🏞｜general-with-images

[boreal falcon]

xD

i dont mean just nudes

not that i saw any nude from your images lol

except one image that indicated it, from your CivitAI site

profile

[bold cave]

lololol

[boreal falcon]

[bold cave]

U can see all of em if u enable nsfw in the settings

u should be able to see widowmaker's purple tiddies

[boreal falcon]

O_O

[bold cave]

Here it is but with Karras in lieu of Euler. Now if I hear one more time that it looks plastic, smudgy, fake or lacking detail I'm gonna staple my ballsac to the ceiling.

thats the power of dreambooth+controlnet working as a team

I didnt use it, but I do have cgi in the neggies

I also like how hard the model tries to get her nose right. lol

to catch em all

@shut sinew and @smoky oak gone real quiet after my magnus opus dropped

https://tenor.com/view/obama-mic-drop-largar-o-microfone-gif-7413222

[smoky oak]

What? lmao

[bold cave]

broooooooo

u gotta try this tool

[smoky oak]

What magnum opus?

[bold cave]

its magic broh

its magic

[smoky oak]

what, the image you sent above?

if so, you can keep that magic to yourself lmao

[smoky oak]

funny, those words are all terms I would use to describe that image

[boreal falcon]

i think you are bit too hyped brother (Sen) 😄 XD

or should i say it in Hulk Hogan voice brotha

[shut sinew]

I have a life man

[smoky oak]

Here is my recreation in like 5 minutes in UU@bold cave

biggest problem is the phone cable

also, its half the res, and higher detail... hmmm

[bold cave]

higher res in which universe

[smoky oak]

hmm?

my generation is lower res, but it has more detail for that res

[bold cave]

more detail in which universe I mean lmao

it looks buttery af

[smoky oak]

well lets see

[bold cave]

guess we'll have to disgree on what makes an image good

[smoky oak]

yours

mine

more individual hairs, more eyelashes, half the res

also way more realistic skin texture that doesn't look like a leather bag thats overstuffed

I mean, if this level of detail per pixel is something you are after, then go off

but I'll gladly stick to this level of detail

or actually my bad, thats my higher res version

this is the lower res one

still manages to have more detail at a lower res than this lol

lololol

@bold caveI mean, you tell me which one has more detail

[bold cave]

right

[smoky oak]

waldur wide

[bold cave]

if u look closely you can see how each part of the img looks different because of the tiling

for example the mouth part looks way different than the eyes

[smoky oak]

and yours doesn't have that issue?

[bold cave]

its luck of the draw with the seeds. also, Im using UU too dude lmfao I told u this tool works in conjunction with UU

the difference is that now they are consistent in pose even at high denoise

which means if u batch a couple of them you can patch one and another

I dont cherry pick gens

[smoky oak]

if thats the case, then you are not using UU to its full potential if you have less than 1/8th pixel level detail

[hardy comet]

what is this and how to fix it

[bold cave]

hold on a minute

[smoky oak]

like it doesn't take a genius to see a quality difference lol

[bold cave]

left doesnt even look like melora

[smoky oak]

I can only do so much

if you value how much it looks like a person over the actual detail, then by all means, I understand

[bold cave]

the thing I keep telling u and peter its that its easy to make anyone look good, but making a specific person look good is something else

like everytime I post a picture of my dreamboothed girls peter comes in and provides someone that looks nothing like her lololol

mine looks better bc thats what melora looks like

not some random girl

[smoky oak]

Perhaps thats just ause your dreambooths/LoRA's aren't high enough quality to upscale?

[bold cave]

also, I can go bananas with the denoise

[smoky oak]

cause I can upscale LoRA's just fine

[bold cave]

they are

another example is using hires fix with facefix

theres a reason noone who does specifics uses facefix. yes it looks more real but its no longer the same person l0l

im telling u that picture of melora is as good as it gets for 1.5

[smoky oak]

Facefix has been dead for months, I don't know a single person that uses it

[smoky oak]

I wish you future success in training better LoRA's

[bold cave]

I dont do loras

I do dreambooth

[smoky oak]

then dreambooths

[bold cave]

loras are less powerful

[smoky oak]

yet they get better results than you are sharing? weird...

[bold cave]

which get better results

I dont see your melora

[smoky oak]

oh, its MELORA

I thought you were saying "me LoRA"

[bold cave]

lmfaoooo

[smoky oak]

yeah sure, I'll make a LoRA for her

[bold cave]

yaaaaaas

lets do a pepsi challenge

on a different channel

and get an arbiter to mask which is which

[smoky oak]

sure

[bold cave]

lets do it in #1011228642472378369

[smoky oak]

I shouldn't need too long to get this done

less than an hour

[bold cave]

let's get a mod to be the arbiter

now the point is not to say that loras are bad

but im telling u that as far as sticking with a realistic subject, mine is way more realistic than a generic that looks nothing like her

[smoky oak]

also wait a damn minute

you didn't tell me your image was img2img@bold cave

[bold cave]

sit down kid

[smoky oak]

Actually looks more like her than their gen lmao

I am looking at her right now

sen, you didn't tell me your image gen was an img2img of her lmao

[bold cave]

I posted the original above lmfao

yeh it is. I wanted to get her in her business suit

dude stop

you cant recreate her accurately because she's a B-list

it looks nothing like her lmfao

u need a lora/db to get her right

ooooh hold on

sytan we need to take this to a different discord. one that has The Office fans

they know melora better than anyone. and they can judge it better

[smoky oak]

ok, that doesn't look like her

his kinda does

[bold cave]

ofc it doesnt this kid's a zoomer

[smoky oak]

all I can say is Sens upscale of her doesn;t look like her at all lmao

gave her crepe skin, and ugly lips

[bold cave]

im saying stop trying to generate someone who you don't even know who it is :kek:

thats what melora looks like dude. she's milf

[smoky oak]

@bold caveAlso, I am a zoomer as well, yet he and I are both basically the best in this server when it comes to realism gens, sooo...

[bold cave]

this is her

well a young one at least

[smoky oak]

yeah, so your eye color is off, your nose is off, and you gave her orange peel lips

[bold cave]

she's supposed to be crackly cuz shes milf lmao

the left even looks MORE like melora

but lets take it to an Office discord

I really wanna know who can make the best Jan

[smoky oak]

bro, the one on the right is a photograph of her

[bold cave]

ik lmfao

[smoky oak]

what do you lean left looks more like her?

[bold cave]

bc it's the more recognizable version of her from The Office

this dude lmfao

[smoky oak]

ok, so apparently these look the same...

Totally didn't make her 10x more wrinkly with caramel plastic looking skin, crusty eyes, and orange peel lips...

[bold cave]

lmffao

Its cfg 7

it gets that oversaturated look

[smoky oak]

and?

oh, thats the sign of a poorly made model

:/

[bold cave]

I can do cfg 4 if u want that more true to skin tone

the ones from senblend 1.6.3 usually go for the photostudio look

not the candid photo look, just personal preference

[smoky oak]

yeah, mine looks way more like her lmao

[bold cave]

no way dude

its bc u arent an office fan

if you knew her as well as office fans do, you know you'd be way off

it looks like she got a facelift

[smoky oak]

yeah, cause this leather face orange peel isn't way off...

[bold cave]

lmfaooo

this is too much fun

Oh Im also doing kate flannery next which is even wrinklier

we can do a challenge on that one. lora vs dreambooth

[smoky oak]

Not interested. there is no point in trying against somebody who can never be wrong. Good luck tho

although, she does look absolutely great in this dress

thats a fantastic pairing of colors between her hair and that gorgeous dress

yes lmao

I was just saying she looks great

Thats the next person Sen is gonna botch with a dreambooth

Please don't waste your time with them, they can never lose in their eyes

even when their images are so massively off

They aren't one to talk, more so insult

But I am down to chat, if you'd like

unfortunately, I do not

I have been hurting for funds for a while now

but I am learning python with hopes of getting a job

[smoky oak]

vibing

[hearty karma]

spider-woman :-)

[crystal bloom]

Made with revAnimated Checkpoint.

[boreal falcon]

imo the first one wins clearly

if i have to compare

[hasty nova]

im out of inspiration so im using a prompt generator script

this is pretty cool, might refine the prompt

Same seed, refined prompt.

Nope, never used that

It's a little standalone tool I helped make a while back that is really customizeable and flecible. Not as fancy as the actual AI-powered ones but it works just as well.

@smoky oak how is the realism? Is it any better than the stuff I was posting yesterday?

(for the last three)

[smoky oak]

Thank you lol

and I didn't even have a LoRA or know who that person was, AND I reconstructed it from the result of the second 😅

[smoky oak]

yeah, its getting better

[hasty nova]

cool

[smoky oak]

but the lighting and details still have that harsh look associated with 2.x, personally

but they look more like humans and less like 3D models for sure

[hasty nova]

Yay!

(this wasn't trying to be realistic) a dim kitchen

[smoky oak]

IDK why

but your name

always reads to me as "Junger lally"

even tho I KNOW its jungle rally lmao

[hasty nova]

Everyone reads it differently 😆

Most people just call me jungle

Though I have heard it as jungler-ally and then per perosn calls me jungler

Anyway i gtg

[smoky oak]

l8r loser

:p

[boreal falcon]

i dont know who used what, but the second one just has more clear...artificial looking stuff

plasticity and co.

hard to explain

i mean there is no perfect AI image but the first one is far more natural

[smoky oak]

yeah, it looks like plastic

the second one I mean

[boreal falcon]

yes

[smoky oak]

my specific jam is doing realism, so it makes sense I would have an edge in it

[boreal falcon]

you use your own model?

[smoky oak]

used to, and I am working on my own mix again

[boreal falcon]

ah okay

i dont generate photorealistic stuff as often

[smoky oak]

photo real is my jam haha

[boreal falcon]

do you have any specific use for them?

[smoky oak]

its where I tend to shine

[boreal falcon]

like i dunno, photobashing or so

or do you just generate them

[smoky oak]

practice for now. Potentially opening up for real life in person LoRA caricatures and stuff

its mainly just a pass time to see how real I can get my gens to look

well, i should say, I do not go for pure realism

I do like some level of artistic flare, even if its minor, of course

[boreal falcon]

interesting 😄

im generally interested in what people do with AI images

i have a bit "complex" route with AI art

[smoky oak]

I did train my own LoRA in order to bring my fursona to life in a way that looked how I wanted him to in the end. I had been using a heavily edited version I made of somebody else's art until then

[boreal falcon]

since i learn 2D and 3D art (including VFX and co.) and animating comes too

and editing ofc very important

and now generative AI is here

[smoky oak]

I have done 3D art for years, but I have fallen out of my interest with it

brb

[boreal falcon]

2 of my friends had the same fate but with 2D art

[smoky oak]

I am back

I had a falling out with 2D a long time ago

and more recently 3D

3D was mainly cause I hate using blender and dealing with its problems that have persisted for years, regardless of huge user outcry for them to be fixed

I realized they are not interested in making it a better experience, but rather an over-bloated one where they have more underbaked features

for now I am learning code, with the hopes of becoming an ML engineer

[bold cave]

Aight I promise I wont post this milf anymore especially since she's been my guinea pig all day but I gotta say ddim is my fav, this is using UU with CNET + seams fix

[bold cave]

bruh ur literally talking to the man in the mirror lamo

olivios, aitrepreneur, SDcourses, literally ALL OF THEM state lora is inferior to dreambooth in terms of subject replication

[smoky oak]

This one looks way better

[bold cave]

yeh but my question is

or my request is if anyone has suggestions for prompts that will make the subject look more like a candid photo

u saw the pix peter posted earlier

that had that grain and homey look

the ones I post all look like studio pix

I wanna make it look more like it was taken on an iphone

[smoky oak]

Ask him, I don't do images like that

I do DSLR style realism

well, I can do it like he did, but I do not enjoy it, and I do not do it as well

[bold cave]

I know using shorter prompts for whatever reason has that effect, but from examining those prompts I see no common denominator

it seems like luck of the draw tbh

[smoky oak]

this image was one I purposely used grain on along with a specific upscaler choice to get a more crunchy look

this is the full scope of the image

[bold cave]

thats more towards what I want although it's not a good example bc it has that same orange light

[smoky oak]

however, it does seem like the VAE I am using runs some form of denoise in the dark dark spots

[bold cave]

the ones that look homemade have that white light look to them

oh shit see I havent even messed with vaes

maybe thats where the secret is

[smoky oak]

good prompting is the best resource

Do not fall for it, it is a scam

[bold cave]

adding "natural light" and removing the photostudio prompts seems to help achieve a more candid look. this is dreambooth and no hires fix nor upscaling yet

also pushing the initial resolution as much as possible helps a lot in creating native details, although probably requires digging thru more seeds since this tends to distort subjects more

a little too blurry for my taste

bro use a custom model already l0l

absolute perfection

Haters: "it LoOks SmoOtH anD plAstiC"

The image:

if over-detailing an image is possible then it's obviously possible to detail an image just right. All with 1.5 and my sb1.6.3 model with dreambooth

just adding natural light to the prompt changed the style, although frankly, I like my previous cyberpunk photostudio style too. Yall are just picky af

nah. I like marilyn monroe tho

o shit she's an example of someone I wanna fine tune come to think of it

this dude wants to pay me to do a dreambooth tut, I might do monroe as the subject

delete that atrocity now

an insult to the OG sex symbol

the asian girl I posted is pretty darned good. But I plan on doing 4k wallpapers of every one of my 30+ dreambooth subjects

[modern kernel]

clown

[bold cave]

now that this tool is up and works so damn well I might get it done in a heartbeat

before, I had to frankenstein gens in photoshop and its super timeconsuming

I think he's saying that ur monroe looks like a clown l0l

marilyn monroe + chucky hybrid

that is a much better monroe although she's got a fivehead

I'll record a video tut since I'm getting paid for it. It'll be lengthy tho

Im revealing the dark secrets of my art. I havent seen youtubers use my method and as you can see it works quite well

sky's the limit with these models

there ya go

but yeh I plan on making the perfect monroe

no

with dreambooth

[smoky oak]

this is probably the least artificial image you have posted yet

[bold cave]

bro u cant be serious. its deliberately overdetailed as satire

if the image isnt so bumpy it looks like a topology map its too plastic for u guys

[smoky oak]

no, thats not the case lol

there is more to realistic skin than just texture

[cursive yacht]

"overdetailed"?

[smoky oak]

Some goofy term they came up with to try and come out on top

I wouldn't worry about it much lol

They like to make a mockery of things they don't understand

[boreal falcon]

Same, although not sure about ML engineery but maybe in robotics too or some other area

[boreal falcon]

users rather blindly hype it often loo

lol

But luckily Blender isnt by far the one and only in the 3D art world

[wispy ether]

[red forge]

/food planner

Imagine/prompt: food planner

[crystal delta]

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

Currently, there is no public bot on the server that generates images. There is an experimental bot available for early server members & Stable Society. You can obtain the Stable Society role by winning our weekly events #⭐｜pow-info #1087493421209485393! 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!

[mystic marlin]

I'm trying over here on the ControlNet Tile+Ultimate SD Upscale method. Details come out good but still working on perfecting this very image here.

[shut sinew]

Curious, what are your settings?

[mystic marlin]

For ControlNet or U SD Upscale, or literally everything? Haha

[shut sinew]

Everything
Im also trying this atm and got some complications

might be because im trying it on realistic stuff

[mystic marlin]

I'll drop the metadata from the image in a second that basically gives everything.

[shut sinew]

thanks man

[mystic marlin]

woman, deep red hair, ponytail, vivid blue eyes, wearing blue jacket, modern, portrait, sharp focus, high contrast, RAW Photo, intricate details
Negative prompt: ((low quality)), ((disfigured)), ((bad art)), ((deformed)),((extra limbs)),((close up)), blurry, out of frame, extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn feet)), ((poorly drawn face)), ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), extra limbs, (malformed limbs), (((long neck))), large breasts, watermark, nsfw,
Steps: 15, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 3501969932, Size: 2048x1536, Model: lyriel_v15, Denoising strength: 0.5, Clip skip: 2, ENSD: 31337, Mask blur: 4, Ultimate SD upscale upscaler: 4x_foolhardy_Remacri, Ultimate SD upscale tile_width: 768, Ultimate SD upscale tile_height: 768, Ultimate SD upscale mask_blur: 16, Ultimate SD upscale padding: 32, ControlNet 0 Enabled: True, ControlNet 0 Preprocessor: tile_resample, ControlNet 0 Model: control_v11f1e_sd15_tile [a371b31b], ControlNet 0 Weight: 1, ControlNet 0 Starting Step: 0, ControlNet 0 Ending Step: 1, ControlNet 0 Resize Mode: Crop and Resize, ControlNet 0 Pixel Perfect: True, ControlNet 0 Control Mode: Balanced, ControlNet 0 Preprocessor Parameters: "(512, 1, 64)"

It's a jumbled mess so read carefully. Haha

[shut sinew]

wait actually i could have just taken that myself

[mystic marlin]

Actually yeah, you could considering the image I uploaded was straight from the Webui.

Just drop it right into PNG info and bam.

[shut sinew]

Yeah idk, its still early? My best excuse

[mystic marlin]

Understandable, it's 4 AM here and I'm still freaking up. lol

[shut sinew]

11 AM here

my upscaled stuff (right) comes out as blurry mess with weird skin details

[mystic marlin]

RIP her complexion.

It still looks good though.

[shut sinew]

i mean there isnt a lot of detail in the shirt but it makes it worse

[mystic marlin]

Yeah it can be finnicky to play with this, I've been noticing that faces get super burnt if I go too high on the upscale multiplier.

Anything over 3 on the Lyriel model burns the faces.

Just makes women look....old....

Or even like they've been doing a little bit of that Walter White blue stuff.

[shut sinew]

maybe it depends on the model you have running while upscaling

like a lot

[mystic marlin]

Someone said that on a video that the model may make you have to play with some stuff a little.

Like the same settings won't work on every single model.

I'm starting to believe that....

[shut sinew]

i just dont know where the skin stuff comes from

[shut sinew]

Yep pretty much tried every setting now, cant make it work

[shut sinew]

only running controlnet for upscaling seems better but still not perfect

[dry crow]

Thats something im gonna try too the next days

[shut sinew]

I get the best results atm without controlnet, at least for realistic stuff, but thats also all i tried it with so far

Are we still going?

Well then yes

[shut sinew]

Dude looks creepy af

[neon crystal]

Wish granted. Here be the eyebleach.

[shut sinew]

You really got a problem if youre still onto that, why cant you deal with me having my personal opinion

My opinion wont change by you sending pretty much the same stuff over and over lol

[neon crystal]

Challenge accepted.

[bold cave]

yo we really need to have a #challenge-a-diffuser channel

so we can dump all the flame wars and debates over which model is best in that cesspool

This needs to be my wife

and I don't even find gingers that hot. To me they're like watered down blondes but this babe is something else

[bold cave]

use ddim, use the seams fix option with 0.2-0.35 denoise and make sure u got a dope model

[bold cave]

Here's a really cute one of gabi belle, another one of my dreamboothed models. If yall want tips on how to use this tool I can put up a video TuT on YT since I was literally banging my head against a wall all of yesterday using this tool so I have gained some valuable knowledge on its usage lol

@smoky oak Thank you also for teaching me to add bokeh to my prompts, that's gonna be a new staple for my gens. It's funny how things work out so much better for everyone when we work together instead of nitpicking each other's work. Also, adding "heavy makeup" to the neggies creates wonders like the one above

I'm adding some of these upscales to my model showcase in my profile if yall want the prompt data although a video tut would explain it better

I especially love the one above because gabi always wears heavy makeup in her videos. someone should send this to her and ask her if this is what she looks like al naturale lol

[bold cave]

As if I needed another reason to feed my diffusing addiction. Now cranking these out is so freakin easy omgg..

[bold cave]

https://civitai.com/models/22402/fantasticmix Can someone try out this model and find out if it's only good at making waifu-looking women? For the life of my I can't understand why people like that big-eyed alien look on girls

Aaand it looks like I'm going to yet again have to upgrade my model since someone just dropped this bombshell https://civitai.com/models/57319?modelVersionId=61735

[smoky oak]

This is definitely the best result you've sent

[bold cave]

I can't keep up with the AI world at this point

[bold cave]

It's outdated now.

People are putting up new models on civit like crazy

I have to make SB 1.7 now lmao...

[smoky oak]

I know, a new version of the best realistic model I've ever used was released last night, I haven't tested it yet

I will have to compare those two models that you just sent to see if they beat the model that I'm using

I love how you were insisting that you had the best model, and how realism could not go any further with 1.5, and now here you are saying that you're behind the curve and you need to make a new one because the realism got better with 1.5... lol

[wispy nest]

good stuff as always, pete. lots of those red heads look pretty damn convincing

[bold cave]

bruh. this is almost a 17GB model

[bold cave]

I guess you can never underestimate people's craftiness. But seeing is believing, the improvement may be marginal, or it can be extraordinary. Thank God I'm unemployed cuz it's time to do a lot more tests

Thank God this shit is fun otherwise I wouldn't be doing it 🤪

[smoky oak]

Good point, I will be testing those two new models that you sent along with the new version of the model that I'm using right now to see if there's any meaningful improvement

@bold cave if you send me your prompts and DM, I can see if there are any other changes I'd make, because there's still something the slightest bit off with the texture of skin in your images. However if you're happy with how it is currently, then I can totally understand that

You know, I don't mean to be that guy, but where are the moderators?

I've had two tickets open for well over half a day, one of them's minor, but the other one is literally somebody sending discord links to illegal servers in DMs because they got hacked, and neither have been accepted

not here in chat

thats what the mod ticket systems are for

actually, looking DMs, the guy who was sending links seems to have gotten his account back

[smoky oak]

what in the hell-

Thats the size of an SDXL model

👁️

OH @bold cave

Its a 1024x model

no wonder

damn, 16.6 GB is gonna take some time even on my fast ass SSD's and internet

[acoustic moth]

damn, im assuming realistic vision is outdated then at this point

[smoky oak]

yeah it is

I have already run tests and it has not been fairing well

its typically one of the worst

These images are mostly from my new go to, which just released a new version, so I will be testing that vs these two new models that sen sent

[acoustic moth]

these look great, ill stay posted to see what your findings are

[smoky oak]

I will be testing the model I use vs these 2 new ones extensively

[acoustic moth]

are you using controlnet at any point in your workflow?

[smoky oak]

nope

no inpainting either

well ok, the fair skinned princess has a little inpainting

[acoustic moth]

just prompt testing then nice, definitely interested to know and start training models over

[smoky oak]

from my tests, yeah, it gets beat in almost every way by modern realism models

[acoustic moth]

hey i saw your message about wanting to use controlnet to create poses, trying something similar did you end up figuring anything out

[smoky oak]

not saying its bad, just that its not the best anymore

here is another image from the model I use

I am just worried how damn long comparisons will take with that massive mega mix model

not everything is a competition man, and i am not saying realvision isn't good, its just not as good as newer options

[acoustic moth]

ah ok, im just experimenting with controlnet still, not getting the results im seeing in guide videos

[bold cave]

It says it's 1024 but it is based on 1.5

How is that even possible? I mean, I tried it once but it was ass

[smoky oak]

gotta know what you are doing, I have seen some people do it

alright, all new models are downloaded

doing a 4 way test

model I use vs the new beta version of it, vs fantastic mix, vs that ultra mix

[bold cave]

Ditto. I'll check out this 1024 model to see what's what

Yo can u pass the links to the newest ones

But if it's not compatible with any 512 model, then it's ass. I'm sorry

And my original statement prevails. 512 might actually be near its ceiling

https://civitai.com/models/57319?modelVersionId=61735

Aight less talk more testing. Time to debunk some theories

No lmao

Its a new one

Training a base model takes a shitload of money

its 1024

[smoky oak]

I am I about to test to see if thats the case

[shut sinew]

The problem with realistic models and the discussion here is that people seem to not see the difference, like they never actually saw real humans

[smoky oak]

I am doing a 5 way test actually

[fringe pecan]

#🏞｜general-with-images https://cdn.discordapp.com/attachments/1076852403082629161/1104028254630846484/imagineAi.jpeg A young girl with purple hair and red eye style，cyberpunk high details 4k, It's a masterpiece, the ultimate detail

[smoky oak]

Alright, first 5 way test is commencing

this is gonna be so slow with how huge these models are bro

[shut sinew]

yup thats the issue

most models are fed with heavily edited pics

so nothing looks real

Like yes your pics look like taken with a phone in 2015, but thats just to hide the imperfections of AI

[smoky oak]

jesus christ man

fantastic mix took 13 seconds to load, ultra mix took 29

ok, so

ultra mix looks horrible-

and I am not sure whyyy... hmmm

also, fantastic mix is not realistic at all @bold cave

looks like rendered anime

[bold cave]

yea some of them are a huge waste of time

[smoky oak]

fantastic mix is right

[bold cave]

For some reason, I dont remember why, I immediately discarded RV 2.0

It had a weird CG effect or some thing I didn't like

[smoky oak]

the ultra merge loosk horrible

[bold cave]

that saves me some time

[smoky oak]

what is this

its supposed to be a picture of a guy lmao

[acoustic moth]

whats left

[smoky oak]

realivision 1.4

[acoustic moth]

looks great, using controlnet for the poses or something else?

[smoky oak]

alright, first test is not looking good

[acoustic moth]

i have 2.0, i feel like the older version is actually better

[smoky oak]

ultra mix is unusable at these settings, which disqualifies it

fantastic mix looks terrible, and the new beta version of the model I am using does indeed look more fake like people where saying

[smoky oak]

1.4 is a lot better than 2.0

2.0 is refined 1.3, whereas 1.4 is an offshoot

I have done dozens of tests between them, 2.0 is always almost identical to 1.3, while 1.4 has a little more pop and tonality

[bold cave]

YES

Im glad Im not alone in that

SB1.5 had RV1.4 which was the bomb but when I tried 2.0 I was like...wtf?

[smoky oak]

1.4 is for sure the best real vision

[acoustic moth]

ill need to get 1.4 then

[static tusk]

did you save the config file for ultra?

[wispy nest]

No, a script to make a video

[shut sinew]

edge of realism and analog madness are better imo

but probably prompt depending

[smoky oak]

analog madness is one of the top performing models in my test

it took 2nd place in the last test

[shut sinew]

yup makes sense

[smoky oak]

let me make sure

[acoustic moth]

sounds interesting, hows that work

[shut sinew]

Do you have edge of realism in your test too?

[smoky oak]

ah, I forgot to rename it

[smoky oak]

I guess I can add it in

[acoustic moth]

rv1.4 taking first place?

[shut sinew]

[smoky oak]

no

1.4 came out second to last

[acoustic moth]

lol

[smoky oak]

Alright, I can't keep holding on to my secret model

[shut sinew]

RV isnt bad, there is just better stuff out there now

[smoky oak]

the winner in my tests has been Cyber realistic 2.0

Its been consistently the best in all of my tests

My magic model is cyber realistic

[bold cave]

Aaaaand we're off to a bad start with the 17GB model

[shut sinew]

doesnt even look human lol

[bold cave]

I don't care how detailed your model is, if you can't get the face right of the most successful female artist in history, it's shit

let alone that

it cant do taylor swift

[smoky oak]

Alright, i am gonna be testing Cyber Real, Cyber real beta, 1.4, edge of realism, and ultra mix

(Fantastic is disqualified cause it looks like barbie dolls)

I am also gonna get rid fo cyber real beta actually

[shut sinew]

from the preview cyber real has a painted/plastic look too, gotta try myself

[smoky oak]

its just worse cyber real 2.0

[bold cave]

It could also be that these models are so different that our go-to prompts dont work here

[smoky oak]

@shut sinewThese are almost all cyber real

[bold cave]

having to re-engineer prompts seems like a pain in the ass

[shut sinew]

looks way more solid than the previews thats for sure

[smoky oak]

It has consistently won my tests for realism and wide reaching realism

so realism for humans, and wide reaching for animals

also, those are jpg compressed versions of my images, those are all about 4k native

[bold cave]

Here is the same model without using 1024

[smoky oak]

This is what the detail on her face really looks like

[bold cave]

bro @smoky oak I'm telling you this model isnt 1024

its still 512 based

and that taylor above is miles ahead of a 1024x1024

Maybe he trained 1024 but somehow adapted it on a 512 space? idk 🤷

[smoky oak]

alright, I have everything I need

We are doing cyber realistic, real vis 1.4, edge of realism, and the mega model, any suggestions for other models?

[bold cave]

Let me finish my tests with the 17gb one. If its not 1024 then it will be fully blendable with SB

I honestly think adding too many models might ruin it

DUDE

look at this upscale

No fucking way!!!!!!

This is endgame

[smoky oak]

looks blurry IMO

[bold cave]

She looks less like swift, but SB can get celebs down pat

Still for a base model looks insane

So if I blend both of them I might have the best of both worlds lolol

[smoky oak]

what do you mean a base model?

[bold cave]

Like just 1 model without experimenting on merging it

merges always come out better

[acoustic moth]

.

[bold cave]

Im gonna do a three way test. 17GB/SB1.6.3/And a blend between both and see which produces the best results

[shut sinew]

you can just look up what i posted through discord

[smoky oak]

had a look, not bad

[shut sinew]

I still dont know what kinda problem you got with me

Always the one asking

[smoky oak]

peter just comes off as having a problem I think

[shut sinew]

Maybe english isnt his 1st language

[smoky oak]

I think thats just how he is

it isn't, I think french is?

Peter, whats your native language?

[shut sinew]

Then it probably makes more sense

[smoky oak]

ahhh, my bad

[shut sinew]

You always sound like you want to fight me

[smoky oak]

but yeah, I had the same feeling at the start boto

[shut sinew]

[smoky oak]

but the more I have hung around him, I'm just chill

[bold cave]

brazilians are like that trust me

[smoky oak]

https://tenor.com/view/rolblox-devinbooker-gif-18412462

YOU'RE GOING TO BRAZIL!!!

[shut sinew]

not biased at all

[bold cave]

theyve earned their reputation huehue

[bold cave]

Edgeofreal looks best

[acoustic moth]

i trained a model with this face, any way to use controlnet or anything else to get more full body poses? right now its only faces

[smoky oak]

I wills ay, realvis 1.4 does load by far the fastest

[bold cave]

I aim to devastate

Bro stop wasting ur time

17GB is it

I dont even have to keep testing lmfao

[shut sinew]

ill make a comparison too with my prompt

[smoky oak]

what are you suddenly doing different?

[bold cave]

THE MODEL

[smoky oak]

oh crap

[bold cave]

the 17GB model is all you need

[smoky oak]

Ultra mix just handed the others their ass

[bold cave]

its not 1024 is what Ive been trying to say

[bold cave]

so its compatible with all my 512 1.5s

[smoky oak]

ok, ultramix wins hard here

[bold cave]

Interesting. edge of real is the worst one here

[rugged pendant]

accidentaly made this

[smoky oak]

I specifically asked for a dreamy and blurry background, and only ultra dud that

[smoky oak]

Thats why I test edge cases, I want the best allrounder

[smoky oak]

heh, thats cute and fun lol

[rugged pendant]

i got that from normal lol

[shut sinew]

Im sure he didnt mean it that way

[acoustic moth]

besides bottom left, cyber looks best

he didnt mean it like that, youre good

[smoky oak]

its not what I asked for, so I need to dock points unfortunately

I specifically asked for strong background blur, and an upper body portrait, not a full body portrait

[acoustic moth]

you ever test full body portraits?

[bold cave]

My model does them pretty well

[bold cave]

BROOO

look at what happened when I added ambient occlusion to the prompt

thats not even upscaled dogg

[smoky oak]

I have to give the slight edge to ultra mix on that one as well

real vis is way too overprocessed

[acoustic moth]

got any examples?

[bold cave]

The censorship bot will probably block them from posting

but let me finish my testing and I'll find some

[smoky oak]

this one is a toss up between ultra and cyber IMO

[acoustic moth]

those longshot faces

[smoky oak]

they would be fixed with upscaling

[acoustic moth]

are you able to do full body poses with controlnet without the face being disfigured?

[smoky oak]

actually, I am gonna run an upscale on the top right image in those

it has been from my last tests, I am testing new options

[acoustic moth]

i actually havnt tried that yet, only inpainting

[shut sinew]

yeah have done that

[acoustic moth]

any insight on model used/settings or dont want to give out your sauce?

[smoky oak]

they very much are not

[shut sinew]

this is pretty old, but highres fix helps, model shouldnt matter too much

[smoky oak]

real vis looks the worst here, too overprocessed

cyber has a blown out sky

Edge of real and ultra mix look solid, but I like the more cinematic colors on ultra

[shut sinew]

with a more stylised prompt

[acoustic moth]

im using rv1.4, weight on 1, steps 0-1, balanced controlnet and prompt, my faces come out completely distorted, also using openpose

[smoky oak]

i am removing realistic vision from the test again, as its still just giving worse results IMO

[shut sinew]

at what final resolution?

[smoky oak]

the differences ARE the comparison lol

you give identical everything and see how much they deviate lol

[smoky oak]

now you are finally starting to get into that realism area

[acoustic moth]

512, havnt touched that, would it need to be 1024 minimum?

[smoky oak]

still not QUITE there, but that is the most realistic so far for sure

[bold cave]

I am super tempted to cut the testing short and just go for the new model

[smoky oak]

I am liking it so far

[bold cave]

but I have to go through with it. I am super pumped to see if a mix will be better or worse

[shut sinew]

you need higher resolution for more details, especially if using a pose where the face is "further" away

[smoky oak]

oh! I am gonna test to see how well it plays with one of my LoRA's

[acoustic moth]

@smoky oak thoughts on realism here?

[smoky oak]

but I probably cannot send the results here lol

[shut sinew]

also as i said, high res fix

[bold cave]

Im gonna have to retrain all my 20+ dreambooth models oh dear...

and all of the new models are gonna be 17gb a pop lmaoo

[smoky oak]

its decent, I wouldn't say super realistic, but it alsod oesn't look like straight up plastic/rubber

[acoustic moth]

what resolution would you suggest

[smoky oak]

Hmm... If only there was a small little thing you could train that can manipulate models

looks to my side

OH RIGHT... LoRA!

[acoustic moth]

over 1024 overkill? or 4k is good

[bold cave]

never

[shut sinew]

im doing 732 by 960 with 2x highres fix atm

[bold cave]

I always go for optimal, not convenient

Dreambooth on this model is gonna be insaiyan, IF it works

[acoustic moth]

ill have to look into highres fix not at my computer rn not sure what that is, but appreciate the tips

[acoustic moth]

trained on rv2.0, a better base model would help i hope

[shut sinew]

Its an automatic1111 thing, not sure if you get that anywhere else

[acoustic moth]

i have auto1111, is it built in?

[smoky oak]

RV 1.4 is better than 2,0 in my tests, but 1.,4 is still worse than a lot of other realisrtic models out now

[smoky oak]

yeah

[shut sinew]

yep

[smoky oak]

I am using it for these high res comparisons

[acoustic moth]

gonna be waiting for your test results haha, good stuff so far

[smoky oak]

I just disqualified 1.4

cause it just looks too overprocessed

We are doen to cyberealistic, edge of realism, and photo real ultra

[bold cave]

this is the exact same last prompt but with SB1.6.3.... honestly.... it's not that far behind. It's still solid as fuck to be honest

[acoustic moth]

ah ok i see, how would you achieve a 2x highres like boto mentioned earlier

[shut sinew]

another stylised prompt

[bold cave]

After SB1.7 comes out, we can finally say we have a model that's better than MJ v5 😈

[acoustic moth]

these all look good

[smoky oak]

we already have plenty of models better than MJ lol

[bold cave]

its not even out yet fool loooool

[midnight owl]

how are yall getting such realistic images? I'm having trouble getting anything good out of practically every model I've tried besides everything v4

[smoky oak]

model and prompting

[midnight owl]

what model do you use?

[smoky oak]

I am actively testing them right now lmao

midjoruney has a post processing filter that brings down the realism

[shut sinew]

Edge of realism and analog madness are my favs atm

on the grids we post you can see a few more names

[bold cave]

https://civitai.com/models/57319?modelVersionId=61735

[smoky oak]

these are considered peak realism out of MJ V5

[bold cave]

try this one for now brutha

There may be vram limitations tho

[bold cave]

no way

are those really from v5

[smoky oak]

yes lol

[bold cave]

damn. maybe Ive been hallucinating too hard

[smoky oak]

LIKE I HAVE BEEN SAYING LMAO

[bold cave]

well MJ just screwed themselves over by removing free trials

[smoky oak]

v5

[bold cave]

its only a matter of time before theyre left in the dust

[smoky oak]

WTF am I even looking at

MJV5

[bold cave]

ok its still pretty damn good

the one peter posted looks like an actual movie still

[smoky oak]

MJV5

[bold cave]

so credit to mj. not saying its better

[acoustic moth]

its not that bad, at first glance regular people will just think its an over edited photo not necessarily ai

[bold cave]

but its still pretty powerful

[shut sinew]

with a custom mix in

[smoky oak]

Oh yeah, not bad at all, but I always see people saying that MJV5 is like WAY ahead in realism, and its just not lmao

it is :/

[bold cave]

something's fishy but who cares. we are all team SD here

[smoky oak]

I am just going off of the top results that come up with MJ V5 photo realism

MJV5

[acoustic moth]

these two as well as some of sytans tests look really good, are you doing a specific photography prompt?

[smoky oak]

while the detail is great on those, the faces are still decently deformed on the black women

[shut sinew]

all my pics i post usually have the meta data in, so you can just browse through my images

[smoky oak]

and then people are calling these "extreme realism"

which translates to blur everything so much that you can't tell

[acoustic moth]

good to know, ty

[shut sinew]

2005 is calling they want their camera back

[smoky oak]

I have no interest in generating low res compressed and messy images

[bold cave]

thats essentially peter's fetish lmfaoooooooo

[acoustic moth]

can you make consistent characters in mj to the same degree as sd?

[smoky oak]

These images are just not as good as people think they are

[bold cave]

MJ has no dreambooth. need I say more

[smoky oak]

THOUGH

I MUST SAY

MJ has this very specific image style of somebody through a window, and these look incredible

Now those, I can see why people say MJ looks amazing

but in just normal portraits, it drops the ball

[shut sinew]

those look pretty solid

[bold cave]

this is my biggest beef with them

you cant test v5 unless you cough up some cash

[smoky oak]

these just don't look realistic, even with all of the media posts of "OMG!!!!!!11!!1!!! MJV5 looks PHOTO REALISTCI!?!?!"

[bold cave]

they can go fuck themselves, AI is free by nature

[bold cave]

they kinda do... u just have crazy high standards

not saying they're better again

just saying theyre still pretty good

there can be. a lot of things are open source and crowdfunded

[smoky oak]

There is nothing realistic about the depth fall off from the lens, or the flawless skin, or inconsistent lighting

but anyways, I digress

MJ's recipe is just blur it so much that you can't tell

[bold cave]

that's what crowdfunding is you knucklehead

[smoky oak]

except MJ is built off a base level of Stable diffusion, which is free and open source, and community driven by thousands of people getting paid 0

Stable diffusion is nothing without its community lol

I used to be a customer of MJ back in V3, and they themselves said they used stable diffusionf or a base gen, then blurred it, then applied their own latent upscaling to it, and then put a post process color filter on it as well as a VAE that could attempt to fix the issues introduced by their latent upscaler

they are from months ago, and I am not in their server anymore

[shut sinew]

Dont they also edit the prompts you put in to add more detail?

[smoky oak]

yes

they also have a blacklist longer than a marathon

[bold cave]

exactly

[smoky oak]

I have heard they fixed that

[bold cave]

not to mention their wussy-ass censorship

[smoky oak]

but I left MJ cause they were making it nearly impossible to generate anything that wasn't G

not PG

but G

[bold cave]

really. did their censorship get even worse

[smoky oak]

When I left, it was so bad that words like "Blood orange" like the fruit, would get your generation privilege's taken away from "trying to skirt our necessary censorship"

[bold cave]

woooooooooow

I even heard that people flat out got their subscription revoked

what a spineless pitiful company

[smoky oak]

Want a fighter whos bruised and wearing bandages? too violent

[bold cave]

thats chatGPT in a nutshell too lmfaooo

[smoky oak]

want a bodybuilder who is working out shirtless?

Shirtless men are too NSFW

like, it would generate men with shirts on IN POOLS

[bold cave]

lmaooooooooooooooooooooooooooooooooooooooooooo

[smoky oak]

MIDJOURNEY BRO

I used to pay for midjourney for like 4 months

[bold cave]

back before 1.5 dropped, it made sense.

[smoky oak]

what? lmfao

[bold cave]

they were revolutionary in the AI field

[smoky oak]

no cencorship?

[bold cave]

now they are just the laughingstock

bruh stop feeding peter when he's in full troll mode

peter just likes to stick to his ignorance

[smoky oak]

yeah, I agree, I'm just gonna stop on that end

[bold cave]

EVERYONE knows you cant do anything even remotely sexual in MJ

he could google it

[smoky oak]

protecting things for your platform is legit cencorship...

[bold cave]

but he would rather be a knucklehead and be confidently incorrect

know the difference between opinion and fact

fact = MJ is censored

opinion = MJ sucks

[smoky oak]

yes, it is a fact

[bold cave]

I am not responding to him anymore

[smoky oak]

The fact is, MJ is censored heavily

[bold cave]

until he backs off from troll mode

[smoky oak]

thats funny, MJ removed the ability to generate politicians for a hot moment, which is literally cencorship to protect political views and political morals... Which is cencorship

They just chose to remove and add the ability to what you generate, which is by definition, a form of cencorship

Anyways, this is just a waste of time now

just move on bro, this is a waste of time

extremely unfortunately, thats just the way open source things work, and there is no changing that

literally nobody has said anything about you being a troll for stating that people will use this for bad

yeah, we weren't talking about illegal use of AI

[midnight owl]

yeah, there will always be people who misuse technology

[smoky oak]

we were talking abotu how midjourney censors their gens, community, and reddit, which is a fact

I agree, the Brazilian thing they said was problematic

not cool at all

[rain gazelle]

hope everyone is having a nice day

[smoky oak]

hope you are as well

[oak osprey]

4k image from SD has duplicates, but incredibly high quality. i love it still

[rain gazelle]

yo

[rain gazelle]

Alice in Wonderlandy

[oak osprey]

nailed it

thats supposed to be amber heard as alice lmao

[bold cave]

are you rendering natively?

[oak osprey]

yes

[bold cave]

this can be fixed in many ways

outpainting, inpainting, latent coupling

[oak osprey]

outpainting could work but this is from a discord bot i work on so i'm not sure how to get masks and stuff in there super easily

[bold cave]

ooooh I see

[oak osprey]

yeah i like reinventing my own wheels so i can learn more and do exactly what i want with the tool instead of try and figure out how to extend eg. stable-diffusion-webui

[rain gazelle]

Amber Heard.. hmm ok, interesting choice.

[oak osprey]

i do reuse libs where it makes sense

[rain gazelle]

Challenge accepted.

[oak osprey]

[bold cave]

ah fuck amber heard is a fox

[oak osprey]

she is pure evil but yeah very nice looking

try a hybrid face combo of Jennifer Connelly and Amber Heard. if that woman existed in real life she'd have all the roles

[bold cave]

I will put that on my to-do list

[oak osprey]

i see what you did there

[smoky oak]

your images are definitely starting to look more realistic

[bold cave]

ty~!

I actually forgot to turn on the CNET on that one lmfao

and that one believe it or not is the one test where SB1.6.3 beat the 17GB model

[oak osprey]

i know it's all highly opinionated but can you all provide feedback on my aliases for these schedulers. i'm sure Euler A could have a better preset name than "fast"

[bold cave]

SB1.6.3 is no joke. It is a pretty solid model

[oak osprey]

i feel so boomer. what is SB?

[bold cave]

my own model. its in my profile. I am testing it vs a brand new model that is outperforming it in almost all tests

my plan is to merge that model with the new one. I just need to know what percentages to use, thats why Im doing some tests

[oak osprey]

interesting. could add it to my discord bot and you can just sling large batches at its queue

[bold cave]

there ya go

[rain gazelle]

ok got distracted

amber heard, alice in wonderland... let's go

[oak osprey]

whatever you do, do not try Millie Bobbie Brown (that gal from Stranger Things)

unless you want the WEE-WOO's at your door step

[smoky oak]

oh god lol

[oak osprey]

how dare you use non-carbon neutral sources of entertainment to generate Greta

[smoky oak]

lmfaooo

we might be getting solar panels, in which my AI generations are 100% free lol

[rain gazelle]

welll.. that was a fail

[oak osprey]

my friend prompted "Andrew Tate in Prison, taking one for the team" which is how we found out that "taking one for the team" is a metaphor SD understands all too well

[smoky oak]

oof lol

[bold cave]

im willing to do an amber heard dreambooth

then I'll blow it out of the water

[smoky oak]

I have a LoRA for andrew tate I made lol

[oak osprey]

💦 mark

[rain gazelle]

the amber likeness is kinda tough

[oak osprey]

i was using urpm

[smoky oak]

AH

[rain gazelle]

I'm also not that familiar with what she looks like, honestly.

[oak osprey]

i'd buy that is her, fwiw

[smoky oak]

@bold caveRace to create A LORA/dreambooth of amber?

[rain gazelle]

nah i can do better

put me in coach

[bold cave]

yaaaaaaas

let me finish the tests

im getting sidetracked af lol

[oak osprey]

put me in coach

"your ticket is first class"

[rain gazelle]

oh man they just keep getting worse

Does that look like her at all?

[oak osprey]

looks nearly identical to her

yours I mean. mine is a hybrid of amber and jennifer

[rain gazelle]

nice well there ya go

[bold cave]

I keep trying to upscale this mongol babe but she keeps deviating too much from the original

[rain gazelle]

whichi method?

Which, rather

[bold cave]

CTU

[oak osprey]

weird. my scales never change

[bold cave]

controlnet tile upscaler. Im hoping the acronym will catch on

[rain gazelle]

Try a simple SD Upscale at .1

[oak osprey]

oh

[rain gazelle]

up to maybe .25

[oak osprey]

that's great if you're wanting animu

[ivory compass]

good day

[oak osprey]

I'm ashamed to say but I have a command that queries gpt35 turbo with a prompt that describes how stable diffusion prompts work, and I can generate a random number of these... they're usually better than anything I can come up with and it's with fewer words in the prompt lmao

the Alice in wonderland series was prompting that with the theme "amber heard as alice"

[smoky oak]

@bold caveAlright, starting my training, data set secured and tagged already

[bold cave]

so is this gonna be a competition or what

cuz I want it to be. We can test SB1.7 vs ur model while we're at it and let the people decide which amber is best

[rain gazelle]

the more cartoony I go, the less Amber I see...

[oak osprey]

currently battling an OAuth issue with threads so i'll be back in a bit to help out with that

[smoky oak]

@bold caveAlright, its done

time ti test

oh boy, you have your work cut out for you lmao

this worked really damn good lol

[oak osprey]

the fine-tuning?

ok my fix is good to go 😄 was using cached OAuth ticket for reconnecting to websocket, shame on me

@rain gazelle mine has gone the complete opposite direction toward hyperrealistic

[smoky oak]

I'd say my Amber Heard LoRA worked pretty good

[oak osprey]

boy i wish she were easier to root for, as a human. she would do great as Alice, with that face.

[oak osprey]

do you no longer have to provide a keyword to get that ?

[smoky oak]

I traied it to be keyword dependant, but the keyword is "Amber heard" so

[oak osprey]

eg. if you try to get Elon Musk out of it now, does it make him female/amber heard-like?

oh

https://images-ext-2.discordapp.net/external/ruNhIZA_nkyZ1QFRSIyhW6e372RZLE6IVcJqPtALeaw/https/tripleback.net/public/discord/1683305528.00363288807fefd4d1f63a74673351a0bbc8708.png?width=1600&height=900

added tim burton to the prompt

me: "hey gpt. what do you have to add to this fight?"
AI: "i got u. add (jumping) keyword"
me: "are you... sure....?"
https://images-ext-2.discordapp.net/external/ObB1x1n-D8GN_SIi7HifhhVPcq-8Jtk-P1L0S3uyuh4/https/tripleback.net/public/discord/1683305654.00734666658cdf997902867bf60106ed0c40466.png?width=1600&height=900

yours and Sytan's make her look stoned

i can hear the vocal fry and stoner mannerisms from here

https://i.imgur.com/hiltbrph.jpg

combined with Billie Eilish

hmm, why no preview

[rain gazelle]

yeah you lost alice lol

[oak osprey]

oh i didn't have alice in those keywords 😄

i forgot

[smoky oak]

Ambiller Hearilliesher

[rain gazelle]

I was just running some random stuff, but I can try Amber again...

I was doing cartoons again. 😛

[oak osprey]

wonder how well this model knows the members of the 1990s girl band, TLC

T-Bone, Left Eye, and Chili

nervously pretends he had to google those names

what the heck happened

[rain gazelle]

I'm not that familiar with her so I'm not the best at determining likeness....

let me see

[oak osprey]

https://tripleback.net/public/discord/1683306197.04192333931fa01ae574c6044a5bea63cb12c5.png

Clyde thinks this is too hot for TV

[rain gazelle]

This look like her?

[oak osprey]

oh i watched the entire johnny depp trial and that one kinda looks like she used a Snapchat filter

[rain gazelle]

Didn't she have red hair?

[oak osprey]

she seems to have so many "episodes" she goes through, yeah probably red hair seems easy enough to imagine

Greta Thunberg?

some of the deformations, man. if i met her in the woods i would definitely die from the Siren's call

did you inpaint for that?

great result without that then

i put your img through the img2img pipeline

https://tripleback.net/public/discord/1683307015.0378299095a70da799530aafa925af3b6ae2ab.png

settings

[rain gazelle]

Any guess who this was supposed to be? Gave her kind of an Asian flair...

[oak osprey]

looks like our girl Amber but maybe i'm looking through Amber-tinted lenses now

[rain gazelle]

LOL yeah, not Amber. 😄

superhero realm... Marvel, not DC

[oak osprey]

oh, i was also pretty happy with the result of Mia Sara. she played Sloane in 'Ferris Buellers Day Off'

using 1970s picnic at the park kodachrome polaroid keywords at the end for that soft focus

the polaroid keyword sometimes adds edges etc

[rain gazelle]

can't get the costume...

hmm... Mia Sara...

I'll poke my loaded models to see which know her.

[oak osprey]

this isn't very close to her likeness, fwiw. it's got the basics but she has slightly more cartoony/anime-like features than this

[rain gazelle]

I kinda wanna try it as a game... guess the person. haha

I'll start w/ an easy one.

[oak osprey]

Ray Jeno

https://tripleback.net/public/discord/1683308085.500352c3f193350d639bf499dd01d1ca542f2a.png

Neverending Dream gets so close

Dreamlike Photoreal has its upsides for this prompt but tbh the face is Wish.com version

[smoky oak]

I am pretty proud of my Amber Heard LoRA

not bad for 15 minutes of work lol

@bold cave

[oak osprey]

my friend would loove it. he's always trying to get more 'realistic' images with eg. stretch marks/light scarring

[bold cave]

I will let my work do the talking

however this looks like it's going to be a while because I am planning on redoing my entire regularization image catalog which is 1500 pix with the SB1.7 model

I still have yet to test the blend to see which percentages work better, but the challenge is on

we need to find out how the judging system is going to work tho. We need a mod to anonymize the pictures and put them up for votes

[rain gazelle]

Yeup 🙂

[oak osprey]

interesting idea, a channel where each user's posts are ranked against their own posts

[rain gazelle]

how bout this?

[oak osprey]

@rain gazelle Bill Nye?

[rain gazelle]

nope

[bold cave]

lmfaooo

u uncultured swine

[oak osprey]

he has a bowtie tho

[bold cave]

thats alek trebek rest in peace

[rain gazelle]

you got it!

[oak osprey]

that's no alex trebek. where's the moustache?

he is a BRAND

[bold cave]

????

[rain gazelle]

Ok... easy one...

[bold cave]

aaahnuld

cmon man

who doesnt know who arnold is

[smoky oak]

Thats a little extra lmao

[rain gazelle]

oh my gosh... this one... LOL

[bold cave]

lmfaoo

the lord of the rings guy

Ik ive seen him

[smoky oak]

makes me think of that one youtuber

[oak osprey]

that's the guy from Wilfred

[bold cave]

oh wait no

[rain gazelle]

I love the ones that are like... off. LOL

Yep that's him!

Elijah Wood

[bold cave]

oh it is

[rain gazelle]

how about:

[bold cave]

adam sandler

just a really shitty one

[oak osprey]

omg if you squint that's Andrew Sandler

erm, Adam

[rain gazelle]

last one of this generation set:

[bold cave]

zooey deschanel

dang im good

[rain gazelle]

dang, speedy mcgee over here

ok i'll run some more

[smoky oak]

made me think of him lmao

[oak osprey]

are you always squinting or something?

[bold cave]

lmaoo

yeh that elijah is way too skinny

[oak osprey]

wonder if any AI can reproduce Bruce Rivers, criminal attorney

[bold cave]

you can reproduce anyone ma boy

[oak osprey]

suit guy looks like young version

[bold cave]

just need to finetune

[oak osprey]

i mean without fine-tuning

[bold cave]

then no

[oak osprey]

😄

[bold cave]

even B-list celebs the AI gets wrong

[oak osprey]

you dont know that

LMAO b-list. it gets Ellen Degenerate wrong

[bold cave]

lmfaoooooooooo

there u go

[rain gazelle]

New set will be ready in about 5 minutes. 😛

Anyone venture a guess? 🙂

no one?

thought that was an easy one 😄

[neon crystal]

That fat assistant guy for Iron Man?

Crossed with Ronald Reagan

[smoky oak]

lmaooo

[wispy nest]

looks like Dick Nixon to me