🏞｜general-with-images

[nocturne oak]

No.

[thin compass]

Cats would totally do this if they could.
"I made you a sammich! Oh by the way bwaauulwgh"
(This was unintentional, the prompt was "a cat making a sandwich out of bread, ham, mustard, cat food, and cheese", but... cat's gotta cat.)

[steep sleet]

Searge img2img

[steel yarrow]

Yes it is.

[celest sigil]

☝🏻 Flux and SDXL were discussing how SD3 had far too much to drink...

[steep sleet]

SDXL img2img

[white pivot]

Quick question, do you like it with the butterfly or without the butterfly ?

[steep sleet]

Without

[languid pebble]

Silence of the lambs ...

[clever oar]

first image its flux second i do with sd 3

[hushed patio]

Quick question!

[clever oar]

Mojo

[hushed patio]

Is there any lora or lycoris for yoruichi pose

[clever oar]

what is it richi pose

[hushed patio]

Search it, it's pretty cool

It's like cat stretching

[clever oar]

she have many poses

if you need pose only...

you can copy it from ipadapter

[hushed patio]

I'm downloading it now after I read all of contents

Thank you!

[languid pebble]

Hi dicordos

[languid pebble]

Investing in crime ^^

[steep sleet]

A nutmeg grater, silk slippers, and a bodkin!

[nocturne oak]

I'd invest in a scam-free & spam-free Discord experience.

[steel yarrow]

what's a go to proper photo grain simulator ? anyone use anything like that that's solid ?

[sturdy pewter]

published my first public model - contains huge amount of research info : https://civitai.com/models/731347

[unique condor]

20 seconds in Photoshop, Lightroom or Krita

[valid urchin]

anyone know, what ai-model this artist use ?
https://www.instagram.com/jasonscarecrowart/

the lightning was perfect and really good realistically

[valid urchin]

which model to generate painting like this?

[thorn talon]

portrait for business man

[ivory glacier]

hello everyone! would it be possible for anyone to point me to a good checkpoint to create translucent figures, humans and/or animals please ?!

[royal monolith]

flux dev, base model, does those just fine with the right prompt

[ivory glacier]

downloadable from Civit Ai ?

[royal monolith]

think it's still only downloadable from the black forest labs github and huggingface spaces.

[ivory glacier]

alright. Thanks for your time!

[clever oar]

mini nuclear bomb explodes in lego toy city

[crisp stream]

It´s Stable Cascade with a fine-tuned checkpoint for Stage C (Invictus Redmond v1.1) and a fine-tuned model for Stage B from ClownsharkBatwing. Here you can find the workflow and the according model-links: #🏞｜general-with-images message

[crisp stream]

https://civitai.com/models/618692/flux @royal monolith

https://civitai.com/models/618692?modelVersionId=699279

[shrewd trench]

Ducks

[clever oar]

indian river?

[languid pebble]

Got the missing better cascade? I think it was A

One part is missing in your linked text

[crisp stream]

yes, it´s the ft vae HQ, don´t know where I got it from though, you can use clip_l instead anyway

[crisp stream]

https://huggingface.co/madebyollin/stage-a-ft-hq

[languid pebble]

Thank you! ❤️

[final grail]

https://tenor.com/view/oceanmam-barber-lip-bite-gif-8052459839889670503

[languid pebble]

It's such a nice and friendly community here ... people helping each others ... I like that!

[shrewd trench]

no i just wrote woman with ducks on water

[subtle meadow]

Only Flux creating "hands" for shark 🙂

[languid pebble]

Good morning coffee!

[subtle meadow]

Made original idea to dynamic to get much more versions. For me SDXL not working because the shark missing or wrong:
A beautiful {25|30|38|40|45} year old {natural|red|redhead|blonde|black|colorful painted} {long|short|cool} {curly|straight} hair model woman is relaxing in a {blue|green|red|gray|black} hammock, wearing a sexy {light|dark}-{green|white|black|red|yellow|brown|blue} {fishnet transparent|elegant|luxury|normal|cheap} {normal swimsuit|lingerie|sexy underwear|sexy swimsuit|one piece swimsuit|bra and panties|silk apron}. She relaxes with her perfect legs stretched out, set against the backdrop of a sunny beach, with palm trees swaying in the breeze and the blue {quiet|stormy|waving} sea shimmering in the sunlight. From the water nearby, an anthropomorphic {funny|creepy|friendly|attacker|angry} shark is {standing|jumping|attack} upright, wearing stylish {brown|green|blue|pink|funny} sunglasses and holding a glass of {whiskey|coke|pepsi|wine|beer|caffee} in one fin. The shark has a wide, happy smile, its demeanor laid-back and confident, as it casually sips the cup of drink while observing the woman. Shark flirting with woman, talk something {erotic|funny} to her and {smiling|watching} her. The scene is {playful|funny} and surreal, blending tropical {windy|stormy|windless|summer|rainy} paradise vibes with the humor of a cool, {drinker|drinking} shark, creating a unique, fun, and detailed full-body photo realistic shot.

[celest sigil]

🤣

[subtle meadow]

As I see booth pic made by Flux? 🙂

[celest sigil]

Yes

[subtle meadow]

Kid shark like milfs (and singing):

[celest sigil]

https://tenor.com/view/shark-tale-lenny-bar-stool-on-chair-spinning-gif-27061076

[subtle meadow]

what a crocodile 🙂

[languid pebble]

Flux changed to male ....

[celest sigil]

Never had that problem!

[subtle meadow]

Maybe female shark....

[subtle meadow]

As I see "he" have success 🙂

[subtle meadow]

Not too dangerous idea to tie up the hammock to the shark?

[languid pebble]

I changed it to male 🙂

[celest sigil]

Ah! I thought you meant Flux had changed it.

[subtle meadow]

(too much image when the woman keep the drink instead of the sea creature. How to hack the prompt if I want drink to the creature's "hand"?)

[subtle meadow]

I don't give buy for yourself:

[subtle meadow]

As I see the sharkcrab is exist and attack:

[solid pecan]

Generate a picture which is featuring with 3D scene, a white chinese buliding beside a jade river, with a gian glowing moon in the background. Jade and golden color theme, Chinese painting in the style of Pixar, C4D rendering, Minimalist Style, warm and peaceful feeling --ar 3:4 --personalize 78yex4m --stylize 550 --v 6.1

[celest sigil]

Here is your image...

[subtle meadow]

...and guess what, I'm swimming calmly in the sea, suddenly I notice that the surfer in black is stuck in one of my protruding teeth, and I've been dragging the poor guy along for hours

[steep sleet]

SD3@ClipDrop

[subtle meadow]

Flux accept any unreal mutations 🙂

[tidal river]

ahah these are adorable and so detailed!

[subtle meadow]

My favorite topic but for today I finish them 🙂

[steep sleet]

SDXL img2img using Mobius checkpoint

SDXL img2img using Mobius checkpoint

SDXL img2img using Mobius checkpoint

[still rose]

Did use some alr used prompts online but it did turn out good

[pallid ruin]

Nice! I like the lighting

[wild sorrel]

yea looks like old school phone camera

[fading comet]

hi

[hearty violet]

hello there

how can I put a11 on white mode?

[steep sleet]

Open .bat file and delete --dark mode

[quiet current]

it follows your system theme, but you can force it to white with --theme light

[subtle meadow]

Bored the shark:

[topaz smelt]

how do I fix this I alreeady did the upcast setting

[inner grotto]

/img

[plush leaf]

?

[crimson dawn]

you have a very old video card and can't use half precesion. You need to launch the UI with --no-half as a command line argument. this is not an ideal situation. This doubles the size of the model in memory. fp16 vs fp32.

[crimson dawn]

RIP to one of the Greatest Kings in history

[topaz smelt]

okay thank you where but I cant type in the cmd where do I have to put that? and I have an Nvidia 1650 SUPER is that that old?

[crimson dawn]

if you're using automatic1111 you'd open webui-user.bat in a notepad, and add it to the line for comand line arguments. assuming you use that bat file to launch your app. It's the recommended way to launch.

[crimson dawn]

1650 is quite old for video card standards

[steep sleet]

Dave Prowse was the 'physical'actor wearing the Darth Vader costume - and James Earl Jones voiced Darth Vader. R.I.P.

[crimson dawn]

Darth is awesome. The voice is iconic. Such a different vibe too. James switches the inflections completely! I'm glad Prowse's voice wasn't used. I've seen the raw take lol. Good times.

James was just one of the truest actors around. Such a great perform. Knows how to embody a character. One of my favorite moments was his cameo on Big Bang Theory and he was acting so juvenile, taking sheldon to ding dong ditch prank Carrie Fisher late at night.

[crimson dawn]

yeah its old. 1650s were a budget line of cards produced after the 20s came out. Long before the 30s. the 50 series are almost here. You're back on a late generation 10 series.

Thats a monumental amount of time in the GPU space. Especially for machine learning purposes.

[crimson dawn]

@topaz smelt i told you where here

oh wow. WOW. where is my head i answered that post already. you weren't asking again. doie

Time to go outside. Get vitamine D

[steep sleet]

ARGS!!

[crimson dawn]

truth

[steep sleet]

🥳

[royal monolith]

@wispy nest

flux's idea of noiseless

[wispy nest]

lol wow

its quite hard to not trigger that effect sometimes

sure does look pretty noisy also

[royal monolith]

kind of old tv scan lines noisy

[cedar ocean]

主题：《从释经到讲道》学习班
时间：2024年10月9日-11日
地点：花园酒店
导师：知名牧师吴荣滁牧师
内容：包括深入研究圣经、理解其中的神学思想和传达方式，以及实践性的讲道技巧，以提高你的领导能力和牧养工作效果。

[celest sigil]

https://tenor.com/view/invisible-gif-4252759544531179440

[languid pebble]

Missing a photo of Trump on the wall

[celest sigil]

This one?

...or did you mean a signed photo?

[languid pebble]

😄

[nocturne oak]

It would've had to have been great in the first place. It was only somewhat good.

[gentle perch]

https://tenor.com/v0b7M0MdqQz.gif

[smoky cipher]

Would anyone tell me how to get as good as the result that we get in Midjourney for this prompt

A moon with a large circular hole filled with glowing yellow electronics.
Details: Intricate details, photorealistic rendering, textured lunar surface, craters, soft ambient lighting, visible wires, circuits, and chips, warm yellow glow, cinematic lighting, depth of field, volumetric lighting.
Style: beeple, Greg Rutkowski, trending on artstation, hyperrealistic.

[wispy nest]

there is a secret method to steal an image

1. use a captioning model to write your prompt
2. use IP adapter to transfer the style
3. use depth and canny edge control nets to transfer the composition
4. use color transfer tool to transfer the color

[steep sleet]

5. use AuraFlow to forward the gold

😄

[royal monolith]

stop using greg rutkowski in your prompts - not only did he make it very clear he hates that, but also his name really doesn't get you what you think it does

[wild sorrel]

Why would he hate it o_O ? It only made him more popular among AI people who never heard about him lol

[pallid ruin]

lmao

[royal monolith]

yeah, and i pointed that out to him. he responded with a nasty gram. he hates it. waste of free publicity in my opinion

[steep sleet]

Flux.Dev

Flux.Dev

Flux.Dev

[crimson dawn]

his art was never in the dataset. like one or two bits of it. CLIP G and CLIP L however, associate his name as a token to a lot of that kind of style. With that in mind how you're just referencing the general style of D&D gamebook art, ALL artists in the field and not rutkowski specifically, you can see the name as a tool to aim for an aesthetic.

As long as somoene isn't publishing it as "This is art by Greg Rutkowski" it's 100% in the clear and he can suck nuts

that being said, refines get far away from the base model's clip alignment

[royal monolith]

doesn't matter if his art is in the data set or not. he doesn't want his name used, so out of respect for another human being, dont' use his name

[crimson dawn]

the tool only works so well depending on which model you use

[crimson dawn]

sucks to be him I guess. His name isn't really that valuable

[royal monolith]

no, it never was in anything but disco diffusion - but people tend to just use what others use

[crimson dawn]

some people demand respect that's just unreasonable. Like some celebrities don't even want fans to look at them. I aint about that

[royal monolith]

so you're okay with us using your name as a prompt modifier then?

[crimson dawn]

Now, another example. Gary Larson. He doesn't want farside published online at all. Deal. I'll never make a farside lora. Even though i love farside

[crimson dawn]

100% . How could I ever be harmed by that?

[royal monolith]

you couldn't be, and it might have some cool effects with stable - but some people just don';t want their personal stuff touched. have you tried using it, btw?

[crimson dawn]

if you start publishing stuff as if i made it and you're acting like me, that's another story. entirely different context

[royal monolith]

well that falls under several fraud laws anyway

[crimson dawn]

Some people are obsessive and think they can control the world. Their name is just a name though. Anyone can use it so long as it's not impersonating them maliciously. slander. libel. there are limits, but mostly it's just a name

[crimson dawn]

yeah. The laws have been long established and rutkowski is trying to extend his name rights beyond those

not about that at all

imagine if JK Rowling said all people who had the last name rowling must change it and stop using it

fuck. that.

[royal monolith]

so being curious, i ran your discord name through flux dev and got these - what it would do in a longer prompt though, not sure

[crimson dawn]

Keep using Greg Rutkowski's name. Since flux1 is a base model (distilled but still close to the base and default clip understanding) it works again. pretty well. Not as good as G though mind you

SD3 has clip G so rutkowski really fires it up there

[crimson dawn]

i'm fine with all of those. it's probably the same as adding any kind of nonsense token though.

[royal monolith]

sure, but a nonsense token that, by itself, lands in a nice area of latent space

[crimson dawn]

RNG

on flux dev it's likely more todo with aesthetic guidance

[royal monolith]

regardless of which specific mechnisim, it's still a nice result. trying something a bit different

[royal monolith]

prompt: sunset by artist "nuuideas" - all have a nice art sort of feel to them. so i think your discord handle might be a good modifier

[crimson dawn]

100% fine with that. I doubt it's working for the same reasons as rutkowski would though.

tokens are just tokens. They affect the vectors that aim into the latent dimensions. If someone says not to use a token because it looks like their name, i'm 100% going to tell them to jog on

https://tenor.com/view/simon-pegg-sod-off-jog-on-gif-8266609749881880561

[wispy nest]

lol this is a really good token

I like the sunsets

[crimson dawn]

nuu ideas are always good to experiment with

[royal monolith]

doesn't matter why it's working, really - unless you're wanting to dig into the wiring.

[crimson dawn]

it's just aiming a vector. not really that big of a deal

[royal monolith]

me too, and just to make sure it wasn't a fluke, i ran it as the only prompt a second time and got really nice art sort of result. so i'm keeping this as something to use

[royal monolith]

sure, but sometimes it's nice to be able to aim at a specific target

[wispy nest]

its like a textual inversion embedding but found manually

[crimson dawn]

yup. which is why not using a specific token because a person is offended by it, is weird.

prompt: sunset by artist "fartfacer".

sometimes a cigar is just a cigar

[royal monolith]

that goes into the realm of being polite. someone's name is someone's name, and if they don't want people to use it, out of respect for another person, don't use it.

[royal monolith]

brighter coloring, more saturation though

not quite as nice

[crimson dawn]

courtesy only extends so far. when people ask too much, they're now the ones being impolite passive aggressively. forcing you into a position where your only stance is to obey their inconsiderate demands, or appear rude to them. Oh well.

[royal monolith]

and more of a vector look to them with sharply definced areas of gradient

[crimson dawn]

it reminds me of the 8 bit digital art era

[royal monolith]

it does - that's another prompt to play around with. what if you change sunset to something else?

[crimson dawn]

another image. i've got prompt game already.

[royal monolith]

might be good for a new side-scroller background set

[crimson dawn]

though as i investigate, it seems "fart" is the token that leads stuff to a pixel aesthetic. intersting.

[royal monolith]

so is it tokeising it to f-art ?

here's the result with a space in the middle. prompt: nuu ideas

still art, but in a different direction

[crimson dawn]

yeh. you're kind of just doing the legwork to demonstrate my point for me. cool.

[royal monolith]

part of what i'm doing with this exploration 🙂

i've found a couple modifiers that flux really likes - stick the term yesteryear on the start of any prompt and see what happens

[crimson dawn]

https://sd-tokenizer.rocker.boo/ says that fart is it's own token on clip L

fartfacer is f-art-fac-er though. But individual tokens don't matter so much when they're in asequence like that

i'm not sure that page is using Clip-L though. all the options are the same

[royal monolith]

you used fartface not fartfacer, however - does it tokenize the same way?

[crimson dawn]

no i used fart facer

[royal monolith]

i read it wrong then

[crimson dawn]

fart face is the more common one thanks to thanos. but wall facers are an awesome thing so i sorta smashed it together

https://www.youtube.com/watch?v=z6hEfK5C-Sg for scientific reference

[royal monolith]

yeah, it's a term that's been around a long time, usualy coming out of the mouths of small children

didn't know they did a skit with it though

[crimson dawn]

small kids know whats up. they're untainted by lame memes

[royal monolith]

which - strange connection happens in old memories - reminds me for some reason of a player i had in a DnD game so can to the session one night and said that someone had called his house and left a message on his answering machine with the phrase "the ice cream has no bones. The ice cream... has no bones'

so now i want to use the ice cream has no bones as a prompt

[crimson dawn]

https://youtu.be/GCWIT4IGix0 more scientific reference

[royal monolith]

youtube rabbit holes

[crimson dawn]

gotta do the research

[wispy nest]

ya gotta pay the tax!

[royal monolith]

i dno't think this is how a Trebuchet really works

[sinful swan]

Кот на столе

[royal monolith]

prompt: Umbra pegasus on the top of a mountain, lighting crashing through the sky

[languid pebble]

Good morning coffee

[dire notch]

1. 法棍：长条形、大且深棕色，刚刚烤好，表面粗糙，带有面粉痕迹。
2. 模特：穿着比基尼，法棍放在她的胸前。
3. 风格：复古风格，柔和光照，暖色调

[languid pebble]

I took a pill in New York ... ... ...

[celest sigil]

Asteroid mining facility

[limber widget]

boy

[twilit olive]

improve this house and make it a little more realstic

#🏞｜general-with-images improve this house and make it a little more realstic

[celest sigil]

Done...

[steep sleet]

Ollama img2img

Ollama img2img

Ollama img2img

Ollama img2img

[steep sleet]

Ollama img2img

[fresh canopy]

to create link for thread without images

[fresh canopy]

Great concept, you should work in advertisement

[slender vessel]

I was making some images for a video I am working on and had SD make me T800 breakdancing at a disco. This is the result. Never change SD, never change! lol

[wispy nest]

lol

[pallid ruin]

Epic!

[royal monolith]

@nimble mason flux really should be in a lab - prompt: Xylose Phosphate

[languid pebble]

Thank you! Unicorns are special fun 😄

Good morning coffee

[languid pebble]

The real reason why Unicorns extinct

[crisp stream]

I know that one with dinosaurs from back in the days 😄

[languid pebble]

Are you advertised by Artstation?

[crisp stream]

wdym?

[languid pebble]

I got a mail from Artstation saying: Moofi published artwork "OR-I-DJINN"

[wild sorrel]

Which model is it? I'm strugglign to make phoenix =.=

[crisp stream]

possibly because you are following me you get notifications? Don´t know much about how Artstation works in that regard

[languid pebble]

Ahhh... possible. I don't really use it. To many social media to watch them all ...

[crisp stream]

🙂

[wild sorrel]

would be great album cover

[crisp stream]

what music? 🙂

dark dubstep? 😄

[languid pebble]

Funeral music for astronauts ...

[crisp stream]

naaah 😄

too energetic 🙂

[wild sorrel]

dark trap 😡
https://youtu.be/SpDkEJk9QKE

or anything dark really lol

[crisp stream]

Haha, fits quite well 😄

@languid pebble

[languid pebble]

Selbsbildnis? 😄

[crisp stream]

wessen? 😄

[languid pebble]

Deines

[crisp stream]

dachte ich mir 😉

[languid pebble]

Ich teste gerade ein neues Flux Model ....

[crisp stream]

can it do painted appearances?

[languid pebble]

It should ... https://tensor.art/models/772796227503912754/FLUX-2-ULTIMATE-DREAM-DIFFUSION-V2

[crisp stream]

@languid pebble

[steep sleet]

Ollama img2img in ComfyUI

Ollama img2img in ComfyUI

Ollama img2img in ComfyUI

Ollama img2img in ComfyUI

Ollama img2img in ComfyUI

[amber pilot]

JuggernautXL DPM++2M Karras

[wispy nest]

really nice meal

[amber pilot]

anyone who can help me understand why this prompt never gives me realistic looking pics?

[wispy nest]

realism means art

you just want to say something like (photo:1.3) of a busy office space, people working

[amber pilot]

what does the :1.3 do? (im very new to this :))

[wispy nest]

it boosts that word a bit

[amber pilot]

Ah i see

Thank you, this is what i was looking for

[wispy nest]

you need to check the syntax

cos I use comfy not A1111

but its probably the same

[royal monolith]

Flux dev

[dusky token]

angry salami

[languid pebble]

You make me wonder what your prompts may be 😄

[crisp stream]

It´s as well with an input image of the 4080 here in open case 🙂

@languid pebble

[languid pebble]

lol

[crisp stream]

🙂

[crisp stream]

btw you can check out stephen gammell if you haven´t yet 🙂

[languid pebble]

Cool!

[crisp stream]

would be interested with what you come up

he´s not the only one in this prompt, yet a pleasant starting point I feel 🙂

[languid pebble]

Childbook Illustrator ...

[crisp stream]

yeah, well, definitely not, at least the paintings I´m referring to 😄

[languid pebble]

I never really did a lot IMG2IMG ....

[crisp stream]

possibly it´s about time? 😄

[languid pebble]

I'll check ... most of the time I know what I want ... but I add some wildcards sometimes, too ....

[crisp stream]

The most interesting part with AI is getting what you didn´t really expect I feel 🙂

[languid pebble]

I'm sure there's a market for it 🙂

[crisp stream]

f markets 😉 😄

[languid pebble]

Arrgs ... need to check my workflow ...

[errant egret]

create an image of a colorful mantis shrimp playing an electric violin as fish look on in amazement. larger than life, pixar style

[languid pebble]

Have a look here -> https://discord.com/channels/1002292111942635562/1237461679286128730

[amber pilot]

Whats your avg gen time?

Thinking about returning my 7900 gre and buying a 4070 super ti (not the same as 4080 i know)

[crisp stream]

What model?

[amber pilot]

I only have juggernaut xl as of now

So have not tested any others

[amber pilot]

might test flux soon heard alot about it that its great with texts

[crisp stream]

Let me check SDXL quickly

[scenic yew]

Yeah it’s the best at text for open source and comparable to the best closed source.

[crisp stream]

@amber pilot would have to wait a bit because I would have to reboot for a clean value. Currently busy though

[amber pilot]

No problem man

Take all the time you need :)

[crisp stream]

With multitasking programs open it delivers around 2.45 it/s

at 1024x1024

[amber pilot]

Yea thats the same for me with 7900 gre

[crisp stream]

wait

[amber pilot]

Windows machine?

[crisp stream]

that´s with dpmpp SDE (karras) taking twice the power of let´s say euler_a, where it´s 4.73 it/s

[crisp stream]

yep

[amber pilot]

Alright, would make no difference in me switching to a 4080 then, maybe better with a 4070 ti super idk really.

[crisp stream]

so you got 4.73 with euler a karras as well?

[amber pilot]

4.5 i believe

[crisp stream]

yeah, so seems similar

[languid pebble]

Good morning coffee!

[languid pebble]

Very dark ...

[steep sleet]

Can anybody please speed-up this Ollama/Flux workflow?

[steep sleet]

Ollama/Flux img2img (took 30 minutes to make!!!) Aha! I changed to "Unloads Model" and now licketty-split!!!

[steep sleet]

Flux/Ollama - img2txt (Ollama) - txt2img (Flux)

[steep sleet]

Flux/Ollama - img2txt (Ollama) - txt2img (Flux)

[steep sleet]

SD3 originals Ollama/Flux img2img

[sterile kiln]

Hi guys, I'd heard of a watermark remover that used Stable Diffusion (I think), do you have a name please?
I'm looking for a good watermark remover to train a LoRA Flux, since nowadays we even monetize digital images x)

[amber pilot]

You sure you dont mean copyright instead of monitize?

[worldly quest]

[sterile kiln]

Hmmm make sense x)

[steep sleet]

If you can copyright AI Art - let the world know; as it is exceedingly difficult to prove agency i.e. "I made that art!", as a court often decides that it is more likely made by an algorithm

[amber pilot]

Not sure if he was talking about ai art

[steep sleet]

Ollama and Flux img2txt - txt2img

[sterile kiln]

I'm looking for a good watermark remover to train a LoRA Flux

[steep sleet]

https://www.reddit.com/r/StableDiffusion/comments/1d5qg6u/how_to_minimize_the_effects_of_watermarks_on/

[royal monolith]

it's the unicorn of allergies hard at work

[languid pebble]

It's throwing up candies ... looks like it's kinda censored in Flux

[royal monolith]

ah. looked like pollen to me

[languid pebble]

They are way smaller 🙂

[sterile kiln]

🙂
https://civitai.com/models/758984?modelVersionId=848667

[shadow stone]

whats the difference in using t5xxl_fp8 and t5xxl_fp16?

[wispy nest]

fp16 better but more vram

[sterile kiln]

Then there are GGUF models and text encoders, which are almost as good, and save a lot of Vram.

[steep sleet]

Ollama img2txt / Flux txt2img - all one process

[chrome yacht]

Get a real job you scamming tramp

[crimson dawn]

As a cat, you should know that's not a real fish. The girl part is just so that a man feels like they're intellectually superior to this person and can't be scammed. Men are easy to fish in like that

[wispy nest]

quanting the Unet goes a bit better than quanting the text encoder

can alternatively lower VRAM burden from text encoders by just unloading them

[shy bluff]

"Cat sitting on the moon, eating a glowing fish, with Earth visible in the background and stars in the dark sky."

[languid pebble]

Take one for free ...

[dry blaze]

Slowly figuring out how to make a good longer multi clip FLUX text to video gen lol

[languid pebble]

IMG2Video?

Ahh... got it. sorry!

[dry blaze]

Pretty much. Blue section on the left makes the image, rest extends the video

Shit, forgot it was a img 2 video lol. Gonna see if i can make it a text to video xD

[languid pebble]

Cool ... but needs a lot of time I'd guess ....

I'm looking for something better than SVD for IMG2Vid

[royal monolith]

add a prompt section in front of it to create the image, then have that image become the one you would have uploaded

[languid pebble]

I'd think it's better to seperate the generations ... cause you don't want to animate a bad created picture

Theres no benefit for text2img

[royal monolith]

why? the animation could be cool

purz does that all the time and his stuff is very cool

[languid pebble]

But better use a good created picture for an animation than just a text prompt

As long as the text prompt has no influence on the animation ....

[wispy nest]

if its gonna take a long time
then yeah maybe want to make sure the input is good

[steep sleet]

Flux.GGUF_Q8_0

[languid pebble]

Easy to create 10 Pictures and animate the best. Harder to create 10 Animations ...

[royal monolith]

most 'not so good' have chunks that are fine, i just use those parts

[dry blaze]

Prompt section is here where it makes the initial image :P

Sadly SVD doesn't have a txt2img yet

[languid pebble]

Cause it's the worse concept ....

[steep sleet]

Flux.GGUF_Q8_0

[royal monolith]

doesn't need to. it won't know if you created the image first with that, and then you just ran the image into svd, or you uploaded an image

[dry blaze]

No, i mean, i got a text to video, but not for FLUX

[languid pebble]

TXT2VID only makes sense if the prompt has an influence in the kind of animation

Otherwise PIC2VID is the better solution ... like SVD

[scenic yew]

try cogvideox 5b, its the best open model rn and only requires like 5gb vram at the least. Image to video support isn't here yet but should come out this month(most likely next week I believe)

[languid pebble]

Thanks for the information!

I'll keep an eye on that

[wispy nest]

img to vid arrived

2 hours ago on reddit i saw it

[scenic yew]

yeah here is the link: https://huggingface.co/THUDM/CogVideoX-5b
its not quite kling level but still great.

Oh wow, yeah just seems to have arrived.

[languid pebble]

At the moment I have pictures of a product willing to make a video ...

[scenic yew]

you can just try it here: https://huggingface.co/spaces/THUDM/CogVideoX-5B-Space

[wispy nest]

What does it mean by "negative values"

[royal monolith]

it thinks you're using photoshop, i think.

[languid atlas]

1-10 what do you give me for the repaired hand?

[languid pebble]

Good morning coffee

[autumn fern]

Can we restore old monochromatic images using Stable Diffusion

[languid pebble]

You could have a look on SUPIR ... not 100% sure if it fits to your needs ...

[autumn fern]

Thanks I'll try that

[wispy nest]

there's a recolor control net out there

SUPIR for an upscale stage would be fantastic too

you might want to look at inpainting for problematic areas

[sterile kiln]

Do you know "Super Menteur" ?
🇫🇷 🥖

[steep sleet]

C'est moi, n'est ce pas? 😉

[sterile kiln]

haha, non mais, je vais te repondre en image 🙂

[steep sleet]

Super Liar en image?

So I just make a two word prompt ... I see!

Flux Q8

Qu'est ce que c'est un Super Menteur?!

[sterile kiln]

It's "Super Menteur" !
Now completely overtaken by his replacements, he's almost outdated.

[steep sleet]

Chirac! Ooh la la!

[sterile kiln]

You may be too young to understand this reference x)

[steep sleet]

Soixante huit

[sterile kiln]

Serieux xD ??

[steep sleet]

Ouais

[sterile kiln]

Ah, alors "Mangez des pommes" vous parlera x)

[steep sleet]

Jacques baby!!! 😄

[steep sleet]

Chirac sur le dos d'un ours!!!

[sterile kiln]

La je travaille Super menteur, il merite un beau wallpaper en 4k

[steep sleet]

Il habite ici

[sterile kiln]

Ah ça, faut demander a bernadette

[steep sleet]

Bernadette qui?

[sterile kiln]

sa femme xD, vous etes vraiment de 68??

[steep sleet]

Ouais, mais j'ai oublier beaucoup!!! MDR

[sterile kiln]

le RPR ça s'oublie pas

[steep sleet]

"l'Etat? C'est moi!!!"

[sterile kiln]

"Oh celui la il va nous faire peter l'emmerdomètre.."

[sterile kiln]

Chirac Alerte à Malibu

[steep sleet]

🥳

[steep sleet]

Flux.Dev+LoRAs - LoRAs = Toy Camera, Wraith_BW, Blue Future

[untold hedge]

La même avec Poutine est on est bon / actuel ^^

[sterile kiln]

C'est tout le coté artistique de la photo, Chirac serait le dernier digne president, le seul capable de monter un ours...
Comme Poutine x)

[untold hedge]

Il est vrai ^^

[steep sleet]

Fan Club Jacques Chirac!!! 🥳

[elfin marsh]

made it a little more crispy

[clever oar]

but its virus ,not food 😃

[steep sleet]

Flux+LoRAs

Flux+LoRAs

[languid pebble]

Good morning coffee

[graceful island]

Had to show these off somewhere

[languid pebble]

Thank god not so bad here

[fickle moss]

donald trumpin anime style,sunny v2

[lavish garden]

Prompt: https://civitai.com/images/29750620

"Snow Moon" - Oberon (Warframe) | Reimagined

[languid pebble]

Young Donald T. ...

[dry blaze]

@crimson dawn Well, i've somewhat gotten the gist of how it works, now to put it to the test

512 x10, and depending on how much memory it uses, maybe by 10x more xD

[wispy nest]

https://images-wixmp-ed30a86b8c4ca887773594c2.wixmp.com/f/2ca4b697-5687-4d7c-a41b-58e19ad75957/di701uo-fe6ee704-d0b2-49ca-b810-978eb472a61e.jpg/v1/fill/w_894,h_894,q_70,strp/counterspell_74_by_furostnova_di701uo-pre.jpg?token=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJzdWIiOiJ1cm46YXBwOjdlMGQxODg5ODIyNjQzNzNhNWYwZDQxNWVhMGQyNmUwIiwiaXNzIjoidXJuOmFwcDo3ZTBkMTg4OTgyMjY0MzczYTVmMGQ0MTVlYTBkMjZlMCIsIm9iaiI6W1t7ImhlaWdodCI6Ijw9MTAyNCIsInBhdGgiOiJcL2ZcLzJjYTRiNjk3LTU2ODctNGQ3Yy1hNDFiLTU4ZTE5YWQ3NTk1N1wvZGk3MDF1by1mZTZlZTcwNC1kMGIyLTQ5Y2EtYjgxMC05NzhlYjQ3MmE2MWUuanBnIiwid2lkdGgiOiI8PTEwMjQifV1dLCJhdWQiOlsidXJuOnNlcnZpY2U6aW1hZ2Uub3BlcmF0aW9ucyJdfQ.AG8ZHbOMufEyyon9J0CvjPI99Jh7v0lTXtn8WB6zFmg

https://images-wixmp-ed30a86b8c4ca887773594c2.wixmp.com/f/2ca4b697-5687-4d7c-a41b-58e19ad75957/di720ff-f4ea1e30-2c7c-4a0f-adf7-8d6f4b5de281.png/v1/fill/w_894,h_894,q_70,strp/counterspell_74_iopaint_by_furostnova_di720ff-pre.jpg?token=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJzdWIiOiJ1cm46YXBwOjdlMGQxODg5ODIyNjQzNzNhNWYwZDQxNWVhMGQyNmUwIiwiaXNzIjoidXJuOmFwcDo3ZTBkMTg4OTgyMjY0MzczYTVmMGQ0MTVlYTBkMjZlMCIsIm9iaiI6W1t7ImhlaWdodCI6Ijw9MTAyNCIsInBhdGgiOiJcL2ZcLzJjYTRiNjk3LTU2ODctNGQ3Yy1hNDFiLTU4ZTE5YWQ3NTk1N1wvZGk3MjBmZi1mNGVhMWUzMC0yYzdjLTRhMGYtYWRmNy04ZDZmNGI1ZGUyODEucG5nIiwid2lkdGgiOiI8PTEwMjQifV1dLCJhdWQiOlsidXJuOnNlcnZpY2U6aW1hZ2Uub3BlcmF0aW9ucyJdfQ.R-FSASE24oMxtygba6qdav_lQiZcMW7EoqRMFkwkS8w

[crimson dawn]

512 sized tiles is default on forge too, but you can go bigger tiles. like if you can do 1024 images, do 1024 tiles

i think

[wispy nest]

iopaint leaves some blurred marks. What can I do?

[crimson dawn]

super cool experiment . it is pretty great to work with. i've used it with flux a bit but those experiments got lost in the sauce

[dry blaze]

Yep, bigger is always better, especialyl when you're dealing with sizes where one square is still within the same "color palette" lol

[crimson dawn]

https://tenor.com/view/shit-jeff-goldblum-jurassic-park-gif-15006020

[dry blaze]

Aye. I've been screwing with flux as well. Heck, thanks to ai-toolkit, i made my first lora! Never understood how to get kohya to work. And despite all my effort, haven't found out if toolkit works for 1.5/SDXL

[crimson dawn]

what's your issue with kohya? it's the dataset structure isn't it?

[dry blaze]

Yep. Way too damn overwhelming. Meanwhile ai-toolkit was just add images, add text per image with prompts, open it's config file, name total steps (i use 100 steps per image, so 600 steps for 6 images), then simply fire up it's .py file and it's config file, and it's going lol.

[wispy nest]

this image from the other day was using tiled diffusion its good
this is downscaled to 1k but the image was 4k

[dry blaze]

Damn lol

@wispy nest What upscaler node/model did you use to shoot it up to a res the sampler dealt with?

CAuse iterative even on 6/10 steps i set it to is a tad slow lol

[wispy nest]

I always use adaptive token dictionary

[crimson dawn]

yeah it seems that way. the toml files are powerful. All i do though?

I make a folder called "datasetname" and inside that folder i make one called "#_token" in my most recent one it was 5_robot. That tells it to repeat every image in there 5 times. or 1_token would make it do 1 image repeat per epoch. that's how others work by tdefault i think, 1 repeat. then in the folder its the same. txt files paired to photo filenames. to build datasets with multiple concepts, you'd put each concept in a 1_token folder. or with the repeats to balance the set. 2_datasetwith10images and 1_datasetwith20images. sorta.

i just ignore toml files, use the folder structure, and the folder name is what trains when you drop captions.

[dry blaze]

huh, never heard of that one Link me it's node

[crimson dawn]

the kohya gui one automates a lot of the tedious stuff too

[wispy nest]

https://github.com/Phhofm/models/releases/tag/4xNomos2_hq_atdgoes into the "upscale with model" and "load upscale model" nodes

[dry blaze]

Aye. That's what i've tried too, but for some reason, my last attempt was just either no lora effect, or directly one of the training images lol

Or no, not all of it, just #_robot xD

Most of the rest just confuses me lol

[crimson dawn]

love that when you get the image directly back out

sometimes i love that

at least you know the training works haha. like a business that can't handle the volume of customers

[dry blaze]

Ah ok. What's the differences between nothing, 135k and 205k?

[wispy nest]

slightly different flavours

its 3 different checkpoints

[dry blaze]

Yep xD

Ship me your SDXL and 1.5 config for kohya, and i'll give it a go :P Howmany images do you use normally?

[wispy nest]

I prefer the 180k (the default one)

[dry blaze]

Ah ok. So no numbers is the final version, and rest is just "opinion" based earlier "spitouts"

[wispy nest]

yeah

[crimson dawn]

i dont really have a config saved. i just use the presets and fiddle with settings everytime. SOON||™|| i'll combine all my ideas and explorations into one preset and use it.

haven't done sdxl for a while. been training on flux since i got back into this stuff. Just recently i figured out that cosine_annealing is just cosine schedule with a minimum learn rate, which is cool beans

[wispy nest]

if that one is too slow, he released versions for drct-l, dat2 or mosr as well

they will all be faster, but a bit worse

they are in the same repo

[crimson dawn]

adaptive token dictionary sounds hot af. but i dont know why it intrigues me so

must explore that later

[wispy nest]

its a 🔥 name yeah

[dry blaze]

I don't even know what cosine is xD

so, i can't really use a upscale model, because upscaling an image is gonna eat vram like mad, unless there's a multidiffusion for upscaling models xD

[crimson dawn]

learn rate schedule. constant would be the same learn rate the whole time. cosine is going from what you set to 0 then back up, in a cosine wave, the sexiest of the waveforms.

annealing just bottoms out at a minimum setting instead of 0. all this time i thought it was fancier math that i needed to be in the right headspace for

[wispy nest]

oh I think you can tile for most upscalers

[dry blaze]

Curious if i can mess with the tiled diffusion node to simply change checkpoint model to upscale model

[wispy nest]

I know you can tile them in Chainner

[crimson dawn]

https://github.com/LabShuHangGU/Adaptive-Token-Dictionary?tab=readme-ov-file#fast-inference according to this, it's fast on cpu?

[wispy nest]

its very slow on CPU

[crimson dawn]

oh i misread

[wispy nest]

I've done it before so its not impossible

[crimson dawn]

idealy, you know the result before you hit bake. so it's somethig you could click before lunch

[wispy nest]

yeah I used to queue up a bunch of upscales and then just leave it

the Span or Compact ones are more suitable for CPU like this one https://github.com/Phhofm/models/releases/tag/2xNomosUni_span_multijpg_ldl

or this one for patient people https://github.com/Phhofm/models/releases/tag/4xNomos2_realplksr_dysample

[dry blaze]

What does it actually do?

[wispy nest]

its just like ersgan but better

[dry blaze]

Ho, so it's basically "nvenc vs X265", where nvenc is miles faster, but much worse?

[wispy nest]

I don't know the reference sorry

can you use cars or animals as an example lol

[dry blaze]

Ah, like when you use hardware video transcoding, gpu is way faster, but it's more a "ignore quality control, just pump out frames", while cpu is thorough

[wispy nest]

oh I see

its a bit different, cos you can run all of these models on either CPU or GPU

just like stable diffusion itself

[dry blaze]

For streaming, it's no issue, but it's when said transcode will be the new smaller compressed movie, you want it to look good when natively played, and not a movie to watch at the summer house :P

[dry blaze]

Gotcha.

You guys wouldn't know of a upscaler model tiler, would you? xD

[wispy nest]

some people have a GPU so bad that
their speeds are kinda comparable to GPU

[dry blaze]

Indeed lol.

[wispy nest]

you can get tiling in Chainner

not sure a bout comfy

[dry blaze]

Meanwhile i wanna do stable diffusion on my steam deck, so the tiled shit is definitely gonna help xD

[dry blaze]

I meant for comfy with upscale models :P

[wispy nest]

not sure, sometimes there is a node for things and sometimes not

[dry blaze]

Cause so far i've made 4-6 or 8k images even if it takes forever with these ones

[dry blaze]

Also, does cannier have a resolution limit?

[wispy nest]

not sure, I haven't hit it

generally with tiling there shouldn't be

but AI apps don't have the most robust code

[dry blaze]

Nope lol

[wispy nest]

its likely a very high limit
since Chainner specialises in upscaling

[crimson dawn]

the different codecs lol. i like that analogy. its not perfect, but i get where you coming from

[dry blaze]

Aye :P

Well, all that remains now, is to wait for nvidia's announcement for the rumored "28GB" 5090, and whether it's worthy of it's 2.5 grand pricetag so i can increase the damn iterations per sec lol

[wispy nest]

unless its very "secret" images cloud is rly cheap

I can see RTX 3090 for $0.160/hr

[languid pebble]

And the 5090 needs 600 Watt?

[wispy nest]

maybe yeah

[languid pebble]

Someone told me it's better to get 2 3090....

[wispy nest]

oh

depends on the model

[languid pebble]

I don't think they will shrink 🙂

[wispy nest]

its more about whether it splits well

[languid pebble]

Flux is today's stuff ...

I wonder where the better SD3 is ....

[scenic yew]

sd3.5 is Api access only and its not really "better" then flux. It's worse in humans, prompt following, text rendering. However, it does seem to look better and more stylish compared to flux without any thing else, also has more knowledge since its undistilled.

[languid pebble]

I think we are talking about different things. If I am not wrong with the new TOS they said SD3 wasn't really finished and they will release another version.

[wispy nest]

well 3.5 is sorta still 3

[languid pebble]

Is there a 3.5?

[scenic yew]

Yes, not open source but some people have api access.

[languid pebble]

Let's hope the best for Open Source ...

[clever oar]

👋

[languid pebble]

I hope you are fine!

[clever oar]

not fine but i hope will be fine

[languid pebble]

Best wishes to you

[wispy nest]

second one was much more anime

main change was I took the hyper lora off

[viral frost]

bizarre unbelievable action photograph of an inlaid grobyk poop beansbattle and statistics infographic eralian greebling wobodogger wearing beh urban scene twist horrific horrific splatter device : thrilled husband legendary artiststroseating homomorphic bean dissection sliced noodles schematic manuscript masterpiece scientific masterpiece bizarre repeating vandalhtaking high tech Hardware excellent use

[royal monolith]

so not only do you want them to have only a few colors, but you want them tiny?

[exotic hound]

I am building a gen-model for our product, it called curtain light,

[royal monolith]

might prompt for a light brite

[exotic hound]

the curtain light contains only few pixels, we wish we could come up a end to end model to generate the low-res pic dirctly

[royal monolith]

you don't want an AI for that, that's the wrong tool - just write a script to draw to the screeen

[exotic hound]

the pic i show u is from SD1.5, then resize ,color enhancement, etc

[royal monolith]

okay, well, it's still the wrong tool. it's like you're trying to use a bulldozer to do the job a shovel would be better suited for

you might try an LLM instead of an Image generator

[exotic hound]

Maybe

then what is the prompt when llm

we ve also tried LLm

[royal monolith]

i dunno. i'd go talk to meta and tell it what I wanted to do, and get it's suggestions, then ask it for specific code and assitance.

but really, all you need is a script that draws colors to specific pixels on the screen - old school tech

[exotic hound]

got cha, I'll try it on 4o or maybe the latest 1o, thx buddy

[royal monolith]

yeah, see what 1o can come up with.

[exotic hound]

it o1 lol. i got it wrong

[royal monolith]

meta is free... just sayin'

[pallid ruin]

nice!

flux

[languid pebble]

Looks like a monday ^^

[steep sleet]

Will it still block the use of living artists - their unique styles - in the prompts?

Flux1.Dev + LoRAs

[scenic yew]

No idea, possibly or possibly not. It should know more camera terms and maybe a bit more styles.

[steep sleet]

Flux1.Dev + LoRAs

[dry blaze]

Oh boi, this will take a while to convert SVD_XT_1_1 to tensorRT, and had it prepare it to be able to generate 100 frames

[flat sand]

they say SD3 is unfinished since months. In the same time bfl trained flux 😅
I don't think they are training sd3 in background. Maybe they do some dpo, but I wouldn't expect any big improvements here

[hidden tusk]

https://www.youtube.com/shorts/W6xmpX-fdmg

[steep sleet]

Someone else loves TouchDesigner!!!! 😄

[steep sleet]

van Gogh-y type stuff - Ollama img2img, with Flux output

[languid pebble]

😄

[bright harbor]

request

another request

[deft walrus]

Flux just rarely knows how to get a battle axe right

[deft walrus]

I like the images here, may I ask what LoRAs you're using?

[deft walrus]

It can do a sword at least.

[prime harness]

Draw a complete cell

[winged anchor]

Generate a girl playing with a boy

May I just ask how to generate images?

[steep sleet]

Hi, and thank you for your appreciation of my work! The LoRAs are by andreac75 = Blue Future, Toy Camera and Wreath BW. Have a great day!

[left blaze]

hello guys, Who knows what negative value is?

[languid pebble]

Negative Prompt? Everything you don't wanna see. But I'd try to leave it empty ...

[steep sleet]

Negative prompt via mat3o - aaaaaaaaaaaaaaa aaa aaaaaa aa

Ollama img2img and Flux

[languid pebble]

Have a look here: https://discord.com/channels/1002292111942635562/1237461679286128730

[steep sleet]

img2img with Ollama, Flux + LoRAs

[steep sleet]

A pirate-themed Furby standing on comically tall peg legs, depicted in the style of an oil painting. The Furby is of normal size but standing on exaggerated peg legs, with stormy seas in the background. The mood is dramatic, with dark clouds and turbulent waters adding to the pirate atmosphere, while the Furby maintains its characteristic fuzzy, round appearance. The scene captures a sense of adventure and whimsy, blending the quirky appearance of the Furby with pirate aesthetics.

[steep sleet]

#🆕｜sd3 message

Borrowed from DallE Theme of the Day

"Minimalist rugged oil painting in faded earthy blue hues, capturing delicate details in vast solid patches. A weary and anxious queen wearing a night robe standing near an open stained glass window in the high tower of a castle, holding a candle in a simple holder. View from out of the window. She gazes into the starlit night, as a distant search party of horse riders at the bottom of the castle run away on a dirt trail into the trees."

[pallid ruin]

You can try flux for free here https://huggingface.co/spaces/black-forest-labs/FLUX.1-schnell 🤗

[deft walrus]

These are great, I'm using Vintage Movie and I love it. lol

[bright harbor]

it's like a value less than zero, so imagine going subtracting value when you're already at zero.

[viral frost]

The perfect gift for that person who seems to have everything:

[near herald]

"I'm looking for some ComfyUI flows that can quickly and accurately paste product mockups onto white templates (well, 80% accuracy is fine, I can do post-processing later). If anyone has this information, could you share it with me?"
some kind like pasting exactly the brand of left photo into the bare bottle on the right

[steep sleet]

img2img Ollam and Flux in ComfyUI

[wild sorrel]

https://tenor.com/view/report-sus-the-living-tombstone-among-us-freddit-gif-22800160

[steep sleet]

A candle burning in the vacuum of space, with cosmic swirls of color replacing traditional smoke and flame shapes. The candle's flame merges with celestial elements, creating a blend of fantasy and abstract art, with vibrant colors and surreal space forms dancing around the candle.

[sweet kettle]

Can I ask for help in general guidance on where to start replicating this process? The processed image is on the left; the original image is on the right. Basically, they turned 2d art into a realistic image while retaining many details of the original art(pose, color, facial expression). I am trying using img2img while using the epicrealismXl model, with denoising strength at 4, but it's not producing the same result.

Also, for reference for other images like this by the original author on Twitter.
https://www.sotwe.com/N0LS0L

[wispy nest]

IP adapter with some reaslistic images as input would help
as well as injecting some noise before sampling

[sweet kettle]

thanks I will try this now, I will search in youtube for how, never used IP adapter before

[feral bear]

is this ai

[royal monolith]

no way to tell without looking at the actual image code, and probalby not even then

i can do that with photos in photoshop, or generate it with ai

[feral bear]

someone is selling an ai and i wanted to know if he is just scamming because my father wanted to buy it

https://www.facebook.com/share/p/pTkHp4Pz8VXV2NTQ/

[scenic yew]

I’m leaning toward ai, human looks weird and code is formatted a bit weirdly too. Could be real tho but human looks a bit too smooth.

[royal monolith]

selling an AI? yeah - no.

[scenic yew]

Sd3 channel is the flux channel lol

This too I guess

[royal monolith]

your father can do really good, AI images, for free - he doesn't need to pay that guy a cent

[scenic yew]

I agree with crystalwizard, flux(open source) could probably create similar images, no need to buy an ai from Facebook, seems very shady.

[feral bear]

yeah idk he was throwing weird numbers and comparisons with stable diffusion https://blog.skytells.io/avagen-a-new-era-in-generative-ai/

but he in his linkedin he was suppose to be a ml engineer at apple so i did not know if he is legit or not and wanted some proof before telling muy father that he is a scammer

[royal monolith]

who cares what he's supposed to be. not one of us is paying someone to generate AI images. but if he can't install it at home, there are sites like mage.space where he cna get a free account.

don't pay the guy - he's probably trying to sell stable diffusion, which is free and open source. he sure isn't trying to sell apple's products, he'd be out of a job fast

[feral bear]

ok thank you for the advice

[languid pebble]

Good morning!

[wispy nest]

they've got an official github and replicate so maybe they will post stuff on there if they make something
they just seem very small and new, its plausible they trained a small diffusion model but doesn't seem worth using

it only costs about $2,000 to train a model the strength of SD 1.5 from scratch these days

[worldly quest]

Ks no zzjkskNs

Ksnsmnsmsn

Najsmaknakmans

[deft walrus]

How many steps do you generally use? And you use euler?

I'm getting decent output (laughing at the horns tho) and using 20 steps with euler, but was curious what others did.

[steep sleet]

Yes, 30 steps, Euler and Simple

And I use the Wraith_BW LoRA

[deft walrus]

Do you see much difference between 20 and 30? I really don't.

I haven't tried Wraith_BW

I skipped the LoRA for speed and just bypassed that node

[steep sleet]

30 is a tad sharper/more contrasty

The LoRA makes a world of difference!

[deft walrus]

Okay, I mean, I keep it in mind but really, it is different but I don't know if it's better.

I'll check it out

I'm loving my workflow tho. I've got a slider node for my steps, an AI prompt generator, a 2d slider for image size and it's working great.

[steep sleet]

Good to know!

[steep sleet]

https://civitai.com/models/679352/wraith-bandw-flux

[deft walrus]

But you don't use B&W? How does it prompt?

[steep sleet]

If you turn the Strenght to 1 in the LoRA - then all output is B&W

[deft walrus]

Hmm, I haven't shared a workflow before, dunno how to share as an image

[steep sleet]

I'm keeping the LoRA strength at 0.7

Windows Key + G makes a screengrab ...

[deft walrus]

How do you attach the json to it to share, tho?

[dawn seal]

What is it about monsters with spaghetti in their mouth?

[steep sleet]

Try this for spaghetti-in-the-mouth output
Prompt = bizarre unbelievable action photograph of an inlaid grobyk poop beansbattle and statistics infographic eralian greebling wobodogger wearing beh urban scene twist horrific horrific splatter device : thrilled husband legendary artiststroseating homomorphic bean dissection sliced noodles schematic manuscript masterpiece scientific masterpiece bizarre repeating vandalhtaking high tech Hardware excellent use

[steep sleet]

Dunno - json is code, screengrab lacks the code

Although a PNG made with the workflow will have the same code as the json - if u d/load the PNG from the server

[deft walrus]

Yeah, that's what I was looking for. Mine is so spread out that it's hard to share and still see node names.

[steep sleet]

Cool - and yeah - too small to see 😄

[deft walrus]

Yeah, hmm

Okay, I only cut out the preview image and I joined the Reactor nodes into a single node, otherwise it seems good

[steep sleet]

Yours is by Searge - I love the nodes he makes!!!

[deft walrus]

Sure is, I love that. The thing I added that really helped with loras is text combine tho. I wish that was included in the Searge LLM node

The extra text is just for LoRA triggers to be included along with 'photo-realistic' that I want in every prompt

There are 3 Reactor nodes with acccompying preview image nodes, but that's pretty straightforward. I just load in 3 different images of 3 friends and if I think one of them would like it, I save it and pop it over

I left the strength on 1. heh

I'm new to having a good PC for images, and new to comfy, so I'm pleased

[steep sleet]

You're doing very well!

[deft walrus]

I'm liking this Wraith LoRA. I have photo-realistic in the prompt but it's very much an art style.

[steep sleet]

Yes, it needs careful prompting as it can "take over" the whole picture

To such an extent it is ignoring the details in the prompt

[somber socket]

Can someone Image 2 image this?
I wanted it to be more photorealistic, but it always takes away the glow of the flowers

[steep sleet]

... won't be long ...

[scenic yew]

Incredibly scammy, the ceo photo itself is ai generated(probably by flux which is open source) which alone tells you enough. The text on his lanyard is gibberish which happens very often if you don't tell flux the text to generate, skin is smooth too.

It's a scam, just use Flux which is open source(you can run it locally or online). It's the most popular model right now which can generate great text, perfect humans with 5 fingers, and has very good prompt following.

[languid pebble]

A lot of scam offers online at the moment offering unlimited A.I. for one fix payment. That can't work.

[scenic yew]

There are already so many free services that run flux, and a bunch of a.i models, no need to pay unless you are training or want to run models on virtual gpus.

[languid pebble]

I only generate local ... don't wanna be censored or my prompts beeing published ...

[viral frost]

Why not?

I just found this prompt fascinating due to its varied and detailed output, and how it seemed to depict otherworldly objects and creatures. Some of the generations give a Voynich manuscript vibe.

[languid pebble]

It fits to my topic ... human - machine ...

Good one @nocturne oak

Damn cool! Thank you!

Or hot?

[outer sandal]

can someone explain to me what this means and why it pops up when i try to use some models? Im very new to this i apologize if this is a dumb question

[nimble mason]

Three finger hand too

Shadows too... This idiot didn't even try

Says his name is hazeem Ali but the next on the project says hazem AI lol

[crystal flower]

Anyone know what style this is? Found it on pinterest

similar style
https://imgur.com/gallery/retro-1T3iPkp

[steep sleet]

Did u do this in SD? If so, how? 🙃

[languid pebble]

I've made it with Flux, prompting only

[languid pebble]

"Create a text design for text: "XYZ" with each letter filled with "Your motiv" ...

[steep sleet]

Ollama/Flux img2img

[languid pebble]

Ho! Ho! Ho! I’m a mighty pirate!

[steep sleet]

This one I used m3t4ll0g0 from @vagrant birch (strictly speaking, it is a Flux creation. Getting m3t4ll0g0 to work in Ollama produces too much Alpaca!!!") 😄

Dolphin-esque

[still mist]

Enrique

[pale steeple]

what would be the easiest workflow for making a person from an image move around a pose i grab from another film?

[steep sleet]

https://civitai.com/models/372584/ipivs-morph-img2vid-animatediff-lcm-hyper-sd

[vagrant dust]

I would suggest animate everyone as keyword for your search. You would need some custom nodes but it should work.

[clever oar]

packaging with a xenomorph skull in the meat department of the store

[pale steeple]

search where?

[royal monolith]

twitter first

[pallid ruin]

Lmao

They didn´t even menction which version of stable diffusion

🤣

[still mist]

Do an Image of Bill Clinton

How can I do an Image of Bill Clinton using laion 5-b

[dire lion]

guys, any idea what image-2-image workflow for ComfyUI I can use to transform screencaps of 3D models into concept artwork? Something that preserves details well, but also changes the style to a digital painting.

something similar to this maybe?

#📣｜announcements message

[steep sleet]

Try this Ollama/Flux img2img combo - the workflow is in the PNG (Open in browser)

[steep sleet]

#🏞｜general-with-images message

[dire lion]

oh I did not mean that specific style

just a workflow that would transfer source style to an image

[dire lion]

does this work like that?

[steep sleet]

Yes, and you can also ask it to replicate a style

[dire lion]

alrighty, I'll play a bit with it

see how it's going

[dire lion]

searched for Flux\Flux_LoRA_Wraith_Bw.safetensors but can't find this

ah think I found it

https://civitai.com/models/679352/wraith-bandw-flux

[steep sleet]

Wraith_BW can "swamp" your image - so to tame it - keep its strength quite low

[dire lion]

no idea what to do with the image combo module

what image do I need to pull with that and where from exactly?

there's a huge list if I click value but not sure where it's pulling it from

think I figured that out too

but now I get this

what model do I need there?

this is the error btw

ComfyUI\custom_nodes\ComfyUI-IF_AI_tools\IFImagePromptNode.py", line 193, in describe_picture
raise ValueError(error_message)

maybe I'm missing ollama?

or selected model is undefined?

[steep sleet]

IF_AI_Tools

Select model - llava2, or llama3 - d/load under Models at ollama.com

D/load the models by typing ollama run llava2:latest

Turn OFF "Keep Alive"

[dire lion]

oh ya I missclicked that 😄

[dire lion]

erm... is this the command prompt method? I don't really know how to use this 😦

where do I input these commands?