🏞｜general-with-images

[proper kelp]

I’ve read a lot about it

[smoky oak]

We have it very different from each other

[proper kelp]

Yeh both sounds shitty

I was 225 on day of shot

I’m 170

Not dieting.. just dying I guess lol

Not funny but kinda is

[smoky oak]

mine is very much numbness, loss of motor skills, and tendon issues, hers is very much muscle tension, joint swelling, and restlessness/aches

[proper kelp]

Oh no

None of that cept the sole of my feet feel puffy and numb wi th needles

[smoky oak]

mine just started one day

[proper kelp]

But mine is like..

You ever scrape your knee real good?

All shines and red?

And stings like the devil?

Imagine stretching that across patches of skin

[smoky oak]

Oh yeah, we get that as well. We have dermals ensitivity along with allodynia, where even a brush against a wall feels like literal rug burn for hours

neurological issues are no joke

luckily, we are medicated for neuropathy, and it helps a lot

she is more consistent, I am more flare based

[proper kelp]

What works?

[smoky oak]

Gabapentin

its for neuropathy, and diabetes nerve damage

[proper kelp]

Only medical pot has helped very much at all that shit only gave me horrible side effects

[smoky oak]

when i take it, it reduces my loss of motor skills and tingling

[proper kelp]

And they kept hacking up the dosage anyway

I was getting dizzy when I’d stand seeing big white spots in the dark

Heh

[smoky oak]

for her its the only thing that allows her to get out of bed, if not, her muscles literally stiffen up

[proper kelp]

Ow

Sorry

I have endless sympathy for chronic pain sufferers now

I mean I always did

But now more than ever

[smoky oak]

yeah, she had it all life and was in denial. Or like... she thought it was normal? Until I did digging when my neuropathy started out of nowhere one day and I was rushed to the hospital

[proper kelp]

Ugh

[smoky oak]

once that happened and I got my diagnosis and was sharing all the issues I have she was like... shit man, I have it too

[proper kelp]

What triggered it?

[smoky oak]

no clue

I had never had it, and then that single night was the worst I ever had

[proper kelp]

Mine WAS the Moderna first shot .. never had any of this get the shot and within hours I’m on fire all over

They’ll end up figuring it all out

After I’m long dead

[smoky oak]

lost all feeling in both legs, and my left arm, and I had almost no equilibrium, and it was just itching, and burning, and sort of buzzing feeling. Like a phone in your pocket, but nonstop and way more intense, but through my whole body

[proper kelp]

Probably

Ugh

Yeah

Mine

Is like

[smoky oak]

and prior to that i was dealing with rapid short term memory loss

[proper kelp]

On its best day

[smoky oak]

(fibro fog as its known as now)

[proper kelp]

If I lay in bed and just ‘feel’

It feels like surging electricity

Through my whole body

I can feel it vibrating

[smoky oak]

like humming?

[proper kelp]

Like static

[smoky oak]

yeahhh

[proper kelp]

Yes!

[smoky oak]

its a very specific tone too

[proper kelp]

I use that word all to w time

[smoky oak]

its a very very specific frequency of humming. For both my mom and I

[proper kelp]

Mine is sharp and stings .. and just bums 24:7

If I get hot

Flares

Sometimes for days c

It’s hell

[smoky oak]

mine is less painful, more annoying. I stop holding things with no warning sometimes. And I will also be standing and my legs will disconnect lol

windows drive disconnect sound followed by me falling over into a wall

[proper kelp]

Sadly I’ve almOt got used to the reg day to day pain and discomfort but man when my shit flares it feels like actual fire on my skin .. it’s horrific and it’s all over my body

Too much

I end up trying to just lay there and breathe through out

It

Like I’m fuckin having a baby

Lol

[smoky oak]

oh! I do have a very specific one that I fucking hate

[proper kelp]

?

[smoky oak]

It has only happened a couple times

my body just decides its hives time

[proper kelp]

Don’t jinx yourself

Heh

Ouch

[smoky oak]

It happened once on my left arm, where it started itching for no reason while I was trying to sleep. Woke up to an arm covered in hives

doctor deduced it was a false nerve reaction

[proper kelp]

Now my skin will swell and get red but no hives yet.. but my legs and covered in scabs and scratches from when I fall asleep and can’t stop myself from scratching the burning itch

[smoky oak]

or a "phantom trigger/response" as he said

[proper kelp]

Yeah

So weird

Flare ups can come outta nowhere

No reason

I’m careful with what I eat I manage my symptoms well

[smoky oak]

I would have to say the most debilitating for my mom and I is the allodynia

[proper kelp]

And stil I’ll flare

[smoky oak]

I have found no triggers, personally. I will go weeks without it and then BOOM

butits never been consistent

but I do know that when it does happen, if I make myself warm it is less bad

[proper kelp]

I’ve been up three days now.. my eye sockets feel bruised I’ve done this so many times . I’ll lay in bed until its just horrible and hot and I get up

Sit up alone all night in the living room

Heh

Hence the lame TikToks

I make

[smoky oak]

trust me, if there is anybody who understands the hell of not sleeping, its me

[proper kelp]

It’s a life wrecker

[smoky oak]

my fibromyalgia is NOTHING compared to my sleep disorders I used to suffer from

[proper kelp]

Depressing as fuck at tens

[smoky oak]

I was literally dying

[proper kelp]

Times

[smoky oak]

but a doctor saved my life after a year of decline

[proper kelp]

How?

[smoky oak]

he found the only drug my body responded to. Literally dozens of attempts, only one worked

the list of sleep disorders I had is a mile long

[proper kelp]

Nice

[smoky oak]

(now believed to be connected to my fibro)

[proper kelp]

They pretty much told my this

[smoky oak]

of all of the pills in the world to save my life, it was fucking benadryl

[proper kelp]

Neuropathy isn’t something you can cure.. you have to cure the undying disease causing the neuropathy

Except sadly they don’t see how they’ll ever undo a vaccine

Not gonna suck it back out

For sure

[green plover]

Not a single artist name used 🙂

[proper kelp]

So I’m just fucked

Nice

I stopped smoking weed ag your age

Hadn’t touched it in decades

Not my thing etc

But now

Can’t live without it

It does nothing for the actual pain imo

It just takes your mind ages

Away

Before it was impossible not to focus on it

[dense tapir]

I wish it were legal in my state for all my aches, and pains (age catching up to me). I tried to learn guitar again and I almost needed surgery plus I have lost some mobility forever.

[proper kelp]

What state?

No medical?

[dense tapir]

I would love to grow some and use as a tea but not smoke it

[proper kelp]

I want to move to Colorado

Heh

[dense tapir]

no medical, no anythingthis is Bible belt country

[proper kelp]

Ugh

Yeah I’m being pressured to move to Texas

[smoky oak]

Sorry, I accidentally unplugged my router lol

[proper kelp]

Fuck

And that

No?

No

Argh np

Lol

[dense tapir]

State next to me I can drive to in 5 mins and it allows, omg, CBD oil to be sold, lol

[smoky oak]

Splendulicious

[proper kelp]

Useles

Imo

[dense tapir]

I know

[proper kelp]

The weed o get.., cost soooo much

And is weak

65 for 3.5 grams of like 13% thx

Thc

[dense tapir]

you know it took my states decades before they allowed the lottery and many times the office would have runs to the other state that it had once a week. I asked them why if they were opposed to gambling in our state did they do it? Hypocrites.

[proper kelp]

Lol

[smoky oak]

So, Incase anybody was reading and was curious about what sleep issues I had, here they are

[proper kelp]

I fuckin hate hypocrisy

It’s a pet peeve

[dense tapir]

They were, in essence, NIMBY types.

[smoky oak]

Hereditary insomnia (life long, never waning), nucturnal sleep alignment disorder, chronic excessive daytime sleepiness disorder, occasional sleep apnea, hyper active sleep paralysis, sleep induced psychosis, early onset narcolepsy (that was a fun one), elongated circadian rhythm disorder, REM sleep behavior disorder, and more I can't remember at the moment

So I got 0 sleep, basically

[proper kelp]

Hmm so finally someone to bullshit with at 4am?

[smoky oak]

For months, and it got so bad my body started shutting down

[proper kelp]

Heh

Ugh

Sorry

[smoky oak]

I don't even remember that time of my life outside of pain and anger, but also being so exhausted I could take even be mad

And of all the fucking drugs, it was benadryl that saved me lmao

[proper kelp]

I made my first TikTok face filter last night and it got accepted and added.. now a bunch of ppl are using it :),

Mainly kids but hey

[smoky oak]

I was averaging middle to low single digits of sleep a week while in my senior year of highschool

[proper kelp]

It’s NF black smile face paint

That’s no good

[dense tapir]

I had to give up learningthe guitar (I own 3) or I would have required surgery. I still have pains when I type as my fret hand the thumb joints, and knuckles are always in pain, and the picking hand I got radial tunnel syndrome. I had never heard of that before but I can barely pick anything up now.

[smoky oak]

It was far from no good, it's unlivable long term

[proper kelp]

And y’all are young damn

[smoky oak]

But again, we have benadryl to thank for me still being around

So you can thank benadryl for unleashing me on this world hehehehe

[proper kelp]

When I was y’all’s age I was made out of steel and diesel and was fuckin shit up 24:7 🙂

[smoky oak]

It's their fault, not mine :p

[dense tapir]

I am past young, lol

[proper kelp]

Oh k

Well he is

I’m 48

[dense tapir]

yeah

[proper kelp]

:/

[dense tapir]

I got 10 on ya

[proper kelp]

Ah k

Cool

If you saw me you’d think I was lying anyway

I’m defected

Freak of nature

[smoky oak]

Let me just say, we are so fucking lucky that the pills I need to regulate my sleep are some of the cheapest widely available pills in the world, and not some insanely expensive prescription shit they tested to see if it worked

[dense tapir]

I never once thought the guitar who have fucked me over the way they did. ffs. I tried learning in 2005 but did not.

[smoky oak]

But alas, it is time for me to flex on myself a year ago, and sleep well :p

[dense tapir]

bene give me a restless sleep so I dream like mad which prooves it was restless as a nice deep sleep you will not remember them.

[smoky oak]

Fucking loser, couldn't even sleep :p

OH!

[proper kelp]

Me at 47 a few months back

[smoky oak]

I did unlock a super power though

2, actually

[dense tapir]

I am a lucid dreamer since a kid.

[proper kelp]

Same

I could fly

[dense tapir]

I have 4 or 5 super powers

[proper kelp]

Heh

[smoky oak]

Having dealt with such severe sleep deprivation, I can go 24 hours with no sleep and feel fucking golden, though I can't sustain it unless I wanna find myself fucked up again

[proper kelp]

For most my life

I’m there now

[smoky oak]

Normal tired is nothing to me, but too much normal tired throws me off into a spiral

But the other, arguably more cool one is

[dense tapir]

I used to (in my youth) go three days no sleep, 1 day sleep, and repeat.

[smoky oak]

With how severe my sleep paralysis got, it doesn't affect me at all

[dense tapir]

I can't do that shit now

[smoky oak]

I got so used to it happening every time I slept, that I kinda just shrug it off and fall back asleep now

Which also lead into me being able to lucid dream more regularly, as I use sleep paralysis as a check for lucid dreaming, which works really damn well BTW

[proper kelp]

I need to make some money from home hopefully dealing in graphic arts.. what’s my best hope? Lol

Atm I was looking into learning AR filter creation and finding brands to promote

[smoky oak]

I'm still trying to find a way to make some money. I am broke with two capital empty pockets :<

[proper kelp]

.. :/

[smoky oak]

BUT AS I SAID

I really need to sleep

It's 6AM

This is really irresponsible of me, flying off my schedule

Naughty me :p

[proper kelp]

Heh

Happened

Happens to

Ugh

I can’t toe

Type

Lol

[smoky oak]

Farewell, fellow sufferers lol

[proper kelp]

Mobiles :/

Later

[proper kelp]

Post

Heh

Didn’t they make it where you can’t generate famous ppl anymore?

I read some crap :/

[fresh hound]

think that was when 2.0 was released intially - but it wasnt removed just needed more training

someone can correct me if thats wrong 😄

[dense tapir]

They only removed most artists and actors but with training are easy to bring back in for both. The body doubles, and look alike arts styles, are in there so takes little to make them back to what they were.

[proper kelp]

Wanna read some crazy declassified CIA bullshit next time you get seriously bored

https://www.cia.gov/readingroom/document/cia-rdp96-00788r001700210016-5

It’s long and weird but cool I guess

[chilly lance]

Yeah its actually really hard to keep out the style of anyone because lets face it most concept artists kinda look alike. Just open artstation and you will see like 3 distinctive art styles max.

[dense tapir]

Well, I was speaking more of the classics, but in this day everyone has some artist before them in their style.

To make it worse in the current me me (not to be confused with meme) I need attention generation they will copy what they think is working for clicks, attention, etc... sad

[chilly lance]

I mean its just buisness, consumers know a specific style so its saver to produce games and movies with that art style. Thats why pages like artstation look like they were made by 3 people.

[dense tapir]

Art was never ever about the business as sad as that sounds. Starving Artist is a name dating back thousands of years for a reason. Let the art flow because art suffers the moment you chase after the bank roll.

Most times their art didn't begin to pay off until they died.

[stoic junco]

Why doesnt it show png info? :/

[grizzled sage]

art suffers the moment you chase after the bank roll

I disagree. I think the idea that art suffers if it's a career is delusional. People are allowed to make a living and earn money. We likely wouldn't have the medium of animation if there wasn't a business incentive there. The first animated movie was made barely 100 years ago in 1904. There was nothing stopping animated movies from being created in the past. The technology was all there to do a full production, but nobody was incentivized to explore this amazing new art medium until there was a huge incentive in the name of business.

[dense tapir]

Disagree

[grizzled sage]

You see that in music a lot too, like somehow a band are sellouts if they make money

Even davinci took commissions for most of his work

[dense tapir]

of course.

[grizzled sage]

The rennaissance happened because of patrons

[dense tapir]

it was never his driving force though and that is precisely why our society is so fucked up right now chasing after the cheddar.

[grizzled sage]

You knew him? Neat

[dense tapir]

as much as you did, yeah.

[grizzled sage]

Well, i'm going off historical documents, not inventing his personal motives

I've poured over translated versions of his notes too. Dude was very much after the cheddar

[dense tapir]

If you don't have the drive all the money in the world will be pffft.

[grizzled sage]

I have no qualms with economic incentives. I think good ol Mikey Douglas nailed it in the 80s. Greed is Good

[dense tapir]

Greed is bad. Sorry, I disagree with that and I hope that one day we go beyond that.

[grizzled sage]

We wouldn't have the internet spanning across the entirety of the globe with the bandwidth it has, if it wasn't for economic incentives

[earnest solar]

Can someone remind me how to avoid this doubling up thing when trying to make images bigger than the standard of the model?

[dense tapir]

I will gladly go back to dialup over this modern fucked up world that all this Internet helped to fuel. Social media is a disaster and the human experiment showing just how humans are inherently stupid, self centered and easily manipulated. Going off the subject but the day is fast approaching where making a living will be hard if not impossible for most so what will they have to turn to? Their drive. I like to own XYZ not because I have to but because I want to. That will become the norm as technology replaces every job eventually.

[dense tapir]

Like Sisko's father and his restaurant in DS9. While SCI-FI now will be how it is in the future.

[grizzled sage]

I wouldn't lol. I grew up in that 1990s world and it was shittier. People are like "electronics are too much!" like.. shit.. my 500mb HDD costed $500 and failed constantly

nintendo carts were $100

[earnest solar]

This conversation is depressing. Kitten kong, push it off the screen.

[grizzled sage]

the 90s was also a LOT more rapey. I think toxic masculinity peaked

[dense tapir]

I grew up in the 1970s, 80s, and 90s and give me any of those over this shit.

[grizzled sage]

oh the leaded gasoline years. gotcha

a big reason why i think toxicity peaked in the 90s

lol. Bringing up the fact that smog was laced with aerosolized lead for many decades isn't something older generations like to see recognized

its sort of scary since all of those people are running things currently

[green plover]

you mean like this?

[stoic junco]

😦

[kindred parrot]

#1072220168534642768

[green plover]

I didn't strip the tags out

[random nest]

background Star galaxy Bella Hadid

[green plover]

Try again. I know for a fact I dragged and dropped it this time

[stoic junco]

yeah, must be something on my end. its currently happening with more images but it was working just yesterday 😦

[kindred parrot]

There is no bot on this server. Please check out #1072220168534642768 for more details

[stoic junco]

btw are you using more or less the same set of negative prompts everywhere? seeing how you're mentioning a lot of stuff that has to do with people and this picture has none 🙂

[green plover]

Yes. I have a standard negative prompt

[stoic junco]

thanks 🙂

[green plover]

"crown" is not part of my standard negative prompt. That is still in there from my Pandora picture who is not supposed to have a crown

[stoic junco]

any chance you could upload the picture somewhere? i just googled it might be something with downloading from discord. would be great to confirm thats the problem

[wispy nest]

have you opened image in browser before saving?

[stoic junco]

nope

[green plover]

Give me a few. I just loaded 1.5 and it looks like I need to mod the style a bit. There is more monet that I want

[stoic junco]

just tried it and the problem persists

[green plover]

https://drive.google.com/file/d/1gDPo09wxyYQO05y_bxc2gPosdAX51gzD/view?usp=sharing That is as far above and beyond as I am going. if that doesn't work, try reinstalling

[boreal hare]

Which look better?

[green plover]

imo the 2nd, but you may want to outpaint the top of the head imho

[wispy nest]

fingers always betray me

[cursive lance]

no is furry ok

[glacial token]

?

[wispy nest]

as long as it is within the rules, no nsfw

check pinned message

[glacial token]

@wispy nest is my picture okay ?

[wispy nest]

yep

[reef sail]

joe biden as marvel superhero

[smoky oak]

Captain Sleepy

damn, this is immaculate lmao

this one too haha

[reef sail]

i'm in disbelief

[smoky oak]

I just found them on google lol

[river flame]

i cant sleep bcus of controlnet

[reef sail]

img2img is pure bliss

[cursive lance]

gta san andreas the movie :

[grizzled sage]

@wispy nest

[wispy nest]

OH MY GOD

ITS

A FLUFF BALL WITH LEGS

Lmao

[glad galleon]

So I'm an artist who makes abstract digital collages, I'd like to train a model, dreambooth, or something on my work, but because my collected work is so visually different from one another, i'm not really sure how i'd even begin. I have over 1000 pieces, but because my work is abstract and communicates messages beyond words, i wouldn't even know how to caption my stuff. Here are a few examples of my stuff, if anyone has any advice on how to proceed, i'd really appreciate it.

[stoic junco]

Thank you so much. That worked, so its something about how discord changes the files.

[green plover]

Cool! Glad it worked

[stoic junco]

looks beautiful in high res btw 🙂

out of interest - why are you using face restoration on an image like that? 🙂

[green plover]

Because I forgot to turn it off from the previous image

😛

Which I'm not sure if I can post on this server

[glacial token]

since there's not alot of dragons out there in the wild i thought i would train my own model in dreambooth with my own dragons enjoy this picture

[green plover]

Daddy?

[glacial token]

my rtx 4090 is on fire from this i hope everyone enjoys this

[cursive lance]

my gt 256 is on fire from this

[dusk thunder]

potato

[smoky oak]

Lmfao

You know, now that I think about it, you could technically run stable diffusion on a high-end Android phone if it was properly ported lol

My phone has 12 gigs of RAM, all of which can be used for VRAM, so I could actually do some genuine generations on my phone lol

Granted, I'm imagining that would be significantly slower heh

[smoky oak]

@stoic pivot I made this image in SD

scenery, no humans, outdoors, sky, cloud, mountain, tree, snow, nature, forest, cloudy sky, winter, pine tree, sunset, landscape, mountainous horizon, gradient sky, bare tree, blue sky

that is what the auto tagger got from it ^

if I increase the scope

then we get this

scenery, no humans, outdoors, sky, cloud, mountain, tree, snow, nature, forest, cloudy sky, winter, pine tree, sunset, landscape, mountainous horizon, gradient sky, bare tree, blue sky, purple sky, evening, day, orange sky, twilight, grass, red sky

the more scope, the less certain, but the more potential details it can pick up on

or thus one

eye focus, space, no humans, glowing, blue eyes, planet, star (sky), extra eyes, close-up, solo, light particles, eldritch abomination, looking at viewer, tentacles, 1other, glowing eyes, scenery, sky, fantasy, starry sky, galaxy, earth (planet), monster, realistic, reflection

you can see how it understands it as something completely different to what we would see

[pseudo aspen]

Beautiful

[smoky oak]

thanks, its part of my wallpaper collection I am working on

[pseudo aspen]

I was attempting to generate some space wallpapers too. I'll inspire myself with your prompt

[dusky plover]

#1072220168534642768 813024

#🏞｜general-with-images cars

#1080946152318443610

#1072220168534642768 cars

#🏞｜general-with-images catd

😆

[glossy herald]

lol are you OK ?

[green plover]

could technically run it on a raspberry pi if you could port cuda to it

[earnest solar]

Cixi-Lu Jedi Lord of the Order of the White Rose.

[mossy night]

Beautiful. What model is that?

Hi! I'm having a little issue. When I use the OpenPose model from control net my image seems to come out way more pixelated. Does anyone know how I can fix that? Thanks

[devout reef]

any1 got suggestions for cool models to use as base merges?

[mossy night]

Woah

[smoky oak]

ping

oh, hello

[earnest solar]

Shirane Gekido

[fresh hound]

this is creepy

[stoic pivot]

Well, that was my point and I wanted to know is there was a way to rank words to see what affect it most and to know what tags the AI know. Like for me "trending on artstation" does nothing and is as good as "trending on martkation", but folk still use that by popularity without knowing what part in that prompt that is "good".

[fresh hound]

its rather close consistently with coca cola text

what upscaler do people tend to use in 1111s?

[stoic pivot]

After a debate about what prompts works that ended in the conclusion that "diamonds is a girls best friend" shall be translated to "diamonds are children's best friend" I had to test prompts and...: "anime, render, " work well but I got better result by remove the separation, it was like the separation took "anime" or "render", but "anime illustration" and "anime_render" gave just that every time with good result.

[stoic pivot]

I am testing popular prompt tags, one of the most popular is "sharp", but when I add "sharp" or the more popular "(sharp:1.3)" I can not see any difference in the 'sharpness", but if I changed the tag to "cute sharp looking robot bunny" I got a square (sharp looking) bunny.

[smoky oak]

I personally have never seen anybody prompt with "sharp"

[stoic pivot]

I used this one to grab some popular tags, https://civitai.com/gallery/208511?modelId=11888&modelVersionId=19806&infinite=false&returnUrl=%2Fmodels%2F11888%2Fcrosskemono

It have many trend tags like "best quality", "ultra detailed" and so on.

[stoic pivot]

Test of "trending on artstation": This prompt does change the image, in the test here I used "playful anime_render, cute robot bunny, forest background, " and then added "trending on deviantart" and then "trending on artstation", but it do not do much for the quality. If I generate 16 images with "trending on artastation" around half of them get "signatures" or "logotypes" imprinted on them.

[stoic pivot]

Blåhaj toaster?

[fresh hound]

its so fun to make hybrids

[sterile kiln]

Me (LoRA), with the prompt of this guy :
https://civitai.com/gallery/150219?modelId=4823&modelVersionId=15236&infinite=false&returnUrl=%2Fmodels%2F4823%2Fdeliberate

[stoic junco]

i think that trending on XYZ / XYZ contest winner does affect the image in midjourney. so people coming from that background might be using it, i know i do 🙂

[smoky oak]

Absolutely hate that

[wispy nest]

Never seen "sharp" as it's own thing, it's usually sharp focus?

trending is a weird prompt too o_O

[smoky oak]

@peak summit here you can see some results I have gotten from realistic vision V1.3

Hands down my favorite model for realism or semi-realism

https://civitai.com/models/4201/realistic-vision-v13-fantasyai

[smoky oak]

aww :>

[fresh hound]

THEY CAN FLY

[glossy herald]

bears

[fresh hound]

I'm so proud of my Son

[smoky oak]

haha, nice

[glossy herald]

Elden bear

[fresh hound]

bear in gta sa. turned out a lot better than I expected lmao

[cedar elm]

I got a DM from this server, it looks like a scam but I’m kinda convinced

Should I trust it?

[smoky oak]

scam

[cedar elm]

Damn I know, it just looks way more convincing than other scam’s though

[glossy herald]

already saw another user get hit by the same scam today. block him, make a ticket to #1010934719455707218 to raise the alert to the staff, and don't keep that screenshot here 😉

[cedar elm]

Sad

[smoky oak]

@kindred parrot Seems that @barren prawn is sending scams

ah, I always forget this server has a mod support, my bad

[glossy herald]

no prob but yeah, it stays the best so we can see what's happening when there is volume like that

[kindred parrot]

Shoot the screenshot to me quick in a modmail and I'll take a look at it!

#1010934719455707218

[cedar elm]

Alright thanks for the advice!

[smoky oak]

@cedar elmyour go, send it in the mod mail :p

[fresh hound]

HOLT - WHO GOES THERE!

hi

[mystic quartz]

Wanted to share a music video I made with SD back in November that's finally been released today https://youtu.be/m_XN_2k5wts

[glossy herald]

anime bear in skyrim elden ring style

[grizzled sage]

that's no Kodiak! thats a mosaic! ... eh.. its still early. little dry maybe

[fresh hound]

oh wow!

love that

[glossy herald]

a bear, intricate, headshot, highly detailed, digital painting, artstation,concept art, sharp focus, cinematic lighting, illustration, art by artgerm and greg rutkowski, alphonse mucha, cgsociety

on a mix model

realbiter

[fresh hound]

rush B!

[glossy herald]

the prompt is a little "cheating" given what I put inside, but it still lands results lol

[glossy herald]

lol thanks SD x)

WOW damn it worked (copying params from the model on civitai)

[glossy herald]

Terminat-child

[glossy herald]

xD

[glossy herald]

my model doesn't like bears

still cute

[glossy herald]

photorealistic, best quality, masterpiece, raw photo of upper body photo, a Bear, soldier of the austro-hungarian empire clothes, double breasted jacket with stripes, extremely detailed eyes and face, long legs, highest quality, fur texture, intricate details, (cinematic lighting), RAW photo, 8k
Negative prompt: airbrush, photoshop, plastic doll, plastic skin, easynegative, monochrome, (low quality, worst quality:1.4), illustration, cg, 3d, render, anime

(replace fur with skin for a human :p)

[fresh hound]

you are a wizard with prompts sir

getting some amazing results

[glossy herald]

nope

I'm a copy paste wizard

and I replace the subject with "bear"

lol

I'm only using the sample prompts that were given as example for my model but yeah, some of those are very nice

I'm taking notes :p

[fresh hound]

the funny thing is, i've been copying some of your prompts xD

[glossy herald]

trying this on a bear

[fresh hound]

god dayum!!!

[smoky oak]

I finally figured out what i was doing wrong with my LoRA's, and they are coming out soooo much better now

[glossy herald]

I can give you the name, on civitai it's RealBiter

can't link it though, it's also part nsfw

[fresh hound]

https://tenor.com/view/noice-nice-click-gif-8843762

[glossy herald]

hummmm

I'm stretching it it seems

photo portrait of a beautiful bear, (very long curlly fur, slim body), (evening dress:1.2), ultra detailed face, insanely detailed, highest quality, fur texture, ultra detailed fur, highres, intricate details, (cinematic lighting), RAW photo, 8k uhd, dslr, hdr, film grain, Fujifilm XT3, gloomy, dark night, depth of field, rim lighting, two tone lighting, dimly lit, low key
Negative prompt: airbrush, photoshop, plastic doll, plastic skin, easynegative, monochrome

[smoky oak]

man, this new LoRA training is sooo much better

[gusty chasm]

nature with fire

[smoky oak]

my Avatar LoRA is looking so much better now

[smoky oak]

I have been trying to make hot cyborg men for a while now

[smoky oak]

bro, my avatar LoRA is working SO GOOD NOW

[glossy herald]

show us samples of bear avatars !

[smoky oak]

Who would have guessed the way you prompt your LoRA captions would have SUCH a big difference

[glossy herald]

well... yeah, I did some things in captioning sometimes, thinking it was genius. turns out it wasn't and bad caption can SURE just destroy a training session

happy you found your way, this loos great !

[smoky oak]

I didn't realize what the point of captioning is

I thought it was to show the LoRA what its seeing, but its to tell it what parts are being added ontop

so if you tell it that the subject has black hair, you will have to specify black hair when using it, but if you don't, it will always associate black hair with it

I used to cram sooo many terms into my LoRA's, but it turns out the more terms you have, the less it understands the subject

works completely backwards to how I thought

[glossy herald]

yeah, it will take your tokens and try to feed into them everything it gets from the pictures.

The more token in a single caption, the slower the learning too, since the weight gets dispersed on all tokens.

And yeah, lots of tricks in what you choose to include or not in your caption

not exclusive to LORA though, you'll find the same on DB

[smoky oak]

I just had no idea, I was being told the exact opposite by some people

[glossy herald]

I understood it better when I tried to fine tune on a larger scale

[smoky oak]

I thought the tagging was teaching the LoRA what to associate with the trigger, but its actually telling what NOT to associate with it

[glossy herald]

everything you give gets trained, even regularisation.
So you can even train more by feeding good and diverse regularisation too, fiting your style or watever

[smoky oak]

I have a feeling i am about to start making WAY better LoRA's

[glossy herald]

adding a token will force lora to ask "what's in the picture that could go into that token". So when you describe the black hairs, that helps remove it from the other tokens.

[smoky oak]

yeah, you're only supposed to tag things that would be alterations ontop of it

[glossy herald]

choosing NOT to include the token "black hairs" is also cool, it just mean that every other token in that caption will get a little of the "black hair" influence instead

it's really a choice

[smoky oak]

@glossy heraldI have a LoRA question, if you don't mind. Maybe you can help me

[glossy herald]

like if you have the token "jacksully" :

• adding black hairs to the caption will make it so that you will later on be able to prompt jack with other hairs, since LoRA tried to not put "black hairs" into that token
• not adding black hairs to the caption will make it so you won't have to bother about jack's hair, it will always be black, no need to ask for it in your prompt

sure

but keep in mind, like I said, I mainly do dreambooth finetuning

[smoky oak]

So, how does one stop overbaking? As in, almost all of my LoRA's get artifacts above .4 strength

[glossy herald]

train less

[smoky oak]

is it that simple?

[glossy herald]

or add diversity in the dataset

[smoky oak]

just less steps?

[glossy herald]

no other way around it, yes

[smoky oak]

diversity, interesting

alright

[glossy herald]

the more diversity in the dataset, the more steps it needs to get trained completly, to understand it completly. and so the more you can train before overtraining

[smoky oak]

very interesting

[glossy herald]

also adding some good regularisation/class data can help push that process a little later on, but it's mostly the main dataset here.

[smoky oak]

that explains why my freshwater shrimp one ate up 9k samples with little to no issue

[glossy herald]

try to double check that you have NO DUPLICATE

[smoky oak]

30 different image angles with 9 different colors

[glossy herald]

anything that repeats for real in the dataset, like even the same angle, the same detail, ... will get trained really fast

it's mostly why we always try to stay away from texts, it repeats motifs the model already knows and focus the training on that

[smoky oak]

damn, so sometimes its better to have less images?

[glossy herald]

(aka why a single signature in the dataset can lead to those on all your pictures later on)

[smoky oak]

I think these images are fairly diverse, what do you think?

these are close, but with very different lighting and backgrounds

[glossy herald]

if they repeat or are too close one to each other ? yes for sure ! less is more ! adapt your number of steps. I try to speak in "repeats" and not "steps". I do 100 to 200 repeats, mearning 100 to 200 steps per pictures in the dataset. You adapt the training step count to the number of picture directly

well, I won't lie, I did lots of datasets, but even now, I can't always say if there isn't something repeating too much. What I do is, when I think I'm good, I train, and then I test it a lot, and try to find the biases.
Biases are things that got learned and repeats themselves even when not prompting for them

like in the disney model, prompting for a girl will give you a princess. bias

it's an ok bias in that example

but sometimes it can bother you

if that's the case, you go hunting in your dataset for the pictures that have that bias

and you remove/change some

it's my process to perfect a model

(I'm working on a guide on that whole training process, I'll link it to you when it's done, but not for a few days still I think)

here is the first part, dedicated to the different training methods out there currently used https://trainingsd-methods.carrd.co/

second part is on datasets and training parameters

[stoic junco]

which model is this? 🙂

[glossy herald]

"RealBiter" on civitai. Can't link it though, it's part NSFW

real great model if nyou ask me

"finetune mix" would be more exact

Hey people around here and other channels, there has been a certain wave of scam by PM recently. If you have been contacted about wining a cryptocurrency prize in a contest, don't hesitate to :
1/ report that person to the mod team through the #1010934719455707218 channel
2/ block them
Also of course, don't follow their instructions further

Lots of new users come around asking for this too, so if you see someone not sure, don't hesitate to direct them to this info please !

[grizzled sage]

it's ahem, Jake, Sully. As evident by the iconic line of the first where smurfette was like "My JAKE!"

[smoky oak]

@glossy heraldSo sorry to bother you again

I have one more question

[glossy herald]

Shoot

[smoky oak]

What happens if the LoRA doesn't really capture the subjects likeness super well, and still manages to overbake?

[glossy herald]

(I don't always just respond to ping in a second, just saying)

[glossy herald]

I'm not sure what you mean then. Any example or more explanation ?

It never manages to get it and just goes to shit before it

[smoky oak]

you can see here how its overbaked, but also doesn't really look exactly like him

like, its much better than before, but that is very few steps, and its already overbaked

its only 450 steps

on 9 images

(450 COMBINED)

[stoic pivot]

I have a model called, "Uber Realistic Porn Merge" and I love it for all the wrong reason, I should use it to make porn, but add "NFSW, nude" as negative prompt and it is good for fantasy and anime too and it often give me good result even with simple prompts.

[glossy herald]

When your model diverges without having had a low point in loss (what you are describing), I'm not always sure. For me it's usually when the training rate is too high, but it can also be mitigated with better regularisation data.
Regularisation data (or class data) is nice to give more room for training, it helps the model to "keep grounded" and let you keep training more before bleeding happens.

[smoky oak]

ok, so I have never used regularization data. Do you have an example of what that is, by chance?

[stoic junco]

awesome, thanks. it does look pretty good indeed

[glossy herald]

Not right now, i need to go for now, but I'll hit you up on that later today if you'd like

I'm on phone

[smoky oak]

oh-

I changed my prompt

it looks a lot better now

like

a LOT better

how weird

[stoic junco]

it would be awesome if you guys collected the info about model training into some doc. would be very useful for the rest of us in the future

[glossy herald]

We have added quite a lot to the #1080946152318443610 recently

But I'm making 2 guides on this currently too

[smoky oak]

wow, it looks way more like him now

[stoic junco]

awesome, i definitely plan to try some training but i dont have strong enough GPU for training so will probably have to pay for it...and having to do less iterations will be tremendously useful

[smoky oak]

its still not quite perfect, but damn, its a lot closer

[glossy herald]

Here is the first one
https://trainingsd-methods.carrd.co/

[stoic junco]

bookmarking this, cheers 🙂

[glossy herald]

And just saying, dreambooth in Google colab for 3 or 4 models per day works perfectly too

[stoic junco]

thats kinda the idea. you have to pay for it, right?

[glossy herald]

I was talking about the free tier.

It's about 3 to 4 models, or even more per day, depending on how much you prepare before starting

[stoic junco]

nice! didnt know. might as well start training soon in that case 🙂

[glossy herald]

At least for learning it's great

It shows it's limits when you start to do style training on big datasets, or full fine-tuning for sure

[glossy herald]

There is some explanation on what is class/regularisation data in my second guide but still a work in progress https://trainingsd-parameters.carrd.co/
Check the terminology page in particular.

[smoky oak]

thank you <3

[stoic pivot]

One more image made with "Uber Realistic Porn Merge - v1.2", then I have stop make this kind of porn 😄

[glossy herald]

And if you have any feedback on it, in particular if things aren't easy to understand, it's welcome. I'm trying to stay superficial and add links to relevant guides for further details, but i want it to be easy to understand

[smoky oak]

Sounds good. I would be willing to share some examples of LoRA prompting/training, and some XY graphs!

[stoic pivot]

Problem with anime and the "porn merge", I wanted to make the female a bit rounder, chubby, did not work, so I tested with fat and obese, but it just made the breast larger and fatter.

[smoky oak]

you would have better results with "curvy, thick, voluptuous, thick legs, wide hips" and such

from my experience at least

[stoic pivot]

"blue hair (fantast female) anime_illustration (intricate_details), curvy, thick, voluptuous, thick legs, wide hips, sitting field looking at flower legs together jacket blue cape, background cloudy summer sky large sun sparkling particles blushing hands up saturated colors full shot full shot seen from side"

I do test, I use "a porn merge" that is a bit crazy, it look like '2the realism" block all fat and curvy shapes on fantasy females 😄
I will soon go back to the normal models.

I wanted "fat" or "chubby" for that had saved words, and I am just in a test for length and with, I will not write anything about it, but some of the result may maybe be a part in some video next month, we have to see.

Odd, a bot blocked one message to be to "dirty", it was fully dressed anime girls with very curvy forms and large breast (not nude), I just added that it was made with "Anything Pony" checkpoint. Will not take any risk buy trying upload it again.

[smoky oak]

@glossy heraldI fixed the discrepancies between the character and my LoRA. Issue was I wasn't specifying a few things I needed. I also gave it a second epoch, and I trained it with the model I am using, and now its insanely good

Training image

SD result

it did a DAMN good job

and only took 6 minutes to train!

here is a better training image

[stoic pivot]

Prompt test detail: When adding "detailed" it change the look of the subject slightly, but when "intricate" is used it change the whole image and make it more intricate, if I use "intricate details" it also add extra details, the interesting part happen when we use "intricate_details" when the bunny is closer to the first image but with a bit more detail. So "intricate" is a good tag in combination with words as 'details'.

[stoic pivot]

"anime_render, intricate_details, full_shot, image of a cute soft robot bunny kneeling to pick flower in a forest with a blue sky as background" Stable Diffusion v1.5.

[stoic junco]

love the exploration you're doing here. intricate detail is extremely powerful in midjourney

[stoic pivot]

Test quality again: Many use "best/worst quality" in their prompt, but there is no definition of what best and quality shall accomplish, when I add "quality" it just do a random change, and the same goes for "best quality" and "best_quality", and I get almost same result with "make nice" as with "best quality". So my conclusion is that words like "best", "worst", "quality", "masterpiece" and so on are pointless word to use, if you want a "masterpiece" it is better to referee to what 'master' like "by Picasso" or "Salvador Dali style".

Model used was Stable Diffusion v1.5

[stoic junco]

Will you be writing an article or some kind of summary for these tests? Seems like a shame to have it just as random discord comments

[wispy nest]

Made this cus I was bored and wasn't getting any help with my other issue. It's had quite a bit of inpainting as it originally looked like she had stepped from a forest to somewhere else but I wanted it the other way around. I wanted it to look like she was stepping into a forest from somewhere else.

[pseudo aspen]

How much curvy if I may ask

[stoic pivot]

It was very curvy and the boob size was comical explosive, but it was all dressed and no nudity. But it was a bot, and as the old saying goes, "You can fuuk with the moderators and managers, but respect the bots tho they have no brain".

One of those odd and crazy v1.5 prompts, "fantasy_render glowing_butterflies glitter crystal_cave floating jellyfish ice_colors dust_particles intricate_details bokeh dark_environment fog rays sparkle night_sky".

[pseudo aspen]

Clyde is unfair. I've seen some images with anime girls in bikini and I get Clyded when I only show bare thighs. Welp, it's a bot

[stoic pivot]

I have no idea what all "reactions" mean, so a thumb up as backup.

Btw, what I write and post here do I see as public domain, do some want to add my test to a better source they can do so, but if some do so I hope they test and confirm what I wrote also work for them. I have ideas, they are not always correct.

[pseudo aspen]

Okay lol

[worldly tendon]

Clyde is rigged

[wispy nest]

There's one simple rule when you work with AI - avoid using words like "bad, good, worst, best" , they are meaningless for AI in most cases

[wispy basin]

How do I generate wolves with easy diffusion that don't look like they were stung by bees

[smoky oak]

easy diffusion?

I have generated some excellent wolves

not with easy diffusion, IDK what that is

[wispy nest]

no idea too

[wispy basin]

It's the windows GUI thing

[smoky oak]

I have made gorgeous animals using realistic vision 1.3

[pseudo aspen]

Simply gorgeous

[wispy basin]

So far I've only tried the default stable diffusion model

I imagine this might be the problem?

[smoky oak]

ah, theres your problem

yeah, the default models are not very good lol

https://civitai.com/models/4201/realistic-vision-v13-fantasyai

What I use ^

[pseudo aspen]

Yeah tbh models like Realistic-Vision 1.3 are far better than the official ones

[wispy basin]

will try that and see how it goes
I am also rather new to prompts and whatnot and usually just try to input something like "timberwolf" and nothing else with varying results

[smoky oak]

oh, thats also a huge issue there lol

you're gonna need to learn how to actually prompt if you want good results

[stoic pivot]

Everything is better than 1.5, even "Uber Realistic Porn Merge".

[wispy nest]

eh, it's not that hard, just look what other people doing , there are some common things you need to think about

default quality prompt + subject, subject details, pose, actions , background, lighting, style
Usually more then enough to get good result. You can skip some of them and let AI think about it

[long comet]

Он понимает по русски?

Я не могу разобраться

Всё на инглиш

Stable Diffusion

Not understand

[wispy nest]

проверил на 2 картинках - нет, только английский.

[long comet]

Эх

А ты можешь помочь?

[wispy nest]

с чем конкретно?

[long comet]

Хочу картинку типо кровь на кафиле

А то в гугле не интересно

(

[wispy nest]

дык берешь переводчик, суешь туда что хочешь нарисовать, переводишь, тоже самое делаешь с деталями , аля освещением, чо там еще может быть над...

и суешь в SD

[long comet]

Ой сложно

Ночь

Спать охото

Не думать

[wispy nest]

дык шо сложного то, открыть переводчик, да перевести что надо

[long comet]

Я хз где искать здесь

Что бы бот

Был

[stoic pivot]

One more image and then I call it a day.

[glossy herald]

Oui enfin c'est bien gentil un traducteur mais ça va vite être juste méga relou si tout le monde se met à parler dans sa langue natale

[wispy nest]

tu as raison

[long comet]

Ладно

[sterile kiln]

C'est sur meme

[glossy herald]

of course i'd love to just speak french with other french people around, but yeah, PMs are there for a reason, and we try to stay understandable to the largest audience, following the rules of the server we join

[long comet]

Cuka blat

[wispy nest]

Короч...
Тут бота нет, если хочешь генерить картинки на SD - посмотри в ютубе видосы, там есть русскоязычные чуваки, которые рассказуют чокак делать, натыкался рандомно.

Предупреждаю заранее - придется либо устанавливать к себе на комп - для этого в идеале иметь видюху nvidia от 8g vram и выше, либо использовать google collab , но там у тебя будет ограниченное количество времени в день, либо придется платить.

А в дальнейшем советую тут писать на английском, если хочешь, чтобы тебя понимали

Если есть еще чет - пиши в личку.

yep...that'll be last message here for her on Russian

[long comet]

Ладно спасибо большое

😉

[glossy herald]

и простите за языковой барьер, надеюсь, вам удастся запустить Stable Diffusion!

[stoic pivot]

Now last, it was running in the background, denoise was to high so it is a bit strange.

[wispy nest]

We have Russian speaking youtube peeps who explain things...I don't really like them , since they are doing like 30mins videos talking about nothing, but they are here...
It's just...laziness to look for things.

[wispy nest]

I like weird images like that...when I have no idea what I'm looking at , but it is still beautiful 😄

Hey guys, any of you studmuffins get kohyass lora creation working properly? I had it running and an update borked it. Also I want to know if anybody is using the newer LoCOn Lora models

[stoic junco]

WDYM? Its clearly a fairy in a sphere thats emiting cosmic rays all over a landscape 😄

[wispy nest]

I feel dumb...I think I generated tons of images with wrong model today lol

just realised they all look not as realistic 😦

[smoky oak]

Bro, my new Na'vi lora is fucking awesome

Mixed with taylor Lautner

ariana grande

The rock

(working on all the artifacts on lora blending)

[wispy nest]

lol

[smoky oak]

but the one that blew me away is I mixed it with my Scar LoRA for Scar from Arcane

My LoRA for Scar

vs what he looks like

and then

mixed with the avatar one

I cannot believe how well they play together

and this is only trained on 9 images

and its only trained on men, but can already do women. I am gonna expand it and try to see if I can get a single LoRA that can do male, female, and both species of Na'Vi in one

[pseudo aspen]

Avatar is awesome movie

[opal tangle]

released v3 https://civitai.com/models/14605/howls-moving-castle-interior-scenery-lora-ghibli-style-v3

[cold berry]

I love that movie! You can post your lora in our #1047197565365538826 forum as well!

[timber river]

What is pow?

[cold berry]

Weekly event! #⭐｜pow-info

[sonic vessel]

custom model. I'll release it soon.

[torn spindle]

do you use this with dreamstudio

[green plover]

@naive copper define minimally

[naive copper]

https://www.law.cornell.edu/supremecourt/text/499/340

more than that I guess

[sterile kiln]

Unless it has been updated since that Reddit post was posted the it doesn't work if --medvram is part of your commandline. And with 8GB only gets me 1or 2 shots before I have to restart the program or else I get OOM errors. I can also confirm that the 75 token max in the pos&neg prompts does make it so only the style adapter is needed.

This guy is right, without --medvram I have clip_vision tiadaptater_style working (ControlNet), otherwise not.
Anyone else who only has 8 GB of Vram and has this problem?

Without --medvram I can't go up to 1400x1400...

[opal tangle]

ty!`

[reef sail]

ai is great for remembering things

i saw this dog at a pet store earlier and i couldnt exactly remember what it looked like, but AI got my back

[misty furnace]

Torne Marsh in Spring, in the style of victo Ngai.

[wispy nest]

Is there a tensorflow installation for a local auto1111 ?

I don't quite follow enough to get everything straight but I have a 4080.

[torn spindle]

hi...can i use dreamstudio api with these models...sorry if it sounds dumb...im a newbie

[wispy nest]

I combined line art into a megalora and achieved this

[torn spindle]

hi...can i use dreamstudio api with these models...sorry if it sounds dumb...im a newbie

[wispy nest]

look up automattic1111 installation on youtube

[torn spindle]

thanks will it work with dreamstudio api?

[wispy nest]

You can install dreamstudio into it yes

[torn spindle]

okay...thanks....will send you a message if i have any question

[sterile kiln]

finally succeeded, mixing a model with a style (LoRA, OpenPose, ClipVision)

[grizzled sage]

clipvision is a nifty tool

[sterile kiln]

BUt atm, we can't use --medvram with it...
Because of this bug...
https://github.com/Mikubill/sd-webui-controlnet/issues/502

[grizzled sage]

is okay i has 16

[sterile kiln]

have 8GB ...

so can't use clipvision higher than 768x768...

[grizzled sage]

total bummer

[grizzled sage]

Illuminati Diffusion with openpose cnet

[reef sail]

these all look overcontrasted

[grizzled sage]

That's because he took tthe style from the overcontrasted google image and put it on those images. They all look the same and that's good, for the purposes of that proof of concept

[smoky oak]

Does anybody know how to fix this? This model is exceptionally good, but it crashes soooo much

[sterile kiln]

and ClipVision is limited by 75 token, so you can't do large prompts

[sterile kiln]

Search on github, on the repo

[smoky oak]

@green plover Was able to save the day for me :D

[wispy nest]

clip vision?

[sterile kiln]

https://www.youtube.com/watch?v=PbDdtPTYm_4

[thorny nymph]

バッグ

bag

[sterile kiln]

ClipVision is really good, but you can use Depth with an image, and use a prompt appropriate to the depth and topology of the google photo (or other). So it gives something completely new

[static tusk]

something like this. any thing, that you want 😆

[sharp thunder]

Sino may trainer dito

[smoky oak]

My toothless LoRA works!

:D

[grizzled sage]

mighty mouse + depth control + illuminati diffusion

[glossy herald]

hey ! no problem in being a nooby, we all have to start somewhere.
No, those models can't be used on dreamstudio.
To use those, you'll need to either install Stable Diffusion on your computer, or run it in the cloud. Check the #1080946152318443610 , you may find more on this to help you !

[smoky oak]

anybody know any programs that allow you to add directional resolution to images you use to train on?

for example what I mean is in regards to a very tall or wide image, where instead of cropping it, you encompass it in the 512x512 grid

[glossy herald]

You may find more help on that in #🔧｜finetune , not sure what to use for that

[glacial token]

this is what chat gpt gave me to do for an anime archer

[green plover]

"I know I've been so hard on you
I know I've told you lies
If I could have just one more wish
I'd wipe the cobwebs from my eyes"

[pseudo aspen]

Testing my DOOM Slayer LoRA. It's my first attempt so I think it's not bad

[glossy herald]

It's quite ok for a first try ! The first pic gives a little "overtrain" feel, the black lines and contours burned. A little less steps next time can help.
Then if you want to even more better it, test it a lot and see if there are not some problems, some things repeating too much for example, and then go hunting in your dataset to improve on those. But yep, already a good first try, well done !

[pseudo aspen]

Thx for advice!

[main agate]

https://giphy.com/gifs/MasterChefAU-masterchefau-masterchef-australia-masterchefau13-Tmzz3gyfpI3kk6B1r4

[fresh hound]

this gives such a prime 90's Arnold era style

[stoic junco]

nice, looks pretty versatile

[ripe cedar]

I wonder if/when there will be layers to use in SD. Where different components of an image can be separated.

[wispy nest]

smart bab, youtube logo

#🏞｜general-with-images smart baby, youtube logo

[dense solstice]

No Dream Bots anymore to create images on discord, you can look for dream studio to easy generate images on a website

Try #1025467151206854736 for help

[crystal delta]

@wispy nest

FAQ: I'm new here! How do I get started and is there a bot?

Welcome! There's currently no bot on the server to generate your images. Start by heading over to #1072220168534642768 to get yourself situated and help find the channels you are looking for! Please make sure you review our #✍🏼｜rules-and-tos and feel free to assign yourself some #👥｜roles as well! Answer any questions your may have at our #1072229020520947753. There are many ways of accessing Stable Diffusion, take a look at #1080946152318443610 to start your journey!

[pseudo aspen]

Another LoRA attempt. Now very close to the character. But AI most of the time misses her "horns" thing (Real art <- left) (AI art -> right) Any ideas how to point on that while training?

[uncut gate]

/love-hard-images

[fresh hound]

lol

[arctic drift]

#🏞｜general-with-images pepsi

[uncut gate]

picture-clock-stars-piano

#images, picture-clock-stars-piano

[fresh hound]

#are-you-guys-drunk?

[uncut gate]

/images, picture-clock-stars-piano

[arctic drift]

#a can of pepsi

[uncut gate]

/imagine, picture-clock-stars-piano

[uncut gate]

/imagine, picture-clock-stars-piano

[fresh hound]

Agent Smith - you are terminated.

[pseudo aspen]

Anybody?

[sterile kiln]

If you hadn't seen it last night, it gave me a lot of trouble, but it's beautiful!

[pseudo aspen]

It is

[stoic pivot]

Thank you AI, all I wished for in my "own" images was water marks and footers so I can't use my own images 😄

Prompt: high viewpoint wide scenic dark mountain photo

[pseudo aspen]

Lmao. Negative prompt time

[stoic pivot]

That is why the UI look like this? Few lines for the prompt and space for a dictionary of words in the negative prompt?

[sterile kiln]

Can anyone tell me what "Guess Mode" is in ControlNet?

[stoic pivot]

I do not know why it looked like that, but a "Reload UI" did fix it

It guess I guess, it like you can skip the prompt and the ControlNet try to gusee the content.

I saw a video about it, I shall see if I can find it.

https://youtu.be/bkkBY8JOBKA

Two images is made with (what is seen as) a professional prompt, "long shot scenic professional photograph of {scenery}, perfect viewpoint, highly detailed, wide-angle lens, hyper realistic, with dramatic sky, polarizing filter, natural lighting, vivid colors, everything in sharp focus, HDR, UHD, 64K".
The other with just "high viewpoint wide scenic mountain golden_hour".
So next test I may try can involve things like UHD, 8K, Hyper realistic, and so on.

[sterile kiln]

A nice logo for my desk, (CN Canny with Guess mode)

[wispy nest]

I feel like we need decent 1.5 model with 768x768 or even 1024x1024...

[stoic pivot]

Realism is hard and folk often think "hyper realistic" is the same as a "extra realistic" something, but realism is the best we can get. Hyper is just a word for "above" or "beyond", so hyper realism is when we remember grass as greener than it is and we as sound to space war to make it better than reality. In AI you can often remove "realistic", "photo realistic" and "hyper realistic" and replace those with "photo" that imply it is a realistic image.
But "realistic" can be useful of you have a fantasy landscape and want to add a "realistic rabbit", but that rabbit will never look more realistic than realistic, if you want realistic to have more weight you can test (realistic:1.2) rather than "Super duper hyper realistic".

Btw. Hyper-realism is also a bit of an artform, so you can try it as a style.

[wispy nest]

yea, realistic is about games, movies , cartoons, but not about photos.

Noone describes photos as "realistic".

"photo" or "raw photo" for actual photos, (35mm or camera models work too)
"whatever* realistic for games, cartoons, movies"

[stoic pivot]

"RAW photo" is a weird prompt for RAW photos often look rather crappy, RAW is not made to look good but rather to have as much data as possible to allow better post processing.

And nothing say realism as much as adding lens flares like reality is seen through a 8-element lens without a lens-hood 😄

[sterile kiln]

Me, supposed to be a GTA V painting

[stoic pivot]

I like it, but I'm not a gamer.

[sterile kiln]

I'm not particularly a big fan of GTA either, but we must admit that their realistic paintings have class. And then it's popular

[unreal zinc]

an orangutan smoking from a coal mine factory chimney

[sterile kiln]

Jeez, guys, someone found how to use Clip_Vision and --medvram 😭 🙏

[sterile kiln]

Could be a good GTA V painting

[stoic pivot]

It sad I have a bit of dyslexia and do not know where to post article like scrobbles and rants, for I found something I now call "prompt conflicts" that are rather common and hard to spot.
I wrote, "powerful intimidating bunny chibi wearing an intricate_detailed spartan armor and war helmet with green bronze and spikes walking down a old stone hallway fire in the background glow bokeh, dynamic pose", but the fire in the background is missing due to the hallway, this is a tag conflict and I can remove "glow", "bokeh" and "fire in the background" without change the image concept.

[stoic junco]

hehe

[glossy herald]

and now in my new round of "will I OOM"

not yet but damn I shouldn't have used Latent upscale there

32 minutes later

I bet it's gonna be crap

it didn't oom

it sure is crap though lol

[glossy herald]

Better but not there yet

let's ask for the help of controlnet

(last one before controlnet)

[sterile kiln]

Scribble?

[glossy herald]

the scribble comes from my skill in paint

but yeah, I'm trying the scribble mode

[sterile kiln]

Yeah, saw this

[glossy herald]

the results I showed were without it though

each test takes 10 minute x)

[glossy herald]

that was 20 minute with controlnet in the end....

but wow

in one single generation, never had something that coherent

this is 5120x1440

[static tusk]

the last one looks very faded like the vae file wasnt used

or maybe its using a lot of the white from the scribble for lighting

[glossy herald]

it's the wrong VAE, yep

[sterile kiln]

how do you get it to be so high in resolution? How much Vram do you have? Which GPU?

[glossy herald]

3090TI, 24GB

[sterile kiln]

-__-

[glossy herald]

even then, it's 100% memory and memory swap all the way

20 minutes for that last picture

it would be faster on a 2.X model but I didn't get controlnet to work on it yet (didn't try tbh)

[static tusk]

I havent tried 2.1 CN. Im waiting for someone to prune them down from the 9.5gb each

[glossy herald]

let's run another one, I added a volcano

[opal tangle]

https://civitai.com/models/17313/netherland-dwarf-rabbit-generator-beta made a EARLY BETA version of a generator for a specific rabbit breed :3

[glossy herald]

that is so cute !

we need more cuteness on civitAI !

thanks a rabbitbunch 🙂

better on the colors

but the composition is bad

[ancient trout]

Smart baby , cartoon , profile pic

[crystal delta]

@ancient trout

FAQ: I'm new here! How do I get started and is there a bot?

Welcome! There's currently no bot on the server to generate your images. Start by heading over to #1072220168534642768 to get yourself situated and help find the channels you are looking for! Please make sure you review our #✍🏼｜rules-and-tos and feel free to assign yourself some #👥｜roles as well! Answer any questions your may have at our #1072229020520947753. There are many ways of accessing Stable Diffusion, take a look at #1080946152318443610 to start your journey!

[wispy nest]

How many swords do you even need , jeez
Why is it so hard to get good sword gen...
Like 75% of them with 2 or more swords everywhere...another 15% - broken hands or something else

[glossy herald]

swords, axes, ... are really really hard to position well

even before controlnet, it was almost impossible to get if you didn't use img2img, or a model trained specificaly on that

[wispy nest]

yea , daggers are mess too, tried them too

[glossy herald]

today though, controlnet should help, but it won't let you use random composition

[wispy nest]

Shurikens usually gets right pose , like before throw , but Shurikens itself aren't even there lol

[glossy herald]

it's funny with cigarettes too

rarely positionned well

lots of things like that, that do seem natural aren't always learned correctly by the model. The fact that there are lots of pictures of swords crossing for example, doesn't make it simple for it to understand how swords work

[wispy nest]

yea 😦

[opal tangle]

youre welcome! sorry I had to reupload it https://civitai.com/models/17316/netherland-dwarf-rabbit-generator-beta

[wispy nest]

no lora on Civitai for swords either 😦

[glossy herald]

sorry with such fur and ears is just too cute

xD I removed highres fix

well at least you can clearly see controlnet and latent couple at work here

[opal tangle]

I love rabbits so much haha, might do a few more of my favorite breeds soon

first I want to expand and improve that one

@glossy herald off topic but I happen to know a few random facts about rabbits? have you ever seen how big some get?

[glossy herald]

I think I'll publish a full library of loras of my pets one day

[glossy herald]

I have seen that yeah ! in the recent years, those breeds became quite more popular

I do have a big cat in the same kind of size

[opal tangle]

yeah they did! another smaller breed which I think is healthier too is more popular

my rabbits are really smart, they can be trained just like dogs to voice commands

under rated and misunderstood animal

[glossy herald]

I had lots when I was a kid

I loved those rabits, but the countryside I was in got hit by a sickness for rabits, we lots all of them 😢

[wispy nest]

About cigarettes, lol...
Wasn't trying to , idea was lowering sword weight, AI made this 😄

aaand...no swords at all with 0.5 weight...yeaaaa....

[glossy herald]

that's some kind of "sand people ninja" look

it's quite cool

not what you were looking for but I like it

face is nice too

[wispy nest]

I'm experimenting with ripped cloth 3rd day now , it's just too good 😄

[glossy herald]

good night everyone, I'll try to experiment with sleeping for a change

[wispy nest]

Goodnight