cause their search is unusably bad

trying to use a model I downloaded, but I cannot find the trigger words

[wispy nest]

no, it was regular video, I even clicked same video I opened before - design isn't there anymore

[smoky oak]

hmm... very weird

[wispy nest]

oh...I hate models with trigger words...

had to keep notes for different models for a long time, now I don't care anymore, can't be bothered to use those models 😄

[smoky oak]

It is a really cool model that has a semi painterly style similar to midjourney, but I can't remember what the damn trigger word was

Found it!

God, you know your site is trash when I have to google the name of the model and it comes up on google instead

[wispy nest]

lol

[smoky oak]

Something is wrong with civit

Some of my models on the same page keep failing

and I keep having to jump start them

[wispy nest]

Something is wrong with my youtube lol, I kept testing and trying to find something that will trigger that another design

and I found it

but...I opened exact same video in another tab - and it has old design

[heady apex]

Kiran had been traveling for hours through the winding roads and lush forests of the countryside. As she made her way closer to her destination, she couldn't help but reflect on the challenges and triumphs of her journey thus far. Suddenly, the sound of an engine sputtering and grinding to a halt brought her back to the present moment. With a sinking feeling in her stomach, she realized that her car had broken down in the middle of nowhere. Determined to reach her destination, Kiran set out on foot, hoping to find help or a means of transportation along the way.

[smoky oak]

nice little story

[heady apex]

Thanks

[wispy nest]

Exact same video , different design 🐧

[smoky oak]

what the butttttt

oh me oh my I do not like that

[wispy nest]

That's new one

[smoky oak]

Actually

I can see the utility in that

[wispy nest]

I don't even understand where's comments now lol

[smoky oak]

read the comments without having to scroll to hell

[smoky oak]

oh wait what

OHHH

thats the description

WHAT

ok

[wispy nest]

2nd is old design, yea

[wispy nest]

yep

[smoky oak]

I wanna try out this new design for myself

I wonder why its doing that only sometimes

[wispy nest]

I have a suspicion on that...it might be cause I use vpn sometimes and watch youtube on different languages, google had bugged out before on me, turning on German language or giving me only Ukrainian subs, maybe sometimes it thinks I'm from one of the countries which test this new design

but then it goes back to normal

can't trigger it on purpose with vpn, it's just random

[smoky oak]

interesting

[wispy nest]

hm
https://piunikaweb.com/2023/05/02/new-youtube-ui-or-layout-on-desktop-site-met-with-backlash/

oh yea comments are there, under description, I'm dumb

so looks like they prioritize more videos over comments now when you scroll and just a little section initially for comments

makes sense, I guess

[brazen anvil]

Makio Fath 04 🔥

[wispy nest]

nice

[smoky oak]

@wispy nestGet prepared

I am mass testing realism models, and I have found one that will bring my gens to the next level

it is beating my current model severely

Model I have been using (custom mix)

New one

WAYYYY more realistic

I have a game plan for a new kick ass mix I think is gonna do reallyyy good

[bold cave]

pair it against my custom model

lets do a pepsi test

and if it beats mine I will bow down and call you master

thats how confident I am on senblend 1.6.3

coming soon to a theater near you since the upload keeps failing midway trolol

Or better yet. Upload it to civitai and I will do 50% my model 50% yours to see if that creates the ultimate model. Kind of like a goku+vegeta fusion

[smoky oak]

That could be dope haha

If you would be willing, I would be totally down to put your model through its paces in my test's. We could try and breed a super blend model haha

All for the benefit of realism for all

I am picking the candidates for my new mix at the moment

[bold cave]

What are your test prompts

cuz I have 5

[smoky oak]

I am picking between 2 different ones right now for my main body model to mix on top. Found the one I was using before wasn't the best option

[smoky oak]

I am testing tons, not just the same ones over and over

I am doing a big lone up and tallying the results of what I like the most

I am actually down to 2 finalists right now

[bold cave]

[smoky oak]

leaning to one over the other, but I am trying some edge cases to see if the numbers stay the same

[bold cave]

also depends if ur using it for nsfw or not, which obviously I do

but yeh general-purpose photorealism is my jam

[smoky oak]

I use it for both as well, but right now I am going for realism above all else

[sturdy rampart]

Do I have to download these files to be able to use the model?? I'm downloading dreamshaper if that helps

[smoky oak]

Download the first

its what you want

FP16 safetensors

[sturdy rampart]

I don't have to download everything?

Sorry, pretty much a noob here

[smoky oak]

All good

[bold cave]

wait whats the dif between fp16 and fp32

I had never seen the two options

[smoky oak]

its 2 different file types, 2 different precisions

[bold cave]

which means?

[smoky oak]

FP 16 is what we run on RTX cards

[bold cave]

oic

[smoky oak]

FP32 just uses more VRAM and takes more compute

[bold cave]

in exchange for?

[smoky oak]

for very very little benefit

[bold cave]

ic

[smoky oak]

some models may actually be worse at FP32

cause it may pick up a little more on issues with the data set like compression and stuff

[sturdy rampart]

What about this one? Say I already downloaded the full model fp16 in another model, do I have to download this one from another model?

[smoky oak]

You always wanna go for FP16-SafeTensors

for every model you download

or FP32 if its the only option. It will still run like FP16 on most hardware

pickletensor should not be used on personal PC's

[sturdy rampart]

Ohh, so every model, I have to download another one of those FP16-Safetensors for each model for it to work?

[smoky oak]

Yup!

[bold cave]

yeh

[smoky oak]

When generating, you switch between them

[sturdy rampart]

Wow, that's a lot of gbs

Okay, thanks for the help!

[smoky oak]

it can be

but I and other people tend to mix the parts we like from different models into one

[wispy nest]

wow, great stuff Sytan 🔥

[smoky oak]

I haven't even merged them yet :p

[wispy nest]

that second one is really good

face looks pretty natural

[coral dome]

sup.
weird video model sd1.5 + make-a-vid pseudo3d conv + attn.
https://huggingface.co/spaces/TempoFunk/makeavid-sd-jax
bye!

[wispy nest]

damn now thats nightmare fuel lol

[coral dome]

ye

[sturdy rampart]

Should I download the Pruned Model or the Full Model fp16 model?

[smoky oak]

I would recommend full, if you have the space for it

[bold cave]

whos ready to see my 4k wallpaper

remove ur panties lest they get crapped on once you see it

[smoky oak]

go for it

I have done 4k, 8k, and 16k in SD

@bold caveAre you actually interested in working together on a mega realism model?

[bold cave]

Ive reached the limit of realism in 1.5

im about to finish it

but it means im gonna kamikaze this alt once I do lamo

its just too good not to share

There really is a limit tho

[smoky oak]

I'll gladly put it through my tests, if you'd be willing to share it

[bold cave]

Its marginally better than my previous blend

Uploads are turtle speed

[smoky oak]

curious to see how it works compared to my works and prompting

[bold cave]

Kk start uploading urs too

[smoky oak]

oh god

[smoky oak]

Mine is still in the words

[bold cave]

oic

[smoky oak]

also, I have really fast upload 😅

500/500

[bold cave]

watch this

Here's how I tested the models

[smoky oak]

alright, so I have figured out which model will be my new base

time to start mixing

this model also seems to be able to do darker skinned individuals far better than some of the others I have messed with

[bold cave]

but that's a sample comparison of all top models

ah nvm, its too spicy for the censorbot

pity. that was a highly informative post

[smoky oak]

Can send in DM's if you would like

I'd love to see

[bold cave]

but essentially you dont want to test a shitload of subjects. u want to test a small yet varied range of prompts

My 5 picks are: 2 videogame characters (widowmaker and zarya), highly popular celeb (taylor swift), niche celeb (kamala harris), ethnic random person (beautiful zulu woman) and full body erotic (beautiful mongolian dominatrix)

That way u have all bases covered

Some excel in random subjects but have no knowledge of VG characters or niche celebs

[smoky oak]

But what about non human realism?

[bold cave]

eh

I would do a separate model for that

[smoky oak]

ah, nevermind then

I am interested in a realism model, not a portrait model

[bold cave]

yeh I really dont care about realistic puppy fur

however in theory, if a model is good with human realism it is probably good in nonhuman realism too

you could add a sixth test for nonhuman but I personally dont care much for it

[smoky oak]

that is far from true from what i have seen

realistic vision can do really good humans, and not good animals, unfortunately

[bold cave]

Could be cuz most individuals who make these models are like me. Care little about puppies and more about boobie textures 🤤

[smoky oak]

Yeah, far from my care

I am after realism, general, not selective

@bold caveDoes your model have noise offset, or is it limited in brightness values?

[bold cave]

I was surprised at the few nonhumans I did with senblend 1.5 if Im being honest wif chu

so dont think theyre so mutually exclusive

and tbh, idk, I could share the .doc with all the tests if u wanna

[smoky oak]

From the 10 or so models I have tested, they very much are

[smoky oak]

I would love to see

[bold cave]

Kk let me see if I can zip it

[smoky oak]

Ok, I have discovered an extremely unfortunate byproduct of this model I am using

it has the best photo realism I have ever seen

but it adds noise into images

and it gets REALLLYYY bad with night time photos

[bold cave]

example?

all dark photos have a crapload of noise

[smoky oak]

my previous model didn't seem to

[bold cave]

some of the "new" models are actually poopie in disguise

there was this one that scored so low I immediately discarded it

[smoky oak]

which model?

[bold cave]

I think it was the NED model? Forgot what it stands for

its just called NED in my notes

[smoky oak]

ah

[bold cave]

but yeh it was the worst by far in all my test prompts

[smoky oak]

worst in my tests was protogen 3.4x

[bold cave]

I akshelly didn't test protogen for some reason this time around

but my senblend 1.5 did have protogen in it

[smoky oak]

the images were just too contrasty and unrealistic

and the colors were way off

[bold cave]

At least I know im not missing off on much

[smoky oak]

it consistently scored worse by a landslide in all my examples

Protogen vs the model I am using (secret for now)

Protogen vs one I am using again

nothing reliably realistic about it

Damn

even my old model is showing severe noise

hmmm

must be a prompting thing

[bold cave]

damn, that's protogen?

[smoky oak]

Yeah, left is

3.4x

[bold cave]

I think it wasnt an accident that I excluded it. It probably just wasnt in the top rated anymore

yeah it looks messed up

maybe lowering the cfg? but u really shouldnt have to

[smoky oak]

on CFG7, worked on all the other models

[bold cave]

Also i think protogen3.4 is pretty old

[smoky oak]

I just tried my first merge and... It lowered the realism of the original model really bad

I need a model thats very realistic with noise offset

[static tusk]

theres a lora for that

[smoky oak]

noise offset LoRA's suck from my testing

compromise massive quality for some meh results

also, slows down generations

god damn this model looks so good

I don't think I am gonna be able to merge it in any way without ruining its results honestly

like man, that looks good

[hasty nova]

I know that Digital Diffusion has 0.02 noise offset

Dunno why I picked that number but that's the number I picked

Very faint

[smoky oak]

SCAM

Do not click it

[hasty nova]

they are doing it in other channels as well

[smoky oak]

Man, I wish i could just delete these links when they are sent

[hasty nova]

like #1045349359044280360

and #💬｜general-chat

[smoky oak]

I reacted with the warning sign

hopefully somebody responds fast

[hasty nova]

I was almost a moderator of this server actually, but my age ended up getting in the way :(

[smoky oak]

I have been in the application process for over a month now lol

[opaque badger]

Imagine/ a dog

[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]

@opaque badger

[hasty nova]

I was up to almost the interview stage

[opaque badger]

Thx g

[smoky oak]

Same here

Thats where I have been stuck. I was told there is a lot going on right now so the hiring process is paused

[hasty nova]

But I was also in the application process for about a month or two

[smoky oak]

So I am just waiting for a chance hopefully. Just adding stuff to the community :p

[hasty nova]

Did you ask or were you chosen?

[smoky oak]

The applications are frozen now from what I was told

[hasty nova]

Well I can't even be one for another 3 years I spose

[smoky oak]

are you 15?

[hasty nova]

Yep.

[smoky oak]

ahh, I see

yeah, cause its kinda child labor since its for a real company unfortunately

[hasty nova]

I mean, I moderate another server as an official volunteer position that I get service hours for

Though I know this wouldn't be for service hours, it's still just a volunteer position right?

[smoky oak]

I really just wanna be some form of mod so I can try and spread more positive and reliable information, while also responding quickly to scams and stuff, cause I am always on lol

[hasty nova]

Or I guess it could be seen as an intership.

[hasty nova]

I am almost always on, at least when I am at my computer, and I check in on things at school too.

[smoky oak]

I have no school or job currently, so I am chronically online haha

[hasty nova]

My only job is game dev and I don't make money from that yet because I realized a bunch of mistakes I made and I am currently taking a break from it due to those mistakes. So I also haven't uploaded on youtube in a while.

[smoky oak]

I merge and finetune

Maybe

[bold cave]

train =/= finetune

u cant make a model unless u got boatloads of gpus and boatloads of cash

u dont have to either. the ones in existence are great and work better when properly meshed

[smoky oak]

I do train LoRA's, but thats very much different

I don't know how you prompt for that style of clothing

oh-hoho

This model does grunge reallyyyyy well

no, not what i was meaning

[hasty nova]

man now I am wondering if my model can do that but I am currently trying to get an achievement for the Stanley parable to play it for the entirety of a tuesday so idk if I have the computational resources to test it out

[smoky oak]

kinda

[hasty nova]

maybe somone could try it for me since my computer is too captivated by the stanley parable?

it's a video game

[smoky oak]

sure, shoot it over and I'll give it a try

[hasty nova]

and it's not

[hasty nova]

https://civitai.com/models/48858

It's a 2.1 model

[smoky oak]

oh...

I think I may pass in that case, I'm sorry

I don't enjoy 2.x at all

[hasty nova]

What's wrong with it?

[smoky oak]

Just never found a 2.x model worth it

always deformities and artifacts

I mean, sure I'll try it

but I am not gonna hold any punches

[hasty nova]

I would say my model is pretty decent, not saying it's perfect but I did finetune it on a large hand captioned dataset and it works pretty well. It has the same issues as 1.5, but less often. And it prompts more similar to 1.5, but make sure to read the model description so you know what specific optional tokens you can use to really enhance your image.

[smoky oak]

Alright, ia m gonna try it

[hasty nova]

Ooo fun!

[smoky oak]

I am gonna hammer the photo realism aspect, as a word of warning

who knows, it may blow my mind

[hasty nova]

Alright. Make sure to include "photograph" at the start of the prompt!

And the words "moody, dramatic, dark, spooky, cinematic" may get the style you are looking for too.

Those were all in the captions

[smoky oak]

lol

I used all of those words

If there is one thing I know how to do, its prompt

[hasty nova]

Awesome

Just make sure to use a good resolution for it, base 512 won't work very well (but hires fix can help with that), but 768 resolutions look better I think.

[smoky oak]

first test is not looking good-

but I am comparing against the best photo realism model I have ever used

[hasty nova]

Ah

[smoky oak]

so its not like a super fair comparison

[hasty nova]

Yeah, my model isn't purely photorealism, its a general purpose

[hasty nova]

and its original intention was for digital art

screw you

it's not a video

Indeed I am

[finite gale]

has anyone managed to be able to make a comic book out of SD?

[hasty nova]

Why?

[hasty nova]

Several people have

[finite gale]

links?

[hasty nova]

Not any rn, sorry

[finite gale]

im curious to see what stage we're at there

i think that it's still not possible yet to do it

[smoky oak]

you'd be wrong

[finite gale]

me?

[smoky oak]

yeah

@hasty novaAlright, unfortunately your model is having the same issues I get with other 2.x models, which is why I don't use them ._.

[hasty nova]

I mean, what I said is better than cursing or slurs or something which I would never do

[hasty nova]

Try sending some pictures + prompts?

[smoky oak]

768 res, but it looks like 300x max

[finite gale]

you have an example for me sytan?

[smoky oak]

your model next to a 1.5 model at 768x768

[smoky oak]

not anything off the top of my head, but I have done enough image generations to know its easily possible

[hasty nova]

what was the prompt?

[smoky oak]

photograph of a handsome man, smiling, realism, detailed, bokeh, photo realistic, teeth, detailed skin, detailed hair, upper body

[finite gale]

to do a completely coherent comic where the characters look the same and have actual panels that are nontrivially complex?

[smoky oak]

gonna try another

[smoky oak]

yup

we can do consistent same character gens just fine, especially if you train a LoRA on them

also, i am not sure why, but your model likes to default to black and white for me

[finite gale]

if you train a LoRA, you don't miss out on more generalized complexity?

[smoky oak]

not if you know what you are doing

[hasty nova]

Try this instead, same seed.
Positive: "photograph, handsome man, smiling, portrait, teeth, bokeh"
Negative: "low quality, low res, messy, smooth skin, black and white, old, black and white"

I want to see what it does

[finite gale]

like if i do a lora on a "particular character", it's not going to now struggle a bit if i make that character do something outside the lora?

like a cute girl trained on pretty settings with flowers, now if she has to wear a helmet and slash a sword...its going to struggle, right?

[smoky oak]

you can utilize control net and iteraticve LoRA's

[hasty nova]

Almost never does that for me.

[smoky oak]

another

the 768x messes up the 1.5 image lol

[finite gale]

control net works for poses, but not for facial expressions -- or is there something that can do expressions now?

[smoky oak]

pretty sure there is, not something off the top of my head right now

[finite gale]

i've seen very little so far for like people focusing on diverse set of expresions

[smoky oak]

yeah, i had a feeling, that prompt is reallyyy not good for either

it looks terrible for both lol

[finite gale]

like, a specific character but in a diverse set of expressions

[hasty nova]

really?

[smoky oak]

you tell me lol

[hasty nova]

that... is weird

I don't get why people get far different results than me

Not that I know of

It's not that popular

[finite gale]

a "character sheet" for example would be really nice to see -- and I'm not sure that SD is really there yet -- although it sounds like you're convinced sytan

[hasty nova]

that exists

[hasty nova]

https://civitai.com/models/3036/charturner-character-turnaround-helper-for-15-and-21

[smoky oak]

oh, now THAT we do have a lot of

[finite gale]

i meant more that there's a diverse set of expressions

[smoky oak]

oh of course you mean anime

blegh

[hasty nova]

If you can't find what you need, learn how to make it yourself.

[hasty nova]

Learn about finetuning, loras, embeddings, all that stuff.

[finite gale]

whatever dude pick a bro-tier comic if you'd like

[smoky oak]

bro-tier?

[finite gale]

okay maybe not, their expressions are all exactly the same

[hasty nova]

@smoky oak , here is an image that @static tusk made with the model, im not saying the model is fantastic and will work automatically for everyone, i am just wondering why its doing that for you.

[smoky oak]

oh yeah, that one looks much better

[smoky oak]

for me, the SD 2.1 model is just not working good at all

[hasty nova]

And I have no clue why. Sorry :(

[smoky oak]

its all good, I have the same issue with all 2.x models, and its why I never use them

[static tusk]

check the latest 2 pics I postd 😄

[hasty nova]

Yeah I saw those! They look great :)

[smoky oak]

I am gonna give both a chance to try high res fix

so final is 1024x1536

[hasty nova]

ah my model isn't great with hires fix and photoreal usually

Unless you are using negative embeddings

[smoky oak]

oh, its 2.x

ran out of VRAM

welp, nevermind then

[hasty nova]

How much vram do you have?

[smoky oak]

8GB

[hasty nova]

Same

I can usually do something that size

[smoky oak]

I am currently only after realism, so I will stick with the 1.5 model I found, as it seems to be the best there is

[hasty nova]

But I can get things like this. I'm disappointed it isn't working out for you, but may you have good luck with your photorealism on 1.5!

I have to go, see you all later

[smoky oak]

uh, gross

[finite gale]

that looks totally like ai to me man

[smoky oak]

same here, also gross

stalking a 17 year old waiting for the OF to drop on her birthday is giga yikes

@hasty novaso sorry to bother you, but what would you typically prompt to get clothes like on that woman?

AHA

I was so close

I was putting in renaissance

but it kept making paintings lmao

thanks!

yesssss

there we go!

[past pelican]

Never change SD... never change. 🙂

[fallow egret]

Previous attempts with inpainting

[hasty nova]

Fairly certain in that image I didn't prompt for clothing specifically but I think I included the tokens "peasant" and "medieval"

[smoky oak]

peeng

[hasty nova]

It was a really simple prompt

[fallow egret]

And btw fuck this guy

[hasty nova]

For a second I thought you were talking about the girl in the bikini and was very confused

[smoky oak]

no, this is not what this server is about

[fallow egret]

what is perverted about de-lewdifying an image?

[smoky oak]

good to know, thanks

[fallow egret]

You didn't say that to the guy stalking a 17 year old

[smoky oak]

This server is for SFW only

[smoky oak]

literally did

wait

my messaged got deleted

[hasty nova]

I think you two are confused a bit

[smoky oak]

what

[fallow egret]

Oh wait

You did I just missed that

[smoky oak]

I had a specific message saying it it was mega yikes to stalk a 17 year old waiting for the OF to drop

oh, its there now

it disappeared for me lmao

@fallow egretMy apologies for calling it perversion, its not perversion. Stalking a MINOR is

[hasty nova]

All of it was deleted

[smoky oak]

this server is just not for NSFW

[smoky oak]

its all there for me

weird

[hasty nova]

Oh I wasn't scrolling up far enough, my bad 😆

[smoky oak]

all good lol

we talk big

[fallow egret]

I don't understand the cause of your confusion

[smoky oak]

OH

OH

OH

DE-LEWDING

I'm sorry

I read it as RE-lewding

I thought you were trying to get rid of the censorship

sorry about that

big misunderstanding, really sorry about that

I am already on edge from the minor stalker above

living for this aesthetic

[fallow egret]

I'm doing the same thing as you, trying to get the AI to draw an aesthetic that I like; I was trying to generate something else and I was like hmm that doesn't look like what I typed very much but it looks like an interesting character

I could make a character.ai character based on the image, if it just wasn't showing so much chest; and I like the way the sides of her arms look so cutting off the lower part of the image entirely would be a shame

[fallow egret]

Looks like Sansa Stark

https://tenor.com/view/sansa-gif-20276581

[smoky oak]

@fallow egretits not the best, but I am trying

Oh!

this one is much better

again, not perfect, but much better IMO

[fallow egret]

Thanks so much; it does look kinda good, what's the trick?

[smoky oak]

this si how I did it

so, I put it into inpaint sketch, then drew over the problem areas, like this

then I sent to normal inpaint, and covred the bottom section

[fallow egret]

I did that part but I didn't know what to type; also I used a different site

[smoky oak]

words I used

[fallow egret]

Thanks, that's helping me understand how this works

[smoky oak]

no problem! glad you weren't asking for what I thought you were asking for 😅

sorry about that, one last time

[fallow egret]

This is actually what the AI I used thinks female Ted Bundy would look like

[smoky oak]

other guy gave me PTSD

[smoky oak]

LMAO

I have an idea

refined it a bit, but it looks more like a pencil sketch now. Do with that what you will :p

just tried to get the shirt looking better

[fallow egret]

The AI is mysterious; I was trying to make pictures of serial killers with boobs to troll a girl I know who loves serial killers too much; interestingly it can do that with Jeff Dahmer but when I try Obama, Putin or Ted Bundy it makes a whole different person, but when I just type 'female Ted Bundy' it seems to create a different character that doesn't have this same haunting beauty to me, and it usually gives the character a fucked up face with fucked up eyes

I guess what I'd really like is to figure out how to get this same character, but in some different settings

[smoky oak]

hopefully some part of this is more that I am not too sure about

this is a big venture from my normal photo realism style lol

[fallow egret]

What did you do to turn it into a pencil sketch?

[smoky oak]

was more of an accident lol

was trying to tweak it with an anime model, but the slight noise in the background kept getting amplified 😅

oops, forgot to switch models lmao

apparently these are what "African Women" look like in anime

[fallow egret]

To be fair anime people don't look like any kind of real people

[smoky oak]

true lol

[fallow egret]

Have you seen anyone in Japan who looks like Naruto?

[smoky oak]

no, but he is pretty cute :>

I would say I have a weakness for blonde hair, but I just have a weakness for men lol

HIS ADULT SELF I SHOULD SPECIFY

wait, he looks nothing like how I remember lmao

I redact all of what I said

I am not sure who I was thinking of

[fallow egret]

that show is too long, I couldn't watch it long enough for him to become an adult, he becomes an adult slower than a real person

maybe you were thinking of Sasuke or something

[smoky oak]

No, I was thinking of somebody with blonde hair

hmm