🏞｜general-with-images

[sterile temple]

refreshing

spicy

most memorable hero name

[shut sinew]

asnconnenns

[sterile temple]

I'm surprised how flexible this model is

[dense tapir]

@smoky oak I think when all of this moves to Tensor Coresis when we see the big boy gains for the lovelace vs ampere cards. Look at this.

Probably need a fast machine to take full on advantage too.

[tired peak]

pretty good@

[tired peak]

these are all great

[sterile temple]

thanks 🐰

[dense tapir]

Those are all fantastic

[sterile temple]

cheers

[wispy nest]

https://media.discordapp.net/attachments/1099462662468927600/1113432800629104640/00017-427526085.png?width=422&height=675

[sterile temple]

[oak osprey]

huh? 2.1 uses open source text encoder and v-prediction, is higher resolution, and better at prompt comprehension and textures due to the higher res. v-prediction models work with high cfg with fewer artifacts etc

also already throws the last layer away and essentially paved the way for sdxl, which does all of this and then some

2.1 is the best, currently it is the state of the art.

see #1045349359044280360

[shut sinew]

Models based on 1.5 tend to perform better imo

[oak osprey]

speed wise? compile the unet with torch inductor and voila.

[shut sinew]

No quality wise

I dont have speed issues

[oak osprey]

haha. no it is definitely better

[shut sinew]

I think it largely depends on what youre trying to do

[oak osprey]

the resolution is higher and has better contrast

especially when you fine tune it

[shut sinew]

Eh whatever you prefer, as long as it works for you

[oak osprey]

i mean, i fine tune models and 1.5 is like a toy

havent used it in a month

[shut sinew]

Do you have anything to back your claims?

[oak osprey]

what you want?

[shut sinew]

Dunno just curious

[oak osprey]

a 2.1 model won the last contest on this server too

[tropic shell]

I love how adorable this is

[tropic shell]

Love the colors and the overall clouds

[oak osprey]

it's currently one of my favourite images

thanks 🙂

my friend who pays for MJ, looked at my model and decided to stop paying and instead use fine-tuned SD models

he also upscaled that with a GAN and made it his wallpaper lmao

[tropic shell]

I can see why. I always love creating that kind of stuff. It's fun to experiment and to have fun.

[boreal falcon]

Well MJ lacks now really behind the controllability of Stable Diffusion Adobe Firefly/generative fill (PS), Leonardo.ai and others

Except for image quality

And you pay on top of that

[tropic shell]

I enjoy studying the relationships between language and the way SD interprets shape, color, form, replication, and textures.

[boreal falcon]

I mean MJ pro subscription is basically more expensive than Adobe Creative Cloud for me

[tropic shell]

Creative Cloud. Haha, don't get me started

[oak osprey]

midjourney is interesting and you can't just look at it in a vacuum of how much you pay for which features. it has a community in its Discord server that draws people in, too

with SD it's mostly local generations, and with MJ it's a group of people in a channel working on it either alone or together. Blue Willow uses SD 2.1 but hides implementation details, so you can do "open-source image gens" with friends on Discord with that.

that's why i wrote my Discord bot, so i could have fun with friends and select whatever model we want to use

plus you can always import the MJ image into A1111 and do CN stuff on it if that's what you need to do

[boreal falcon]

@oak ospreythats why i say those generative AI images competitors have kinda their niche of users and customers

[oak osprey]

don't bother trying to explain, just ignore anyone who isn't directly answering the question

[boreal falcon]

for me personally it definitelly doesnt make sense anymore to pay for MJ

for others it doe

doe

does

[oak osprey]

i would never pay for controlnet, in its current state

that's the new inpainting method that i'm calling ADHD Outpainting

[boreal falcon]

lel

[oak osprey]

my understanding is it cascades diffusion from the mask point outward to generate the rest of the image, it started with a jar of shrek on a plate at a picnic and then made that

[boreal falcon]

many people seem to use controlnet

[oak osprey]

it has many features and some are more or less fleshed-out than others, eg. Tiles works really well and is state-of-the-art but OpenPose gets little attention and needs drastic improvement

it seems like sometimes they're spreading themselves a bit thin on development resources/brainpower but i get it, wanting to just get it out there ahead of anyone else. this Inpainting stuff was rushed out to release to "beat Adobe"

[boreal falcon]

a key guy from the Firefly development team plays around with it too now to implement it somehow into Firefly

but i dunno, never played around with Controlnet in SD

there wasnt even ControlNet when i used SD i think

[oak osprey]

Controlnet's license allows it to be taken wholesale by Apache and integrated into Firefly, without even returning improvements to the community. all they have to do is state they are using Controlnet as a base and what improvements they've made on top

they can't call firefly Controlnet though even once it's integrated. the Apache license doesn't allow the inhertiance of trademarks.

[boreal falcon]

i have to rewatch that livestream again to see what he exactly said again

but he definitelly spoke about experimenting with Controlnet for Firefly

[oak osprey]

oh and Adobe can sue contributors of ControlNet if any patents are infringed upon, thanks to the license chosen by Controlnet

[boreal falcon]

but that one has to be for web UI or API when it comes out

Photoshop will have its own implementation of it i guess

[boreal falcon]

huch?

[oak osprey]

yeah, probably. everyone is spending a lot of time reinventing the wheel this year, so that we all have our own versions of everything

[boreal falcon]

true

[oak osprey]

cooperation is now a bad word

[boreal falcon]

each according to their workflow and needs

[oak osprey]

nope it's Not Invented Here syndrome

Not invented here (NIH) is the tendency to avoid using or buying products, research, standards, or knowledge from external origins. It is usually adopted by social, corporate, or institutional cultures. Research illustrates a strong bias against ideas from the outside.[1]

it holds us back quite a lot

[boreal falcon]

that sounds like SD community tbh lol

part of it

[oak osprey]

yeah exactly

ComfyUI has one diffusion implementation, different from A1111's, and both of those are different still from the Huggingface Diffusers implementations, and each of them have their own unique bugs and issues that make it hard to reproduce the stuff the researchers did, that came up with these tools

in Diffusers if you want to run long prompts, you need to use Compel to handle prompt embed generation or the community LPW pipeline, and A1111 supports compel and some internal parser that was allegedly taken from NovelAI, lmao

[boreal falcon]

how is NovelAI even doing today

[oak osprey]

no clue, i've never used it 😄

it seems centric around Anime or NSFW, neither of which are things that interest me

[kind quartz]

@dense tapir are you using deep depth focus?

[dense tapir]

Never heard of that. Is that an extension?

[kind quartz]

no, it is just prompt words.

i love those small models in gardens. Probably reminds childhood when building models

[neon crystal]

coughs

[dense tapir]

tilt shift photography always reminds me of Mr. Roger's Neighborhood.

[kind quartz]

i also used radial blur, but it is something absolutely worthless there.
Will try tilt shift how it will work then

[oak osprey]

i love how it doesn't know how tracks work

me: "What do you want to see today, GPT?"
GPT: "Create a sleek design for a go-kart/locomotive hybrid that emphasizes speed and functionality while still being energy efficient++. Focus on integrating aerodynamics and hybrid technology-- with a nod to vintage locomotives-- to create a futuristic and eye-catching vehicle. Minimize --unnecessary adornments that would add drag and weight to the final design."

it has an autonomous companion

[kind quartz]

not sure what you were using. but i personaly prefer deep depth focus. Tried now tilt shift and somehow weird results will try again.

[oak osprey]

mine is tilt-shift and deep depth focus

the go-kart stuff isn't

@kind quartz have you tried cutaway diagram ?

shows what's inside

[kind quartz]

no, will try now

[dense tapir]

I have no idea if SD can do it right just saying that is what the technique is called.

There is a whole lot of stuff stable diffusion can't do right so I wouldn't be surprised.

[kind quartz]

i think they can very close because generated depth maps are very very close to reality

dont check this image too closely, but i like the shot 🙂

[dense tapir]

I dunno but if MJ were able to be used in automatic1111, and trained for, I am not so sure I would use SD, or SDXL, or anything SAI but I can;t so a moot point, and I stick with SD.

[kind quartz]

now i cant get rid of it 😦

[wispy nest]

made some orcs today (all high res)

[wispy nest]

https://media.discordapp.net/attachments/1099462662468927600/1113424476361732177/00008-3123334449.0.png?width=422&height=675

[oak osprey]

Midjourney and Stability work together on stuff

MJ is allegedly cascaded diffusion which is how ControlNet works now

[dark flare]

Someone's having a bad day!

[dark flare]

wtf

[kind quartz]

i have him driving bike 🙂
Not sure i posted it maybe yes.

[dense tapir]

LOL

[kind quartz]

it is raw output, no any additional things 🙂

[dense tapir]

pretty sweet then

[oak osprey]

i tried to make him race but he just flies

torpedoooo

so the light/darkness of that image is definitely from the terminal SNR i applied during training

trying a little harder now

hmmm

[dense tapir]

Well

[wispy ether]

Stable diffusion

Susion

[kind quartz]

have you tried that new sampler DPM++ 2M SDE Karras?

[wispy nest]

I don't have it yet, but I guess, since I'm using ++SDE karras as my main thing I should check it out

What even 2m means?

[dense tapir]

2m and 2s

[kind quartz]

it is twice faster @wispy nest

[wispy nest]

You mean this 2?

[dense tapir]

the S is ancestral

[kind quartz]

no, i mean new on totall bottom of karras

[wispy nest]

yea I don't have it yet

[wispy nest]

S isn't for me then, what about 2m?
What it means?

[dense tapir]

why not for you?

I way prefer euler_a to euler

[wispy nest]

I don't like them both tbh 😄

[wispy nest]

only if I'm doing art \ anime

[wispy nest]

hm , interesting

[dense tapir]

See the image above?

[kind quartz]

i think you need git pull for it probably

[dense tapir]

here is the same prompt in here is SDE Karras version of the same prompt.

[kind quartz]

image above is iam playing with neutral prompt 🙂

[wispy nest]

I know all of them can do good images, I just prefer ++SDE K for most of my prompts

I had really good images even on DDIM while testing, this doesn't mean I should use it for everything

[kind quartz]

you will have almost 2x speed.

[wispy nest]

for me it's also about the results, not the tool ;P

[dense tapir]

If I had a 4090 I would be doing 20 step ++SDE for all

[dense tapir]

LOL, don't let Sytan hear that. hehehehe

I will be so glad to finally move to it/s instead of s/it

[wispy nest]

Let me know if anyone will have matrix on new things
or at least article

[oak osprey]

DDPMSolver is probably the current state of the art for image samplers in my opinion

[wispy nest]

I like Euler, DPM++ 2M KARRAS, and DDIM the most myself, the DDIM is the least "detailed" ones of them in my opinion, but the rest is worse in general :P

[kind quartz]

with 1.4 i used to like HUEN or HEUN 😄

[dense tapir]

I can't even run the unipc one cause I don't have tensor cores

[wispy nest]

yea DDIM usually isn't as detailed, but for quick tests that's just too good

[wispy nest]

to see if prompt works \ if weight is fine \ to see what models understands and whatnot DDIM is just perfect

[kind quartz]

dont criticize much it is still playing with neutral prompt. So it is not much contrast but evening and soft light

[dense tapir]

I gave up on it as it was just ugly and I couldn't control it

[kind quartz]

point is to control it more. I will play with it occassionaly.

[kind quartz]

i cant make it rain!

[dense tapir]

See, this is why I would use this if I had a faster card. Above 11 steps this is 22 steps.

I would use this at 30-40 steps most often as it helps define straighter lines, and the lighting is greatly improved

[oak osprey]

https://github.com/bghira/SimpleTuner/blob/main/inference_ddpm.py

here's a simple script that'll work in your venv for A1111 if you want to try DDPM scheduler

[dense tapir]

40 steps

Even has caustics

[kind quartz]

yes and what sampler?
I can try dpm++ 2m SDE

[dense tapir]

The above is all from DPM++ SDE Karras

[kind quartz]

i show you caustic fail

[oak osprey]

oh, u like caustics?

[kind quartz]

with that new one you can have two times more passes therefore could be very helpful to you

i played alot with Luxcore 🙂

[oak osprey]

the best is an 8bit caustics experiment gone wrong

[dense tapir]

LOL

It isn't new, though, as I have had it for six months, roughly.

[oak osprey]

a burning tree encapsulated in a vacuum sphere

[dense tapir]

For the work I like to do I would use this at 40 steps, as I have said since it was introduced, but it is so slow it takes my card 2m to do one image.

[oak osprey]

wild

[oak osprey]

whoa

[dense tapir]

Yeah, I am not feeling this neutral prompt thing

Changed from perp to salt

[wispy nest]

much better than when they bring half dead animals :P

[oak osprey]

intergalactic horse travel through portals

ah okay, josh kirby brings a lot to the model 😄

giant loaf of bread?

[wispy nest]

brings a lot, but maybe they forgot cohesion at home? ;P

[oak osprey]

magic > cohesion

if i wanted cohesion i'd use negative prompts

😄

my negative is: malformed, disgusting

[wispy nest]

two more negatives than I use, I've gotten too lazy for them most of the time :P

[oak osprey]

my discord bot lets users set up ongoing negatives/positives, and the positives are added to the end of whatever prompt you throw in

so you don't have to ever type them again if you didn't want to

[wispy nest]

yeah, I got something similar, but I've noticed that a lot of prompt words people use doesn't always cause bad results. I mostly add negatives I 1. know work 99% of the time, and 2. I know won't remove some result possibilities. Like, "malformed" can give good result. And for me, it does that more often than not even :P

[oak osprey]

interesting, yeah, i removed artifact from mine because it prevented me from making good indiana jones gens

trying to get my coworker's cat to be Napoleon

omg

[smoky oak]

you know, that Articus 2.1 768 model gives me some hope for the future of 2.1

Its a big step in the right direction. Probably the first 2.1 model I have used thats not like... a huge pain in the ass/messy

[wild sorrel]

I updated a1111 and still don't have it btw

[oak osprey]

no, teddy! you're not qualified!!!

[oak osprey]

teddy did not listen

[dense tapir]

As lovely as that image is that is 100% not what I asked for. :/

[oak osprey]

❤️ ❤️ ❤️

[dense tapir]

cfg was at 7 too

trying cfg 18 now

damn, even less what I asked for

[oak osprey]

SO CUTE

[dense tapir]

I give

no neg

[oak osprey]

ahhhhhhhhhh

[dark flare]

I will now take full credit

plz send Bitcoin

thx

[smoky oak]

Hey, I am not making any money off it :p

just stating that its nowhere near as bad as most 2.1 models

[dark flare]

I think we should all pitch in to get you a new card... with the number of images you generate in a day, just imagine all the extra time you might have for...... well, making more of them I guess 😉

[dense tapir]

LOL

When I finally get a card I can go back to training as I love that more than the genning.

[wild sorrel]

[oak osprey]

you will now submit to steamboat willie all of your valuables

we're pirates on the open sea, it's always a pirate's life for me

[dark flare]

Go back to? So you had a bigger rig but it went away?

Speaking of training, what's your favorite way? I was trying to find a post someone made here in the last few days, about "good luck with dreambooth" and saying (I THINK?) Kohyasomething?

[dense tapir]

No, I used to suffer through colab, but it wasn't enough time given per day to do anything as deep as I needed plus free colab accounts can't use webui apps so I haven't been able to do anything even if I wanted to.

[dark flare]

ahh yes, zee colab

[dense tapir]

kohya_ss

[dark flare]

It's the bestliestness?

[dense tapir]

I made kohya_ss work on colab but not allowed now. No, kohya for 2.1 has a lot of internal bugs, and issues, and improper implementations that 1.5 will never see. Only thing we have though.

[dark flare]

Ah yes. I actually want to train on the "artius" 2.1 model, see how I can ruin a good thing 😆
So Kohya is the ONLY 2.1 trainer?

[dense tapir]

I think the only trainer period. There was another but it never left 1.5 and was superior, but no updates for a very long time.

you know, this is sus to me

it is trying to make a human female and this cat. prompt is the cat but behind the scenes.

[dark flare]

There's a long-haired human hiding in its ear

[dense tapir]

yeah, and the facial features I can see the human trying to get out. I am sure some furry out there would love it

[oak osprey]

i wouldn't recommend fine-tuning an existing 2.1 model, the loss is pretty high on those right out of the box

but fine-tuning from base 2.1 has really good results

[dense tapir]

it is the seed

[dark flare]

Yeah, we'll see how bad it gets. I hope it works, it's a gorgeous model

I thought "embedding" or whatever it's called these days was non-destructive?

[oak osprey]

i've had the best results so far with a mix of thoroughly captioned synthetic+photoreal data over a wide variety of terms. my model is trained on MJ 5.1 images plus national geographic animals and flickr portraits

[dense tapir]

I changed the seed and this is cute

[dark flare]

Anyway, thanks to this, I was able to find the original post that I was trying to find 😛
#🤝｜tech-support message

[dark flare]

Yeah I'm only interested in narcissistic training 😉

And/or maybe training my wife+son's faces, see if I can do a nice family portrait or something

[dense tapir]

Been having some folks run training tests on AMD 7900XT. They ran into a snag, but disappeared on me.

good luck, as my Nephew turned 18 in Feb and was attempting him but it only got so close. A real Dreambooth would have been spot on.

[dark flare]

ah, so Dreambooth is superior but not beyond 1.5?

[dense tapir]

I don't know now as I stopped working with DB due to 2.1 requiring more than colab could do (the free one). It isn't so much superior it is that if the model has nothing similar it can fully insert while lora/lycoris has to have something to latch onto.

Well, wth?

[dark flare]

OK I should force myself to bed now 😛

[dense tapir]

night

I upped from 12 to 18 steps

[sterile temple]

And I have to force myself to get up now and get ready for work 😞

[oak osprey]

can someone fix their faces?

[neon crystal]

How would you rate it against Realism Engine? I’m interested to know if it follows prompts better.

[smoky oak]

I have not tried realism engine

[neon crystal]

Ah ok I kinda never felt the need to use anything else till now. Excited to try pseudo s latest for sure. I like the “temperature” of the samples that are shared here which feels bit more on the brighter side

[oak osprey]

khoya isn't the only 2.1 trainer. i've made my model using my bghira/SimpleTuner on github

[smoky oak]

I am not sure if realism engine is 1.5 or 2.1

[oak osprey]

recently updated its layer freeze strategy to allow freezing it at some percent through training, you can freeze just the first or just the last, or, between two layers. added the terminal SNR to it, so it patches the betas schedule and corrects the noise it's trained on. allows you to pertubate the offset noise if you choose to apply that during training, to offset the instability it introduces when using static values.

i'll be updating it soon to have two learning rates, one for the unet and one for the text encoder, as well as more ways to label and caption data, group concepts into folders and add that as a keyword to the caption for each image in the folder, etc

[smoky oak]

ah, it is 2.1

yeah, artius has to be better

[neon crystal]

Experiments with mixing LORAs

[smoky oak]

realism engine looks terrible, unless I am on the wrong model

[oak osprey]

it comes with a few helper scripts to test the models you create and to caption datasets, as well as a tool that downloads images from a csv file from Kaggle datasets. there's a movie processing script that'll let you pull a 4k movie apart into frames and pull faces from it, discarding duplicate-looking frames.

[oak osprey]

have you ever tried controlnet tiles at 1.0 strength? it's good times

there's so much noise in those images. it will have a lot of fun

[neon crystal]

Works like a charm for my needs. Working class Realism needs a bit of prompt hackery you know lighting hue temperature etc but yeah I’m a happy user with no complaints for art stuff especially with following the prompt and mixing styles

[smoky oak]

if high level realism is what you are after, I would hard recommend 1.5 instead. It has way more realism potential as of the time being, IMO

[neon crystal]

Agree. No question there

[oak osprey]

oh, negative embeds work great too

[smoky oak]

ah, alright, if thats the case then by all means haha

I would say that artius looks way better, even just off the demo images I saw

screw it, I can just test them together lol

[oak osprey]

stay toon'd for my new yootuube tutotirlas

i wouldnt trust any testing that's done through A1111 unless your goal is to simply compare how models behave in that tool only

[smoky oak]

damn it lmao

[smoky oak]

I was just about to test realism engine, and civit is down lmao

[neon crystal]

Coolio, I’ll give artius a spin as well and see

[smoky oak]

damn lmao

I had just clicked download lol

[scenic fjord]

Gangsta presidents lmao

[oak osprey]

Artius is great

these are classics from my best-of gallery

and it really does people quite nicely, like rick moranis

[kind quartz]

weird...
here i got mentioned it. It updated on 28th. may.
sd_samplers_kdiffusion.py

[oak osprey]

finally perfect photorealism

[wispy nest]

heh, didn't expect it to "work" :P

HOW do you guys s[ecify negative prompts and styles??

???

depends on what you use to make the images

#1101178378452750447

then I can't help you as I use auto1111's webui that can be run on your computer, if it's powerful enough :P

thx

[static tusk]

type your prompt, then click to the right of the prompt box and more options are available like Styles, Negatives etc, with their own box in the chat section

[wispy nest]

thx broseph

[wispy nest]

lol love it

[oak osprey]

love how there's a kid in the background like, wrestling a bear cub

what negatives do i use to fix this?

it's so close to being good, or even great

[wispy nest]

there are many ways, but few without many details :P

[oak osprey]

it's just supposed to be a family portrait

[static tusk]

negatives wont fix conjoined bodies

[oak osprey]

how do you know lol

[static tusk]

maybe reduce the frequency a little

Ive used them for 6 months

[oak osprey]

i found another seed where they aren't but it added like grandma James May

god i love it, it's like a bad photoshop

it almost, ALMOST, seems like, SEEMS like.. because i have no children in my training data, it instead, tries to stuff an adult face on there

[wispy nest]

were my first result on having three people in the same image :P

[wispy nest]

this one made me laugh! :P

sadly, it's not yet good enough to fix the typical woes :P
Like fingers, face, and broken spines :P

[sterile temple]

the sun sets on another day

[timid acorn]

25 murals on the walls of Le Havre, France

[rigid robin]

Anyone can help me with controlnet sideviews? It's not following the pose I want.

Settings

[woven dawn]

Try adding the word "profile" to your positive prompt. Its the proper word for 'sideview'. Profile, facing away, looking at viewer

[oak osprey]

can you not be gross?

[wild sorrel]

not gross

[earnest solar]

well, we're back on... I think?

[dense tapir]

yes

[oak osprey]

huh?

[earnest solar]

I've been having issues with CUDA all day

[dense tapir]

did you update?

[oak osprey]

what kind

[earnest solar]

yeah

[earnest solar]

Look in #🤝｜tech-support it's all there

[oak osprey]

i just got the tiling pipeline working with SD 2.1 😄

[oak osprey]

ouch, that looks frustrating

[dense tapir]

Well, this is why I don't update for months at a time as this stuff is just a house of cards.

[earnest solar]

I think mine updated automatically?

[oak osprey]

mightve screwed some math up

[dense tapir]

Well, it only does that if you left in the command to do so. I updated Linux yesterday to spend the rest of the day getting everything to work except greenwithenvy which apparently no longer is updated or works. Means bye bye Linux for me.

greenwithenvy is like Afterburner in Windows. For me I need the undervolt and fan curve.

[latent atlas]

https://www.instagram.com/p/Cs85keMoBxH/?igshid=MmJiY2I4NDBkZg==

veiled.vision.studio

Stargaze Commander 🌑

Instagram

[oak osprey]

fixed it!!

huzzah

[wild sorrel]

went too high cfg on low prompt

got no details aside from actual prompt lol

[full reef]

I think it was one of the most beautiful

[oak osprey]

this thing is hard to wrangle but really neat results

[dark flare]

Hey folks 😎

Do you have a favorite set of regularisation images? (For "man" class)

Or do you generate your own?

I was really excited about the FFHQ dataset until I realized it includes all kinds of humans 😉

I found this:
https://github.com/tobecwb/stable-diffusion-Regularization-Images/tree/main/sd2.1/man
but somehow I'm not smart enough to know how to download it.

Thoughts?

[dense tapir]

Yeah, old news but there are a shit ton of those from Nvidia

Far more than I have space for

[formal surge]

(╯°□°)╯︵ ┻━┻

[formal surge]

hey

[dark flare]

Googling "nvidia man regularisation images" wasn't obvious enough 😅

[dense tapir]

LOL

[dark flare]

As in, the results .... I can't find a big shiny obvious "DOWNLOAD" button

[dense tapir]

each folder has 1k, if I remember right, so you right click and download the folder

[dark flare]

The FFHQ? Or Nvidia

[dense tapir]

FFHQ is Nvidia

[dark flare]

ooooooooooooooo

OK, that I did not know.

But unfortunately it includes women and children.

I guess I have the time to delete all non-man images but .. . . .. . . . .. . . . . ..

[dark flare]

He disapproves of my laziness, clearly

[dense tapir]

LOL

[oak osprey]

go to Kaggle datasets and find a high resolution set of humans

you can use BLIP2 to caption it, and that's typically good enough for fine-tuning. worked really well for me.

[dark flare]

I guess in the amount of time it's gonna take me to find a set of MAN ONLY images, I could just render a thousand of my own 😆

[oak osprey]

the BLIP2 will caption them and you can just delete any with 'woman' etc

[dense tapir]

Even on a T4 that was over an hour to do

[oak osprey]

would recommend not generating regularization data the way the tutorials usually request you to do

[dense tapir]

4090 could do it in under 5m

[oak osprey]

when you pull them directly out of the model, it just keeps the garbage the model knows

[dense tapir]

no more than 10m

[oak osprey]

BLIP2-captioned human photos from Kaggle ended up actually improving the outputs without catastrophic loss/forgetting/overfitting on trash

[dense tapir]

The above image is with some effects I was working on last year but never worked out as I didn't have enough time on colab to really dive in.

[oak osprey]

keeps complaining about the cost of cloud compute and then goes on about spending $1500+ on a GPU to do local image gens. gotta love it

cost $40 to train my model

[boreal falcon]

@oak ospreyhypocrites in a nutshell

[dark flare]

I took a screenshot of you arguing with someone the other day 😆
I was gonna say "can't we all just get along?"

[boreal falcon]

just like fools that complain about subscription model of softwares but proceed to pay on a monthly basis for Netflix, Amazon Prime and Disney+

[dark flare]

Adobe Creative Cloud was the GREATEST IDEA EVER
I got the latest greatest stuff, and after 10 years, still paid less than if I had bought CS7/CS8/CS9/CS10/etc/etc/etc
I have no idea why anyone would complain about it

[oak osprey]

it's the resulting mindset of someone who never grew up

[boreal falcon]

true, basically i can afford several expensive softwares because i dont have to pay full price at once

[dark flare]

YES VERY MUCH

[boreal falcon]

and in best case i even get that money back if i earn on top of that

but even before that

[dark flare]

EXACTLY
I used it to MAKE MONEY, pay my rent, etc etc ..... it's a no-brainer

[boreal falcon]

plus you can make a deal with some of them, i did with Adobe on CC

[oak osprey]

yeah davinci resolve cost me like $500

[boreal falcon]

i forgot, do you get updates forever?

like major ones

[oak osprey]

yup

[dark flare]

OK, so, um, I'm sorry, could you help me a bit further? 😅

Where should I be of the downloadings....
https://www.kaggle.com/datasets

[boreal falcon]

well thats really good and affordable

[oak osprey]

on the left side it says dataset type. you can filter it by png/jpg

[boreal falcon]

im not even earning money from this all (yet)

[dark flare]

FL Studio is one program that is INSANE, I bought it once like 20 years ago and I'm still getting lifetime free updates ...... how they're still in biz, I have nooo idea

[boreal falcon]

but the future might be different 😄

[oak osprey]

Jody, there's a dataset called "Celeb-A"

[boreal falcon]

FL studio, sounds familiar

[oak osprey]

fruity loops

[boreal falcon]

aaah someone told me about that one

[oak osprey]

have not heard that name in forever

my old roommate used to make stuff on Ableton

[dark flare]

https://www.kaggle.com/datasets/zuozhaorui/celeba ? So many wimins.

So I run them through BLIP2? Is this a Python thingy?
*clearly didn't google it before asking

[oak osprey]

another roommate used Pro Tools and some expensive-ass sound board he got from a defunct recording studio auction

[dark flare]

OOP! WOMEN WITH HEADPHONES!
*gets started again

[dense tapir]

hehehehe

[oak osprey]

blip2 is a way to use the stable diffusion model's text encoder in reverse to caption images rather than, turn captions into images. capiche?

[dark flare]

Would it ever incorrectly classify a man or a woman? noooo 😱

[crimson holly]

New sota t2i?
https://raphael-painter.github.io/

[oak osprey]

one or two wouldnt matter

if you really do care then you can go through and delete/fix the issue ones. that's much less work than captioning them all by hand.

[dark flare]

That's a lovely environment thar

[dense tapir]

🙂

[dark flare]

Believe it or not, I actually googled before asking 😉

Is this a thing I can do in auto1111? Or is it still a feature request...

On https://huggingface.co/docs/transformers/main/model_doc/blip-2 I see a bunch of code that ... uhrr

Is there a repo somewhar?

😇

[oak osprey]

mixture diffusion is ❤️‍🔥

[dark flare]

Just gotta match them horizon levels 😅

[oak osprey]

@dark flare https://github.com/bghira/SimpleTuner interrogate.py here is my script i use for blip

[dark flare]

Thank you!

I assume it can do batches...

[oak osprey]

plop into gpt that whole script. it will explain for you

[dark flare]

hehe nice.

I just ran interrogate.py in a CMD window and it spat out errors, too bad there's no requirements.txt

[oak osprey]

expert mode

[dark flare]

hehehe indeed!

[oak osprey]

load the venv for a1111, should work fine

[dark flare]

Meanwhile I've already generated 1600 regularization images 😆

[dark flare]

Only one error this time! An improvement

pip install clip_interrogator ftw

[dark flare]

Just what I need, another one 😆

[oak osprey]

[oak osprey]

@dark flare i've used that script to caption frame grabs from The Hobbit and it knew hobbits vs wizards etc

i think the reason salesforce made blip2 is to get more descriptive captions for their alt tags in their ecommerce stuff

not to necessarily, caption training data

[dark flare]

Hang on, I'm perilously close to finishing ... err, getting this thing to work

OK so I've created a folder for
process_directory('/models/training/datasets/man')
but I don't actually know where to put it....

I keep getting
The system cannot find the path specified: '/models/training/datasets/man'

Should "models" be in the SimpleTuner-main folder? or the auto1111 venv??

[dark flare]

OK I think I can get it to work but I officially don't know what the script considers my os.path to be

I also can't specify an "absolute" folder, like F:\StableDiffusion\blip2\SimpleTuner-main\models

So it's relative to what? gurgg

[oak osprey]

the cwd

i have never really used windows so i can't quite help with that

but i'm sure ChatGPT could

[dark flare]

oooooooo

amaze

[oak osprey]

😄

if you want to make the changes and open a pull request i can review and include it so that it works on windows out of the box for others

[wispy ether]

[dark flare]

By the eyepatch of Odin, it's workiiiiinnnggg

[wispy ether]

Odin doesnt have a eyepatch

[dark flare]

NOW .............. okay ............... these FFHQ images are lovely, but the faces are cropped so tightly.

Whereas the regularization images I generated have all kinds of full-body images.

Wouldn't that be more accurate?

[dark flare]

Well he does in the Marvel universe anyway 😆

[wispy ether]

No

There he got a scar

[dark flare]

Google images might disagree with you

[wispy ether]

What

Jk

[dark flare]

hehe 😄

[dark flare]

Quite awesome, I gotta say!

What do you think an "arafed" and "araffe" are?

[oak osprey]

a great mystery

i have asked that. and the effect it has on images is also weird

[dark flare]

I might use a magic renamer program to fixem

[oak osprey]

get gpt to make you a python renamer. lmao

[dark flare]

One thing I'm wondering though ... I have 1000 images in a folder, why did it only do 949? Actually, looking at the log, it seems it processed all of them ... but some didn't save? I are confuse

[oak osprey]

no idea

[dark flare]

nah, I already have a good one

It's a Windows prog though so I can't tell you about it 😛

[dark flare]

Hey, who's posting images at a time like this?!? 😆

Really cool!

[dark flare]

I ran the script twice. Believe it or not, the results were identical -- over 1000 images processed, no apparent seed or randomization 😛

[dense tapir]

I had someone with a 7900XT do some lora training testing for me and I am amazed. "You should be proud of your card, btw because it has no optimizations, has no real ROCm support that will bring you tensor core usage (w/e AMD calls it) AND through all of that your 5.98-6.08it/s compared to this T4 on colab 1.06it/s".

[dark flare]

You wrote that to them, or they wrote that to you? 😅

[dense tapir]

I wrote that to them. I am impressed

If AMD actually goes full on with 5.6.0 rocm I can see that card getting about 12-16it/s

for training

XTX is about 30-35

[dark flare]

WELL DONE

From a company that ... well, I thought was dead 10 years ago

I'm running a Ryzen msyelf 😛

[dense tapir]

I mean, how is it even able to do that with no support at all amazes me the most.

Been a Ryzen since 2016

I used to flip back and forth since the AMD Athlon 700 that was a card not a chip. Still that system too, lol

Actually, before it a K6-2/500

then Core2duo to 2016 and Ryzen. For what I do I don't see myself going back to Intel

Did you see their new chips are going to be 350-400 watts? Insane

[dark flare]

ahhh yessss, when CPUs were cards... trendy trendy

[dark flare]

Good thing everyone has solar panels nowadays?!?

[dense tapir]

Slot based and the heat sink was the big pita for it

most don't and none in my city due to being the city of trees and no incentives

Can't even see my house from Google Earth as I live under a canopy of trees. 75-150 foot tall pines, oaks, and I forget the other. Magnolias here too but not that big.

[dark flare]

Washington state?

[dense tapir]

No, TN

[dark flare]

ha, well I was a little off

[dense tapir]

hehehe

[oak osprey]

that explains a lot

[dark flare]

heheh here we go

[dense tapir]

just a little

about to storm here I think 😦

[dark flare]

That guy means bidness

WELL SHARD. Out of 1000 images, only 336 were identified as men 😛 guess I'll download another couple batches

I wish the script was a little LESS chatty .... just saying "man" and "woman" would be nice, idenfitying the main subject only and not all the deets

[dark flare]

WHO SAYS PINK AIN'T MANLY?

I'LL KILLEM

[dense tapir]

HAHAHA

[oak osprey]

chatgpt can write a function to add to the script that only keeps a list of words if they're in the caption

[dark flare]

Well like "there_are_two_people_on_a_boat_with_a_lake_in_the_background.png" ..... the MAIN subject is a woman, just call it "woman.png" 😄

[dense tapir]

Here ya go.

[oak osprey]

you can't have a million files with the same name

[dark flare]

That's why we invented numbers!

[oak osprey]

you are only supposed to be captioning these so you can delete photos that aren't men

[dark flare]

Come to think of it, THAT is why there are only 949 end files! It couldn't have dupes

WHAT A SILLY SCRIPT

If I knew anything I might fork this script and create my first git thingy

[oak osprey]

i used it to caption 30,000 files

[dark flare]

And did you get it to auto-delete the ones you didn't want?

[oak osprey]

nope, i just don't keep dupes because they're harmful to training

[dark flare]

Well the image isn't a dupe, only the description

[dense tapir]

Okay, this is a bit meme worthy "Still waiting for the raid party to form".

[dark flare]

hehehe, excellent

[oak osprey]

the description means it hits the same space of the model

[dark flare]

ooooooooo

Someone knows things about stuff

[oak osprey]

i'm also lazy

[dark flare]

VERY WELL, I ACCEPT

[oak osprey]

it could flip the image horizontally

but then, what would you caption it? re-caption it?

that's so slow lmao

[dark flare]

In my script it'd be man00001 and man00002 and...

[oak osprey]

it doesn't quite matter because you're going to use these as class data

and you give a single keyword as an argument for that case

the captions are literally just to help you filter your data out here

[dark flare]

Yeah but my precious missing 51 images 😢

[oak osprey]

are they dudes?

[dark flare]

I will never know 😄

[oak osprey]

we'll never know

GO FIND MORE DATASETS

how many class images do you need? firstly, is your model already very photoreal capable?

[dark flare]

hehe, I bookmarked that celeba one

I have no idea. I am 9 minutes into a tutorial video and I started it five and a half hours ago 😵‍💫

[dark flare]

Why can't there just be a quality regularization repository lolololz

[oak osprey]

i want to bring this kind of celeb understanding back from baseline 2.1 into my fine-tuned model. that's danny devito if he were in MASH, a show from the 80s

[dark flare]

Because I wouldn't have learned as much, that's why

[dark flare]

Now I have the theme song in my head, thanks

[oak osprey]

that sounds like you need professional help

that song is like a dark emotional abyss

[dark flare]

Well I heard it so many gosh-darn times

[oak osprey]

my mom and i were brainstorming of movies from the 40s/50s/60s that'd add good aesthetics to 2.1

Singin' in the Rain is in pretty damn high quality 4K

i want to find "Hello, Dolly" though

[oak osprey]

or 1969's Hair

[dark flare]

I consider myself a movie nerd but it seems I've seen very few pre-1970's movies 🤔

[dense tapir]

My fav are Westerns and Film Noir

[oak osprey]

well there's My Fair Lady, or, The Long, Long, Long, Trailer, or.. Ben Hur...

there's Ocean's 11

Gone in 60 Seconds

[dark flare]

ah yes, the Heston films, and Spartacus ... and Lawrence

[dense tapir]

Yep

[oak osprey]

Heston's films can stay in the past imo, The Omega Man is a cinematic tragedy

[dark flare]

heheh

[dense tapir]

I still will watch those

I own a bunch of those on Blu-Ray but then the wife passed 10y ago and I just stopped watching films for the most part.

[oak osprey]

mixture diffusion is so tricky

[dense tapir]

I did just watch all but the last ipman last month. Loved the first one, second was okay, but the 3rd fell a bit flat for me. Still was okay though.

Yes, I love a good martial Arts flick

Especially period pieces

In 2023 I find more enjoyment from Indie than Hollywood

[dark flare]

First one blew me away.

[dark flare]

That and "Drunken Master 2" which I watched a few times a year (until I purged all my stuff and moved to a straw hut in Nicaragua)

[dense tapir]

Classic Chan

[dark flare]

Now this is inspiring, I always love a good galactic skyscape

Or whatever they're called....

....I should go to bed

[dense tapir]

A better one

[dark flare]

I had a WILD FANTASY of hitting the "GO" button on this TRAINING thing before bed though 😆

[dark flare]

I dunno, I like the first one more for some reason

[dense tapir]

Cause it showed more skin

[dark flare]

I guess so!

Just a dirty old man in the end

[dense tapir]

I had to get rid of one of my extensions

I use it for wild cards but when I upgraded each time I start automatic1111 it goes checking stuff online.

throws out a ton of spam so rip

[dark flare]

Like the latest sports scores and your bank balance and stuff? 😉

[dense tapir]

RIP

[dark flare]

😦

[dense tapir]

That was annoying af

Not sure what I hit

[dark flare]

One time firing up a1111, it told me something along the lines of "if it's taking a really long time to start up, it's because we have to uninstall and reinstall stuff"

[dark flare]

But as far as I knew, I didn't have any extensions 😛

[dense tapir]

I dunno if I were not going to do video with this I might actually buy an AMD

as it is I need something cause this 1060 just can't handle this

[dark flare]

This is the 6GB card?

[dense tapir]

seconds per it instead of its per second

aye

[dark flare]

ha ... argh

[dense tapir]

some of these are 5-8s/it

so 40 steps is almost 3 mins

I figure the 5090 will be 2500ish and I will just save for it instead of a 4090

[dense tapir]

world might implode by then so best to wait it out

[dark flare]

The most I would ever spend on a graphics card is .. .. . .. . .. $700 ??? even that feels insane, but I did it last year
Normally $300ish

[dense tapir]

Now you know my quandry as 229 was max then 279 then 349 now 2k+? Come on

[cyan snow]

if y'all want a unit just for AI, you should probably go with a TPU.

[dense tapir]

I don't have a second slot now

besides I am not sure this training stuff for SD will do TPUs

Heck, I can barely make the training work on AMD

[cyan snow]

TPU can do anything with AI, it's very efficiant

[dense tapir]

There once was a guy in here who said TPU is a big PITA as it has to keep sync, etc...

[dark flare]

Well, high five, I just hit the TRAIN button .......... now to go to bed, and wake up to an error 😉

Is 10,000 steps in kohya_ss similar to 10,000 in DreamBooth?

'cause I found, when I went past 2000, it looked like crap.

But here I just did what a tutorial told me to do 😛

[dense tapir]

which TPU will train SD? Which make and model?

[dark flare]

Always gotta have that little guy at the bottom for a sense of scale...

[dense tapir]

I know diddly squat about TPU beyond the rudimentary stuff.

[dark flare]

I never even heard about TPU until now 😉

Night folks!

[dense tapir]

Night

You can use TPU on Colab but it never worked for training, or inferering SD

[cyan snow]

before i got my new GPU, that's what i used, it worked pretty well, but nothing is as good as running it locally.

[dense tapir]

I agree, locally only but Colab TPU just errored out for stable diffusion

[cyan snow]

idk, i think google's colab uses TPUv4. if i recall correctly

but if what you want is pure power, than id say A100 is by far the best

[dense tapir]

yeah, big money

[cyan snow]

that's for people that run servers that sell AI stuff

[dense tapir]

yes

[dense tapir]

H100 seems to be Jensen's big boy now

Hopper

[brave saffron]

Instagram avatar, monkey with money and a diamond chain around his neck GT

[cyan snow]

bro wtf

[oak osprey]

i use one to make free models

[cyan snow]

well, you are a special case

[oak osprey]

https://giphy.com/gifs/ace-ventura-put-me-in-coach-im-ready-to-go-just-gimme-a-chance-uVpjHrGt9Ihnx9hRJi

[oak osprey]

Singin' in the Rain 4k is 48GB

that's so much training data 😄

[dense tapir]

@cyan snowhttps://www.datanami.com/2023/04/05/google-claims-its-tpu-v4-outperforms-nvidia-a100/

I think this will eventually move to ASIC

[cyan snow]

https://github.com/Stability-AI/StableStudio

Yo, is this official?

I mean, it is a WebUI made by stability AI, so should it be the best UI for stablediffusion?

[dense tapir]

no idea

doesn't seem to do embeddings, loras, extensions, lycoris, etc...

[cyan snow]

It has a plugin that adds all A1111 features to it, keeping the architecture of this UI

[dense tapir]

damn, a plugin :/

I'll stick with automatic

[cyan snow]

Making A1111 a backend, and Stable studio an interface

[dense tapir]

Yep, and why I said I will stick with auto

is theirs webui based?

[cyan snow]

No, it seems to be entirely different

[dense tapir]

Well, that is a positive

I don't have to run a1111 and it too, right?

[cyan snow]

I don't like A1111 because of what he did, making Stability infinitely better

[cyan snow]

No, I'm not setting it up

[dense tapir]

huh?

[cyan snow]

What? He stole code, and did malicious things. Well, according to this community

[dense tapir]

I don't like Node.js but compared to gradio I am unsure which I hate more.

probably gradio

Not sure what Yarn is though, but I don't like projects that go political on the first page so they can somehow virtue signal. Keep politics out, so I will refrain.

[cyan snow]

A1111 also changed his UI's architecture to be compatible with leaked models.

It doesn't feel morally right to use his WebUI

[wispy nest]

trying some cyborg stuff

[wispy nest]

then dont lol

nobody forces you to

just make a fork and remove the leaked code to make it ethical again

the beauty of open source

i legit dont care about things like that

[wispy nest]

there is probably over 100 devs working on the A1111 webui

i wouldn't throw out their work either just for one dude

inb4 its opensource

fork it and make your own version

didnt know any of that, but glad i stopped using auto and stable months ago now lol

i'll use whatever has the best features

miss me with your idiology war, its just code

ideology

dudes made some edgy mod 5 years ago doesn't make his code bad

coders... stealing code?

no way

[oak osprey]

[wispy nest]

FYI i have autism and i didn't speak until i was older because of it, thanks for your correction tho.

[oak osprey]

you don't get any sympathy for me because you have no empathy for others

[wispy nest]

you were the one who was all smug, miss me with your idelogy war, but then you get defensive lol

yep

i guess now you are free to attack me personally

yeah but he can criticize shit like that but you cant criticize him he has autism bro

you can do racism if you autism

yep

so he is racist because he made an edgy mod 5 years ago ?

thats your point?

you got any proof of that

[acoustic vector]

If i wanted to generate specific pokémon, how would it work? For example, if i wanted to make art with the pokémon named "Dondozo" (image below), how would it work?

[wispy nest]

so the mod is the base for your entire argument that he is racist ?

nah i'm just trying to think critically here because i dont even care about his personality

and i dont know jack shit about his personal opinions and personal life

ok asking questions makes me a bad person ?

[acoustic vector]

is this twitter or somethin

[wispy nest]

i wont jump on a bandwagon or just start talking shit about someone i dont know

[oak osprey]

@wispy nest https://rationalwiki.org/wiki/Just_asking_questions

that's your current tactic? really?

Just asking questions?

try again

[wispy nest]

bro whats your problem

the guy made a racist as hell mod, nuff said

[dense tapir]

a mod to what?

[acoustic vector]

what mod

[wispy nest]

?????

what are you even talking about

take ur meds

you have some issues man

maybe take a walk and get some fresh air

i said i have autism and that because of that my language skills are bad

and you are using that as a way to attack me for your idiological bullshit

literally stop talking to me, go get a grip

you act like a child

winning arguments on the internet lol

you dont even know me

you just make these assumptions

no the context is that people say edgy shit all the time and they are not racist

if you ever went on the internet in the 90s you'd know

[dense tapir]

Those were the good old days. Geocities 4 life. 🙂

As we used call it "geoshities"/

[wispy nest]

yeah calling racism edgy is fucked up

[dense tapir]

I think the Internet turned to crap the day Netscape died and Google began.

[wispy nest]

but you know people here think 'woke' means you cant be racist

you're just looking for a scapegoat to let out some built up stress and anger, i'm not some kind of stereotype you can project your shit on sorry

we're definitely hella ot for images

who ?

blocking people ?

lmao

why are you so agitated

lmao

yeah and i have autism and i was sexually abused as a kid, lets cry about it together

you're in a frenzy

keep going

[boreal falcon]

whats going on

[wispy nest]

idk dudes losing his mind because i didn't jump on some kind of witch hunt without any sort of evidence

as if i'd join even with evidence

[oak osprey]

oh just kids excusing racism as "being edgy" and other gaslighting

[oak osprey]

i didn't reach out to you and beg you to "join our cause" you just disagreed and got tangled up in something you're not even prepared to discuss

[wispy nest]

i think that you should stop using his code then and move on with your life

[boreal falcon]

what do you use now? 😄

[oak osprey]

edgelords don't like nuance in their discussion

[oak osprey]

i have never used a1111's code in my life, i don't know why you keep saying that. this is the kind of self-introduction into an argument that you did earlier. are you actually wanting to continue discussing it?

[wispy nest]

if you dont use it, whats your problem

[wispy nest]

i havent used stable in months, i just check in her for a new stable model and images.

[oak osprey]

i have already explained it, and you said my problem isn't real

[wispy nest]

you want others to not use it too ?

you cannot control what others do

[oak osprey]

i don't give a shit whether others use it or not, that cat is out of the bag

[wispy nest]

well then there is no problem

[oak osprey]

there's better platforms to invest development time in, and it's worth letting people know what a shitty person the project lead is

i don't know why you keep wanting to inject your defense of him into there

[wispy nest]

good, let people know that he made some racist mod 5 years ago, maybe dig deeper into his childhood and find out if he used a racial slur when he was 5 😂

[oak osprey]

do you two have a romantic relationship or something?

the mod is still up on his github, dude

[wispy nest]

racists always excuse other racists

[oak osprey]

he didn't even delete it

[wispy nest]

and if you dont do racism youre a woke mind virus or whatever

[oak osprey]

he actually updates the mod for each rimworld version

[wispy nest]

how does that make you feel ?

angry?

[boreal falcon]

same 😄

i was never primarily a SD user tho

[oak osprey]

the other guy said he didn't want to use some architectural scaffolding called yarn that he doesn't even know what it does, because, "they went political on their front page to virtue-signal"

[wispy nest]

me either. im in like 5 discords i dont even use any more lol

[boreal falcon]

used MJ, in between SD and now full Firefly guy

but i do manual work as primary

generative AI is just one of many tools i have

[wispy nest]

is firefly the adobe thing?

[boreal falcon]

yes

[oak osprey]

yes

[wispy nest]

can you do a new image with it? i thought it was lik eonly to modify an image

[oak osprey]

and the ControlNet guy probably fell down some stairs rushing to get it stolen implemented for controlnet extension for sd-webui

[steel yarrow]

I got bad news for you firefly bros: https://www.teamblind.com/post/Racism-at-Adobe-Uc27Jg7u

it's contaminated, better go back to using python to generate

[boreal falcon]

nice try

now let me start Photoshop

[oak osprey]

never left, lmao

[boreal falcon]

[wispy nest]

lets just acknowledge that almost everyone here eats meat, just for the pleasure, not really out of nescessity

[oak osprey]

https://github.com/bghira/SimpleTuner/blob/main/inference_ddpm.py

[boreal falcon]

you can

but

[wispy nest]

so the moral arguments go out of the window in an instant

[oak osprey]

free your mind, neo

[boreal falcon]

Firefly web is better suited for very stylized stuff

[wispy nest]

im vegan bro

[boreal falcon]

based, im vegetarian

[wispy nest]

i'll have to give that a look later

plants also feel pain, fyi

i was trying dalle today it seems like it regressed

[oak osprey]

brahman hindu here, no meat, sorry

we're all morally superior to you

[wispy nest]

plants have no pain receptor or a brain dude

magic man youre just like a low level maga troll

[oak osprey]

so true

[wispy nest]

get some new material, not that you didnt get some mileage out of the act here

[oak osprey]

probably subscribes to Jordan B. Peterson

canada's national shame

[wispy nest]

the amount of stereotypes in your mind are fascinating

[boreal falcon]

tbh Peterson doesnt sound that bad

there are far worse ones

[wispy nest]

maybe associate me with some andrew tate bullshit too

[boreal falcon]

he is at least intellectual

[wispy nest]

manosphere lmao

[oak osprey]

he's like a gateway drug, homie

https://www.youtube.com/watch?v=hSNWkRw53Jo

[boreal falcon]

Andrew Tate is a whole other story

[oak osprey]

here's a 3 hour deep dive for you

[boreal falcon]

and his followers

[oak osprey]

i will congratulate you if you can even get through 1/4 of this video, i think it took me like 2 weeks of watching just to get through it because the frustration is so dense

JBP can't actually answer a single question. ever. i don't think he's done it. legend has it, he's still working on an answer for the first question he was ever asked.

[wispy nest]

pain is an abstract concept, its a response to loss of structural integrity. and plants have that response, plants also communicate with each other and intelligently respond to changes in their environment. in a sense plants do feel pain and they do not want to die, they have survival mechanism. thats besides the point anywas, i'm just cringing at your "moral superiority"

nothing that exists has moral superiority

[oak osprey]

gotta love how he loudly talks over the interviewer to keep them from trying to reign him in

[wispy nest]

lol. i dont give a shit what you eat dude

[oak osprey]

you're on step 1 to becoming a Proud Boy, congrats

[wispy nest]

but plants dont have pain receptors or a brain, end of discussion

keep the stereotypes going. for someone who claims to be against stereotypes you have a lot of them on your mind

[wispy nest]

thats a very close minded view about life

[oak osprey]

saying plants communicating with each other is like saying the planets communicate with each other because they're spinning together.

[wispy nest]

no its just factual

planets are communicating with each other through gravitational waves