🏞｜general-with-images

[languid pebble]

https://websim.ai/ that's the service ...

But they will still all need food ...

[languid pebble]

Sounds like there will be a new SD3 soon ...

[pallid ruin]

Nice image! 😋

[clever oar]

thx😎

[steep sleet]

England are through!!!

[clever oar]

face control?

[languid pebble]

Sausage control ^^

[late sorrel]

Wurstchen control 😮

[languid pebble]

Best meat only!

[late sorrel]

at least 6 inches

[languid pebble]

I'don't know inches 🙂 England won?

Congrats! 🎉

I think he worked to hard ...

[late sorrel]

lol

[brisk phoenix]

Лев

[wispy nest]

I really like the furnishings, look at that white deer on the wall LOL

[wispy nest]

some one do

cat smoking from a bong on a couch floating in space over the earth looking at the moon

[royal monolith]

attempt one

[wispy nest]

this is way off

its supposed to be on a floating couch in space

not in a room

ok that is actually amazing

[spring mist]

my bongs just arent good lol

[jade lotus]

@royal monolith

[royal monolith]

well those are different at least 🙂

[jade lotus]

I don't think it would like to have the horse ride the astronaut, for nsfw reasons

[royal monolith]

none of the models are successful at putting the horse on the astronaut's back. or any other human. not for NSFW - but just becasue - how many images do you think it saw where horses were riding people instead of the other way round?

[jade lotus]

Where there is a will, there is a way

But yeah, Haven't seen any astronauts on a horse either

So brutally close

[jade lotus]

Yes?

[royal monolith]

wellll. he's standing on his shoulders, not really riding on his back, but at least they both look... mostly normal (really long legs on that horse though)

[candid surge]

"Hey, anyone wants a ride?"

[unique condor]

https://tenor.com/view/celeryman-celery-man-paulrudd-paul-gif-4808594

anyone know why this keeps happening when i use this lora?

[merry iron]

Thank you for using comcom analytics.
"comcom analytics" supports all community managers (moderators and server owners) by stats, visualization, and analytics.

If you have any questions, feel free to ask us!
Your dashboard
Help
Support server

Other languages
en: help
ja: help Japanese

[signal urchin]

Cat

Cat

[languid pebble]

Miau!

[vagrant dust]

here you go

[languid pebble]

Text is so frustrating ^^

[vagrant dust]

Did you tried depth maps as an alternative to sd3 or txt loras?

or QR-monster Controlnet?

[languid pebble]

Would be my emergency idea ...

I might be old fashioned ... but I like the idea it's created as one piece 😄

[vagrant dust]

Yeah but as for now at least longer text make to much headache

[languid pebble]

Is already cool but it's only "Dumdidi Sommer" 🙂

[vagrant dust]

Did you even tried to get a city skyline of an hanse city with blue church roofs ?

[languid pebble]

It should be Hamburg 🙂 I think it fits quite good

Could be the southern part of the port ...

[vagrant dust]

Yea the michel and some of the warehouses. And the text should look like on a poster on top of the image content

[languid pebble]

Do you think "Hanse City" might work better than "Hamburg"?

[vagrant dust]

Well most "Hamburg" pictures look like these,
So it seems the harbor part and the canals are photographed more then other parts. Obvious the Townhall seems to be a good picture too....

[languid pebble]

Pretty expensive town hall ... but pretty nice, too!

[vagrant dust]

The problem with city of hanse results into images from the time of the hanse...

If it is an important image i would puzzle a depth map with the different parts together like a collage and then send in thru the ai 🙂

[languid pebble]

I'll try ... my idea was Hanse City might have a bigger focus on the port while Hamburg is just a big city

[vagrant dust]

"images of the cities of hamburg, lübeck, copenhagen and malmö... 🙂

[languid pebble]

I'd think no real Hamburg 🙂

SD Forge I'd set batch to 100 and hope the best ... there's often only one bad point ...

[vagrant dust]

The last image is pretty sure from lübeck.

[languid pebble]

Zhank god no hands 😄

[languid pebble]

Thanks for your time! I need to go outdoor before Hamburger Weather starts 😄

[candid surge]

It's a nice feature that sd3 can sign the image for you - with a bit of luck

[analog panther]

@dry crow look at this

in one i used a refiner and in other i didnt

same images

why?

[dry crow]

because vae doesnt change the image much

its small details only like a hair strain

the first sdxl vae could cause strange stripes in images, thats why there was a fixed vae quickly after

[analog panther]

oh

so should i use it or not?

the refiner

[sturdy pewter]

Upcoming tutorial video intro - Live Portrait - this is lower resolution and quality

[keen depot]

"Modern open-plan medical facility with natural light, wooden structures, perforated brick reception desk, and seamless indoor-outdoor connection. Features healthcare staff, wood and earth materials, warm and natural atmosphere."

[celest sigil]

Nice quality, but is that a human mouth? 😄

[nimble mason]

lol

[languid pebble]

Like a coffee? 😄

[wispy nest]

is that by any chance Kolors?

[languid pebble]

Clownshark makes me prompt for kittens 🙂

[wispy nest]

cute kittens 🙂

[languid pebble]

What is Kolors?

[wispy nest]

i see some people on X posting about Kolors

im not entirely sure but i think that's a model based on sdxl architecture

[languid pebble]

It's SD3 ...

[wispy nest]

ahh ok. those animals come out pretty ok

have had some good images on scenery and animals that look good

[languid pebble]

Getting a strange kinda humor ... accepted disformed SD3 hands cause I thought it could be funny that some people think: The artist was unable to do pretty hands ... while others will know: It's SD3 😄

[wispy nest]

@languid pebble https://huggingface.co/Kwai-Kolors/Kolors

[languid pebble]

Looks interesting ...

[nimble mason]

cfg++ is pretty nice with RES at low levels^ last few had it

[wispy nest]

i wonder if i can run it on comfyui

[languid pebble]

I have no idea ... I don't know enough to try that ...

[wispy nest]

the file folder section looks like it can be run like unet

[wispy nest]

isn't cascade unet?

[languid pebble]

Clownshark is our Pro for Cascade ...

[wispy nest]

ok

this kolors is pretty new, can't find much resources on how to have it running

[languid pebble]

TBH I stopped trying to be the first user some time ago. First ones have all the work and bad results 😄

Early DeForum has been so complex and so much stupid work ... and it's pretty easy now ...

[wispy nest]

found this https://github.com/kijai/ComfyUI-KwaiKolorsWrapper

[languid pebble]

"it seems to be based on SDXL architecture with clip(s) replaced with the GLM model they are using"

[wispy nest]

yes, im curious how good the prompts and images are with it

[languid pebble]

Installing .... 🙂

[wispy nest]

sweet ... im still reading up on it

[languid pebble]

A lot of conflicts here ...

[wispy nest]

did you grab the files from unet, vae, text encoder?

[languid pebble]

Downloading the ChatGLM at the moment.... installing requirements got me the conflicts ...

started with git pull in costum nodes folder

[wispy nest]

ok

[languid pebble]

But I am not really good in this 😄

[wispy nest]

those files are pretty big

the wrapper explain the steps but will take me some time

think i'll grab chatglm3-fp16.safetensors

12gb in size

[languid pebble]

Maybe try a smaller version first?

[languid pebble]

I did the same mistake downloading the big one ... but you can still do if small one works ...

[wispy nest]

hmm

so chatglm3-8bit.safetensors should be ok for quality

[languid pebble]

For first try for sure!

If it doesn't work you didn't download the big one ....

[wispy nest]

is that kolors?

[languid pebble]

Lumina 🙂

[wispy nest]

im kinda confused with the intallation process

i did git clone the repo but not seeing the exact tree shown on site

[languid pebble]

in your costum nodes folder? there should be now a ComfyUI-KwaiKolorsWrapper

[wispy nest]

yeah that folder, but not seeing that path

ohhh i see it now

[languid pebble]

maybe you started your CMD in a wrong folder?

Ahhh...!

[wispy nest]

no, i was looking in the wrong path lol

[languid pebble]

Rebooting server ...

downloading model now ...

[wispy nest]

nice, does that image work to custom install any of the nodes in comfyui?

[languid pebble]

UserWarning: local_dir_use_symlinks parameter is deprecated and will be ignored. The process to download files to a local folder has been updated and do not rely on symlinks anymore. You only need to pass a destination folder aslocal_dir.

[languid pebble]

There's a workflow in the costum nodes folder in the example folder

[wispy nest]

warning message can be ignored i think

[wispy nest]

this ? example_node.py.example

[languid pebble]

Ah yes. looks like it's downloading

[wispy nest]

oh i see now kolors_example.json

did you have to download all 7 text encoders or are they downloading now?

[languid pebble]

D:\ComfyUI\ComfyUI\custom_nodes\ComfyUI-KwaiKolorsWrapper\examples in my case

[wispy nest]

yes D:\ComfyUI\ComfyUI\custom_nodes\ComfyUI-KwaiKolorsWrapper\examples

[languid pebble]

I think it's the thing you have downloaded seperate ....

[languid pebble]

it's downloading 4 files now ...

[wispy nest]

so far i downloaded the chatglm3

3-8bit

[languid pebble]

I think that's the text encoder or not?

[wispy nest]

i thought chatglm3 is the model

[languid pebble]

No .... model will be downloaded automatically

[wispy nest]

the model goes into unet folder i think, the image shows fp16

lol im bit confused

chatglm3 isn't the model?

[languid pebble]

I think it's a LLM Model ...

[wispy nest]

ok cross checking with you... download the chatglm3 model into checkpoints folder, then run the example.json to load up the nodes?

[languid pebble]

Nodes where already installed here ... have you installed the requirements.txt?

[wispy nest]

yes

im running the comfyui manager now

[languid pebble]

Restarting Comfy if it was running and load the example.json

models will download if you try to create a picture ... I hope 😄

Including text encoder as I can see now ....

[wispy nest]

just restarted comfyui after installing missing nodes, the example json file shows correctly now

[wispy nest]

i see its fetching 4 files in console

[languid pebble]

and 12 more afterwards ...

[wispy nest]

not sure but looks like console progress stuck on 25%

unless its downloading a huge file

[languid pebble]

I thought the ChatLM is kinda replacement for the textencoder ... but I think I was wrong ...

Check your traffic in the task manager

[wispy nest]

its working

[languid pebble]

Almost midnight here ... phew ...

[wispy nest]

getting all these bunch of errors, cause i closed the update while it was running to refresh the downloads ... do i need purge the cache to fix this?

[languid pebble]

No idea ...

[wispy nest]

think i'd have to donwload those files manually and put them in respective folders

[languid pebble]

It's downloading ChatGLM3 here now ...

[wispy nest]

ok, i already have chatglm3-8bit.safetensors in this path D:\ComfyUI\ComfyUI\models\checkpoints

[wispy nest]

sweet

do you see this in your local tree

[languid pebble]

Sample Picture ... it works

[wispy nest]

sucess

[languid pebble]

I don't see that ....

[wispy nest]

i dont either but image shows it might be here D:\ComfyUI\ComfyUI\models\diffusers\Kolors

[languid pebble]

Ahhh ... yes! Now I have that ....

Yes ... there it is

[wispy nest]

ahh im missing that tree

i see the rest

but not text encoder

[languid pebble]

It was downloaded when I first tried to generate a picture

[wispy nest]

oh that explains it

[wispy nest]

is that kolors lol

i keep asking lol

[languid pebble]

Yes it is 🙂

[wispy nest]

ahh nice 🙂

did you have to use negative prompt?

[languid pebble]

Interesting ....

No negative

[wispy nest]

sweet, i was wondering, that example on site didnt have any negatives

[languid pebble]

I always start without ...

Pretty good ....

[languid pebble]

Text not perfect 🙂

[wispy nest]

damn

these are good Mojo

i like the image texture and im guessing you are not using any lora or stuff

also no face detailer on those faces

[languid pebble]

Only example workflow ...

[wispy nest]

yeah, looks pretty decent for something basic

[languid pebble]

Took some of the prompts from here to compare: https://docs.google.com/spreadsheets/d/1vViQh45B-txERc5wnDVh_aNfHWLuEIrOCz7_HGXpPZM/edit?gid=0#gid=0

[languid pebble]

I hope you will get it to work soon!

[wispy nest]

ty, hopefully i will, the thing my isp is working on their server atm, and im getting some slow downloads

[wispy nest]

im down to the last 12 files

[languid pebble]

YaY! 🎉

Large ones .... will take a while!

[wispy nest]

brilliant, kolors im guessing

[languid pebble]

Bad luck no real blood eyes ...

[languid pebble]

Best wishes to you! I'm sure it will run after the download! Past midnight ... need to sleep! Good nite!

[wispy nest]

thanks man, gnite. im almost done too 🙂

viola... first run with Kolors

[wispy nest]

Kolors
a girl in bikini at a beach. sunshine, ocean waves, palm trees.

a woman sitting on a couch, watching tv, living room.

[ocean turret]

robotic

[ripe cedar]

Is there a way to generate landscape like trees and grass directly from a side view? Like a side scroller

[burnt wren]

Genera un portero de pelo color café claro, ojos cafes claros guantes amarillos y con un traje de portero que diga compensar, que el traje sea naranja con blanco y que sea modo caricatura

Generates a goalkeeper with light brown hair, light brown eyes, yellow gloves and a goalkeeper suit that says compensate, the suit is orange with white and is cartoon mode

@crisp stream Generates a goalkeeper with light brown hair, light brown eyes, yellow gloves and a goalkeeper suit that says compensate, the suit is orange with white and is cartoon mode

[nimble mason]

Here is the image you requested.

[pallid ruin]

I love this one, amazing 👏

[wispy nest]

I love this one
Sci fi is my favourite theme for AI images, and sci fi objects in a jungle is a great sub-theme too

[ocean turret]

create a robotic man.thanks.

[jovial tiger]

it does amazing things when upscaling lumina as well

[wispy nest]

the last one is nice, it does blue well

[ocean turret]

Create a robotic man ,thanks.

[nimble mason]

Here is the image you requested.

[languid pebble]

Congrats! Well done!

[modest kindle]

Does someone know how to make images like this?

[unique condor]

QR code controlnet. Wanna say one of the main models was named something like QR monster? I'll double-check(yeah it's one of the main ones, should be an sdxl one as well)

[steep sleet]

SDXL Checkpoint = Mobius. Made using ComfyUI backend in SwarmUI.

[modest kindle]

thx

[tropic shell]

Oh! Love the castle!

[celest sigil]

Kolors

[rugged dagger]

Wheat wire coils intricately woven, shades of grey and carbon black, forming an abstract pattern, desktop background, bokeh effect, soft shadows, subtle textures, high definition, digital render.

[nimble mason]

made with stable castleade 😉

[wispy nest]

thanks! have you looked into the new controlnet union?

[languid pebble]

I think not yet 🙂

[wispy nest]

this is really nice what they did with sdxl controlnet

#✨｜sdxl message

[languid pebble]

Yes it's a powerful tool. I don't know why but for me it feels a bit like cheating 😄

[rugged dagger]

Another example of how you guys can turn AI art into fun masking stuff.

[steep sleet]

Who remembers "The Prisoner"?! SD3@ClipDrop

[candid surge]

sd3, 19 steps, euler, line art

[jade lotus]

@shut sinew

[languid pebble]

Is he Admin here?

[rugged dagger]

If you're gonna spam at least use a good image.

There ya go. Merry Christmas.

Made in SDXL + Davinci Resolve

[rugged dagger]

Sorry about your eyes.

[eager wasp]

Kolors

[nimble mason]

last one was mean-following sampling after branching from 2->4->8 total steps at steps 1->2->3, respectively

this is following greatest gradient descent

[nimble mason]

following the median

[eager wasp]

Kolors once again (it does better for me when translating prompts to chinese, aaaargh)

[nimble mason]

30 steps

[jovial tiger]

I'll allow it

[hearty violet]

hi there

exist to much difference between this sampling method?

or its more depending of wich whas trained the checkpoint?

[hearty violet]

other question

[hearty violet]

how I use controlnet on a especif region? likethis bench? this paint dont work

[crisp stream]

how does it look in a varying style (not high contrast)?

[nimble mason]

Can do anything

[crisp stream]

don´t mind either

[crisp stream]

@nimble mason yet have you got an example?

🙂

[wispy nest]

uh their comment got deleted
but control net with canny edge and depth map
combined with multiple IP adapters
is how you get good img-to-img

[wicked lagoon]

ok i will try that

[wispy nest]

there are different types of IP adapter
the "style" ones tend to use CLIP ViT
but there are ones that use dedicated object detection models such as for faces and clothes, and these tend to be based on the You Only Look Once series of CNNs

[wicked lagoon]

woah thats complicated does civitai provide those ?

i haven't seen any object detection models before

[wispy nest]

You Only Look Once is actually super old
the paper came out in 2015
the best way to use this stuff is comfyui
people have made comfyui nodes with lots of different models

InstantID and FaceID are particularly good

[nimble mason]

will get one later 🙂

[haughty roost]

Wasn't that detailed of a prompt you know:
Dramatic moment from the Tour de France 2024 as cyclists navigate a challenging ascent in the French Alps. The scene's landscape composes of towering snow-capped peaks, winding roads, and lush green valleys. Cyclists sweat as they climb, spectators are lining alongside the roads, cheering and waving flags, and there are official Tour de France vehicles, colorful team jerseys, and the alpine flora and fauna

[nimble mason]

which model

[haughty roost]

Stable diffusion 3

Yeah I think you could learn a thing or two from my prompting cause as you can see, I used some keywords for very high resolution and detail, especially the texture in that skin

I mean you can see every cell in his body

[languid pebble]

Good morning coffee gives you wings ... 🤪

[obsidian light]

Please follow the structure of the house in the uploaded picture to generate the exterior decoration renderings of the house. The decoration style is modern and fashionable, not rural style. A half-walled courtyard is added to the floor, and there are flowers and plants in the corners.

[rugged dagger]

One of the problems you're having is the amount of tokens you're using. You're gaining nothing from the end of the prompt. Tokens are words and symbols used within the prompt. And then, on top of that, your filter or lora adds additional tokens.

Short version: You're using too much stuff.

[languid pebble]

Good morning dicordos!

[haughty roost]

Can you check out #🤝｜tech-support I did use a lot of words I guess in there but still, even without a lot of words and only like 4-5, I still get horrendous looking photos while other peoples look like they took it with their phone

[rugged dagger]

This google doc was put together by Emonz over at Playground. If you learn the words you can avoid using some other outside stuff. https://docs.google.com/spreadsheets/d/1839zMxYbzgrpeN4T7waJ0teZVRhkWhELdu082vHMEgE/edit?pli=1&gid=0#gid=0

^save that

[languid pebble]

Have you checked the information on civitai.com about the best settings for your used model?

[rugged dagger]

There are a lot of variables.

[haughty roost]

Did not know that was a thing

I don't see it?
https://civitai.com/models/497255/stable-diffusion-3-sd3

[languid pebble]

Some models like a special Sampler ... more or less steps ... more or less CFG

[rugged dagger]

☝️

[languid pebble]

I didn't get it's original SD3 ... for example it dislikes some Samplers. You can have a look at the picture descriptions ... what others used.

[haughty roost]

I don't know all the terms and stuff, I started this whole thing like 6 hours ago

[rugged dagger]

That's why I provided the above google doc. The information for smaplers is included. #🏞｜general-with-images message

"Smaplers" oops. Trademark? I like it!

[haughty roost]

No but still, I feel like if I do just edit my prompt with those words it'll still look bad. I feel like it might be my end

[rugged dagger]

Samplers sir, samplers

Dreambooth, Juggernaut, they're all in there.

[haughty roost]

I searched dreambooth and juggernaut, I don't see it in the list?

[languid pebble]

I wonder where the picture generation information is gone at CivitAI ...

[haughty roost]

No I mean here

[rugged dagger]

Samplers are on the right side of my image of the google doc. Those are the ones compatible with the model/lora you're using.

Playground calls the "Filters".

[haughty roost]

There's no sampler for this one?

[rugged dagger]

Steps, prompt guidance recommendations, etc. It's all i nthere

[languid pebble]

You can try you prompt here: https://glif.app/@Oliveira/glifs/clw44qfbl0000m0zztwqk2tnf If it looks better it must be a settings problem ...

[rugged dagger]

If you don't see a sampler it's just a prompt extension (which increases tokens). Can be used universally.

[haughty roost]

Oh trust I've tried my prompt with dall e 3 and it does work, not the prompt having some hidden code or anything

[rugged dagger]

Stable Diffusion is far more advanced. It will yield muc better quality. However, sampler, step, prompts, etc all have to be lined up. It's a pain in the ass. That's why sites like Playground exist.

StabilityAI want you to have a doctorate in astrophysics to understand their software.

[haughty roost]

So what can I do? Can one of you guys try my prompt and settings and see if it's actually art for you?

I feel like it can't just be my settings, something weird is going on cause I see countless of generations that look actually pleasing

[rugged dagger]

Sure. One sec.

[haughty roost]

How did that even autocorrect for you

[rugged dagger]

Dramatic moment from the Tour de France 2024 as cyclists navigate a challenging ascent in the French Alps. The scene's landscape composes of towering snow-capped peaks, winding roads, and lush green valleys. Cyclists sweat as they climb, spectators are lining alongside the roads, cheering and waving flags, and there are official Tour de France vehicles, colorful team jerseys, and the alpine flora and fauna

[rugged dagger]

It didn't. I fixed it. lol

^That image was Juggernaut

[haughty roost]

No not that

Sorry

Two terracotta warriors stand back-to-back in dynamic martial arts stances, defending against Mongol warriors on horseback. The Mongols, dressed in traditional clothing with spears ready, charge forward. The battlefield is dusty and sunlit, highlighting the warriors' detailed armor and intense expressions. The terracotta figures exude resilience, standing firm and coordinated against the oncoming threat.

And then 512x512

[rugged dagger]

It's a bad prompt.

[languid pebble]

Every A.I. uses a different prompt system. Dalle-3 prompts don't have to work with Stable Diffusion in the same way. Midjourney it's the same "problem" ...

[haughty roost]

See now something's just wrong with my medium then that's gotta be it

[languid pebble]

512x512 is a bad idea ... use 1024x1024

[rugged dagger]

He's correct. But even then, every AI has a max amount of tokens it will accept.

Stable Diffusion is 75 tokens? I think.

I could be wrong.

That includes the commas between your words. Each comma is a token.

[haughty roost]

Oh it's cause I see everyone else only use the comma style with sd

Like male, cowboy hat, riding horse, strong wind

I just put that in and hopefully I get something good

[rugged dagger]

I do as well. But my purpose is to organize and keep it clean. They're not required.

[haughty roost]

Also little change

And 1024

[rugged dagger]

The commas help me identify a word I want to move closer or further to the front of the prompt. The further the word is down the prompt the weaker it gets.

[haughty roost]

Now I got this weird thing

[rugged dagger]

Don't use Heun until you really dial in your finished product.

[haughty roost]

Is there a different version of sd cause I might need to switch

Ok, what do id witch it to

[rugged dagger]

The most generic result for SDXL is Euler A

[haughty roost]

Is that euler_ancestral?

[rugged dagger]

Yes

[languid pebble]

That's your prompt using SD3 via Glif:

[rugged dagger]

Much bettere quality. Mine was Juggernaut XL.

[haughty roost]

Nah I am so confused

Oh god, here's the new one

With dpmpp_2m_sde_gpu

[languid pebble]

Some samplers absolutly don't work. That's normal ...

[rugged dagger]

Sd 3.0 Realistic Vision Pro

[haughty roost]

Ok Creative guy

[rugged dagger]

Two terracotta warriors stand back-to-back in dynamic martial arts stances,

[haughty roost]

Just tell me what to put in

Show me how I get the sd 3.0 realistic vision pro thing

[rugged dagger]

Which version of SD are you using?

[rugged dagger]

For Comfy I am not the best person to ask tbh.

Mojo would have better insight

[haughty roost]

But is it the median?

[rugged dagger]

What VAE are you using?

[haughty roost]

What is that

[rugged dagger]

You have to have a VAE to process your images. Head over to the link Mojo gave you above.

[haughty roost]

https://glif.app/@Oliveira/glifs/clw44qfbl0000m0zztwqk2tnf?

That?

[languid pebble]

"So far I am having the best luck with dpmpp_2m (sgm_uniform or simple) at ~30 steps and 2-3 CFG "

[haughty roost]

Why such a low CFG? Don't you want it higher?

[languid pebble]

Above 8 it's pretty hard ... can't really say why ...

[haughty roost]

I just entered that

I still know it's gonna end up super bad

[woeful valve]

does anyone know how to create something like this with a video?

[haughty roost]

Ok this is LEAGUES better but still very bad as you can see

[languid pebble]

DeForum with Video Input ...

[haughty roost]

Missing an arm, leg, weird hair, and also its got that weird style

[woeful valve]

thank you mate

[languid pebble]

SD3 isn't good with hands, yet

[haughty roost]

So

Now what

[rugged dagger]

Playground V2.5 is very good with hands.

[languid pebble]

Try till you get good hands or no hands 🙂

[haughty roost]

But I wanna keep usingthe comfyui local thing

[languid pebble]

Or use another Model like a SDXL costum model from civitai

[haughty roost]

Can you send that to me?

[rugged dagger]

Playground isn't just a cloud based AI art generator site, it's also a local download.

Huggingface I believe

[languid pebble]

A list of SDXL models: https://civitai.com/search/models?baseModel=SDXL 1.0&sortBy=models_v9&query=SDXL

[haughty roost]

https://civitai.com/models/131843/cinemax-alpha-or-sdxl-or-cinema-or-filmic-or-nsfw
I think I wanna try this one

Also what does this mean?

[languid pebble]

Maybe if you want to generate online? Download the model and generate local. But you will need a SDXL Workflow for comfyUI ...

[haughty roost]

This all so confusing

[languid pebble]

It's been like that for all of us when we started. A lot to learn first if you don't want to use a webservice ...

[rugged dagger]

Welcome to Stable Diffusion, the worlds capital of confusion. 😄

[haughty roost]

Ok I got these 2 things:

[languid pebble]

Drop the safetensors into you comfyUI Model Folder, unzip Workflow and drop it into your comfy

Model Folder here for me: D:\ComfyUI\ComfyUI\models\checkpoints

[haughty roost]

Got it loaded up, now can you maybe make the prompt for me this time?

I want 2 terracotta warriors back to back fighting off mongols on horses

But hold on, I got a reference photo

[languid pebble]

Start with an easy one is good to lern and than make it more complex

[haughty roost]

Not the best but that's the idea. Not side by side but literally like ><

So can you please whip up an easy one and then if that actually works now with this new model I installed then I will try complicating it

[languid pebble]

two cyberpunk warriors fighting side by side, explosions

[haughty roost]

I mean it looks cool but not what I was going for

I wanted them more ancient chinese looking

[languid pebble]

You can change that ...

[rugged dagger]

cottoncandy style hyper-realistic smoke rising from a 1920's car, surreal trees background,

[haughty roost]

Looks so cool

It's way better than before just not looking that good

[haughty roost]

What model

[rugged dagger]

Playground V2.5

[haughty roost]

But like what model

Or is it SD2.5

[rugged dagger]

Playground v2.5

[haughty roost]

See I keep getting insanely bad photos even after switching to an SDXL model, lowering CFG, making sure the prompt is super short, and adding more steps

[rugged dagger]

SDXL is best at 1024 x 1024

Shorten your prompt. Less tokens = more

[haughty roost]

It's already so much shorter than before

[languid pebble]

There are tons of youtube videos like "Getting started with ComfyUI" ... maybe watch some related to SDXL first?

[haughty roost]

Here is a better image but the guy in the back has a questionable waist region

[rugged dagger]

3D pixar style candstriper holding a old radio in front of her

I used Real Cartoon XL on that one so it dropped the "candystriper" elements.

But hopefully you get the point. Less can be more.

[haughty roost]

Bro just stop flexing that your AI actually works 😭

[rugged dagger]

😂

[haughty roost]

Ok this is SUPER short

[rugged dagger]

Gotta run. Good luck gettin your stuff set up. Have a great day everyone!

[haughty roost]

Wait 1 quick thing

Can you just lend me your pc real quick so I can use it for the image generation

[rugged dagger]

Now describe the art style. Do you want it to be realistic? Cartoony? Surrealism? Add a flavor to it.

[rugged dagger]

Yes. Np! Check your mailbox.

[haughty roost]

That was quick

Oh yeah that idea sucked so Im gonna look up a tutorial on installing SDXL cause I think this movie one someone made is bad

[rugged dagger]

Yes good idea. And use the Finetune thread here. There's experts that hover around that thread.

They can help you.

Good day.

[languid pebble]

Need to run for lunch! Have a good day!

[haughty roost]

Oh shoot, I found this model on civitai called realvision 4.0 or something, and I found out you can click on demo pictures to see how they made them, so now I know the best settings

[languid pebble]

That's what I tried to tell you 🙂 But didn't find on my own at some places 😄

Good luck to you and have a creative day

[haughty roost]

morning*

Been up ALLLLLLLL night trying to get this working

Stability owes me 8 hours of sleep

[haughty roost]

Yoooo I finally got a nice image out of this whole thing

I just did 2 things with photoshop but that's it, this was realvision 4.0 and then like 4xupscaler

[nimble mason]

how do you know that's bad? maybe he just got home from a bad day at the battlefield

[crisp stream]

what sampler are you using?

[languid pebble]

Hi Moofi! Hello Clown 🙂

[languid pebble]

Hypnotoad ...

Look into my eyes ... take your creditcard ... ... ... 🤣

[solemn venture]

Is it possible to recreate a pony image ( which has used xl loras and so) with sd 1.5?

[candid surge]

depends on the prompt and image

but mostly 90% no

but in theory, with lots of work and a bit of luck, you might be able to find the right combination of prompt, various loras and a good checkpoint to get close

[solemn venture]

Would you recommend me to use controlnet for reference?

[candid surge]

idk, never used it much.

[languid pebble]

Can be pretty hard to impossible. Tried to redo an old picture and had ALL the information. I can only guess an update made it impossible. I can do something like the old picture using the old picture + ControlNet and Prompt, Seed, Sampler, etc. ... But maybe you are a lucky person?

[candid surge]

just cheat and use img2img. thats what I do if SD3 refuses to get some poses or bodies right. I then use sd15 or sdxl/pony to make the 1st image, then img2img with sd3

[languid pebble]

A missing or new XFormers can also make a difference ...

[solemn venture]

I don't really know how to use img2img

Should i just put the img and copy the prompt or something?

[candid surge]

what method are you using for image generation?

[solemn venture]

I always use txt2img + seed + Controlnet

[candid surge]

what software?

[solemn venture]

Forge

Not updated if that's your concern

[candid surge]

never used forge, no idea

just google for a forge img2img howto or tutorial

[solemn venture]

Forge it's the same as a1111 i believe, i used both and they look similar to my noob eyes

[solemn venture]

I'll do that, thanks

[earnest cargo]

honestly I didn't expect sdxl to be that good in knowing it's artists

good thing I found a nice database to test the artists it knows

[languid pebble]

How would you treat cheaters?

[crisp stream]

https://stablediffusion.fr/artist-style

[rugged dagger]

This was my first real attempt at AI Music. So much fun! https://www.youtube.com/watch?v=b8j1Ao1knNU

What do you guys think? Billboard top 10? 🤔

[earnest cargo]

interesting, I was using this one https://huggingface.co/spaces/terrariyum/SDXL-artists-browser

[crisp stream]

there are several 🙂

[earnest cargo]

well more is good :kekw:

tks

[crisp stream]

w

[languid pebble]

Good morning coffee

[nimble mason]

that's awesome

[languid pebble]

Wish you a good day!

[clever oar]

you too

[nimble mason]

me makin coffee

[languid pebble]

Don't roast it to much ^^

[clever oar]

what you wanna drink?

[languid pebble]

Lifeboat ...

[languid pebble]

Had the same problems 🙂

[wispy nest]

koala bird is amazing

better than real birds

[celest sigil]

Did you run out of colour?! 🤣

[rugged dagger]

Made another AI music thingy mcthinger. This is like crack. The cover images are both SDXL. https://www.youtube.com/watch?v=4thm5oSKiR8

Thank you @devs for all you do. We appreciate you!

[candid surge]

sd15

[languid pebble]

1.5?

[candid surge]

yes

[languid pebble]

Still good ...

[candid surge]

somewhat later sd1.5 models are a little better at resolution up to maybe 1024, at least that's what I do

[languid pebble]

I've started with A.I. more than 2 years ago ... there are a few of my first pictures I really love!

[candid surge]

cant take away the emotional historical aspect of it!

[languid pebble]

I redo some of them ... you can say the redos are better ... but the old ones are a bit ... dunno how to say that ?dirty? ... that's a missing thing with the latest models ...

[royal monolith]

gritty

[languid pebble]

I kept some old pictures as an "Mojo learns A.I." galery ...

[candid surge]

I've justed posted this old merged sd1.5 model. There is also the fractal lora to go with it. Check it out at https://civitai.com/models/569901

[languid pebble]

I've found the page ,,, stupid discord 😄

[eager wasp]

Kolors stuff again

[jovial tiger]

Ella/kolors

[eager wasp]

Very brave ghost

[lusty apex]

Fox

[eager wasp]

More spooky 👻

[wild sorrel]

Ghost of Tsushima + Hollow knight

[nimble mason]

looks like it's casting a hex at a halloween party

[lavish garden]

Prompt: https://civitai.com/images/19238270

"The Meeting" (1/2)

[trail lotus]

Fluently XL + LoRAs using HuggingFace diffusers + custom Python code https://github.com/InServiceOfX/InServiceOfX/blob/master/PythonLibraries/HuggingFace/MoreDiffusers/morediffusers/Applications/terminal_only_finite_loop_main_with_loras.py

[burnt wren]

un portero de guantes amarillos con camisa y pantalon de la seleccion colombia, pelo marron y ojos marrones, y el traje blanco y naranja claro

[languid pebble]

Good morning coffee

[languid pebble]

I'm blue dabedi dabedei ... sing

[languid pebble]

Death is a challenge ...

[languid pebble]

A BMW? ^^

[nimble mason]

"very high quality, sharp, detailed,the dark theme, rustic cottage beside a reflective river under a dramatic sky, standing on a beach under a bright sun, car with swirling colors, bright colors and sharp lines"

[languid pebble]

Bad reflection in my case

[celest sigil]

Why do you always say "Good morning" to your coffee? 😄

[languid pebble]

Good morning to you! It's our name for the first coffee of the day 🙂

[frigid radish]

Is there anything like this: https://badnoise.net/latentcoupleregionmapper/ but for regional prompter? I feel like i'm going insane trying to divide the ratio up as I want it 🥲

[nimble mason]

There's something so damn cheery about it that even though I typically see that msg and image on 3am I wanna get up and chug a whole pot and start my day

[languid pebble]

Something we started in a German channel ... starting the day with a "Good morning coffee" picture ... I am the only one left doing that as kinda experiment to see when I'm running out of prompts/ideas 🙂

[glossy gyro]

do people make a new env every time and save a requirementd.txt using stable diffusion every time or just have a main location for it ? i want to know how much space yu are dedicating to SD

[unique condor]

hey, we're almost twinsies

and i only use one venv for comfy

[blissful vale]

/water colors

[quiet current]

I'd say one venv per client (A1111, sdnext, forge) is good practice

also if you're curious

[wispy nest]

kolors

kolors

kolors

[gentle blaze]

Testing Kolors 👀