🏞｜general-with-images

[unkempt spindle]

tell me about ut, video editing is still alot of work. I sat about 10 hours with a project I started just for fun, mostly cause it is shocking how similar they are, they straight up stole from eachother. Remixing 3 different artist into a single song and a single video, with sound and footage from all 3.

but yeah, Im not doing that again

[languid pebble]

There's a point you have to say: from here I can't do it professional alone any longer ... but I've already won competitions with my videos ... I bet because nobody wants to do videos 😄

Without A.I. creating a story, song and video takes at least 10 people? To this point I was able to do it alone ...

With the new video generators you can do scenes longer than 10 Seconds so you would need more video editing skills ...

[unkempt spindle]

AI video is great if you can use it right, I guess you got to stylize it correctly, its incredible dull with images and footage that obviously are AI made if it tries to be too realistic I think

https://www.youtube.com/shorts/IqNDaQPLHDI This is still the best AI video on internet atm 😄

[languid pebble]

It can be pretty realistic nowadays, too. But needs a lot of trials ...

[unkempt spindle]

show me a video with Will Smith eating spaghetti properly and Im convinced

[languid pebble]

For what kinda video would you need that? A.I. has no idea what eating is ... I think it's damn hard to train ...

At the end it's a question of money ...

[unkempt spindle]

yeah, its more about showing AI video generation and what flaws it got. It has problems with hands, people running and will smith eating pasta

[languid pebble]

I wouldn't try to use Will Smith ...because of his personal rights ... even though celebs can give some consistance ...

[unkempt spindle]

No I wouldnt, I think that was were it all started with AI video...there was alot of videos of him eating pasta..which was funny at the time 🙂

[languid pebble]

Or Video2Video for MC Donalds commercials ... looking pretty strange cause the A.I. doesn't know what eating is 🙂

[unkempt spindle]

its still pretty funny tho https://www.youtube.com/watch?v=UQmgKIWFnHc

[languid pebble]

Yes ... it's a complete new game and we have to learn how to play it ... and some weeks later we don't like what we did any longer cause there are new technologies ... but a few old pieces we will still love ...

I think it makes a difference why you create a picture. You just wanna create something cool ... there you go. But maybe you really have a special idea or concept ... I think that will survive new technologies ...

[royal monolith]

he's totally bought into the eating pasta thing to the point of making his own version. i don't think he'd care

[languid pebble]

I'd hope he has enough of humor ... but first A.I. real commercials with his face ... ... ...

[unkempt spindle]

Kling, the chinese company actual got pretty close to someone eating spaghetti, but it sure aint will smith https://www.youtube.com/watch?v=pkXyuaV_2XU

[languid pebble]

I'd say it's A.I.

But maybe worth the joke ...

Strange beared ,,,

[royal monolith]

if someone's making money off his face, he's going to want a cut

[languid pebble]

They can already sell it ...

[royal monolith]

https://arstechnica.com/information-technology/2024/02/will-smith-parodies-viral-ai-generated-video-by-actually-eating-spaghetti/

[languid pebble]

The actor of Darth Vaders voice has sold his voice ...

[royal monolith]

he's smart

[languid pebble]

And we will keep the voice ...

Can be a win-win-situation ...

[languid pebble]

Nini!

[late sorrel]

lol

[hearty violet]

@languid pebble hi, did you try reforge?

[languid pebble]

I've installed it a few days ago

[hearty violet]

and?

[languid pebble]

Looks good, worked fine ... AFAIK without SD3 support yet

[hearty violet]

good ^^

did you use sd3, is good? people was critcizing it a lot must I imagine that it's good in some way

[languid pebble]

ReForge ...

[hearty violet]

good shoot

[languid pebble]

SD3 ... hmm ... you can play with it and it might do some things good. But it's not really a model to work with yet ... needs finetune by community and SAI has been talking about a better version they work on ...

[hearty violet]

understood

[languid pebble]

It's not to hard to play with it using ComfyUI ... we've installed more complex models still in work here 🙂

If you want to try SD3 without installation: https://glif.app/@Oliveira/glifs/clw44qfbl0000m0zztwqk2tnf

[languid pebble]

Egypt is cool 😍

[languid pebble]

Even though the Sphinx is smaller than expected 🙂

[radiant saddle]

Using ICLight and IPAdapter for better integration of masked objects in generated backgrounds. Rendered with SD 1.5 + SDXL refiner

[languid pebble]

My Pharaos die cause of broken neck ^^

[cold rock]

hi

[hearty violet]

hi guys

exista CosXL to automatic 1111?

[cold rock]

Hi

How to use prompts and where to use?

[languid pebble]

Sounds like CosXL is a model and for sure you can download and copy it into A1111

[cold rock]

Hi how to use prompts and where to use can you explain?

[languid pebble]

If you wanna prompt in Discord you will need a subscription. There are a few free webservices you can use like https://www.craiyon.com/

If you want to use here you can check: https://discord.com/channels/1002292111942635562/1237461679286128730

[cold rock]

Oh thanks bro

[languid pebble]

Enjoy!

[hearty violet]

ahhhh I understand, thanks!

[maiden storm]

Any suggestions for a ControlNet that can provide consistency between new images / the source image?

I've been playing around with img2img, I basically create video's by rendering every frame and then using ffmpeg to put it back together.

The problem is that there's too much "changes" between each picture.

Any suggestions for a controlnet to achieve consistency? Attached is a WIP (audio plays FYI!)

[quiet current]

animatediff, mov2mov, deforum, etc there are many tools that are aiming for temporal coherency

[maiden storm]

Yeah, but any specific controlnet that's used with those tools? Tried temporalnet but did not achieve consistent results

[quiet current]

Controlnet by itself won t help much with temporal coherency.

you need something else on top of it to feed it the previous frame(s) and constraint the whole thing.

[maiden storm]

Ahhhh yeah 😦 I tried using animatediff but then that tips the iceberg in terms of vram usage. I was hoping for a controlnet that somehow uses keyframe compositing

[wispy nest]

this is not doable with just image models

your video is already better than most attempts because of the control net

[robust path]

DrawThings has Kolors now 🥳

[maiden storm]

@quiet current I'm using img2img batch, I'm curious if you happen to know a way in a1111 to basically use the last generated image + denoising set to low, which could possibly achieve similar results / consistency? Ex. generate one image, the next image looks at the previous image and just builds on that, etc.

[quiet current]

To my knowledge it can t be done out of the box. img2img workflow is "one image in, one image out". If you want temporal coherency you need "multiple images in, multiple images out" or '''one image in+ motion vectors/optical flow, one image out'' or something alike.
You always will need something that can get informations from the past/future and something to """"""smooth""""""" your outputs (eg : ebsynth).

img2img is not made for that, you need extensions to get proper results

[maiden storm]

thank you for helping so much, yeah 😦

problem is running out of vram so been just generating frame by frame and stitching it back together afterwards.

i'm going to explore a bit more than give up and switch to something like animatediff (again) or mov2mov, etc. the biggest constraint is me trying to do it with SDXL rather than 1.5 like other people

[rain wind]

https://www.fotor.com/referrer/s6darmhg?location=editor_unlock_download

[unique condor]

hi i have problem inpaint. i try make bobs big but it work not. how maketh i bobs big???

[gentle blaze]

I believe the test has been successfully passed

[languid pebble]

Purrfect picture!

[pallid ruin]

Okey DMD2 is my favourite model now

[languid pebble]

Is there a comfy Workflow?

Ohhh... I see ...

[pallid ruin]

The workflow is in the image, you can drop it in your comfyui window

[languid pebble]

Thanks .... Asked to fast 🙂

I need a second 4TB SSD 😄

[royal monolith]

at this point, just create your own private discord and start storing your images in there

[languid pebble]

It's more about the models 😄

And I'm looking for kinda document management A.I. ...

Idea is to scan all my important documents and let it fill out all forms for me ...

[stoic lark]

DMD2? Is that the actual name? What build is it for?

[languid pebble]

https://gist.github.com/comfyanonymous/fcce4ced378f74f4c46026b134faf27a

I hope this is the right one 🙂

[stoic lark]

Noice. Thanky.

[jovial tiger]

Does that work with existing sdxl or a whole new model?

[languid pebble]

It works with SDXL ...

If you add a lot of steps pictures become abstract like my heart queen above

Workflow should be included there, too

[wispy nest]

its distilled

a bit like distilbert

[languid pebble]

I bet you are the SAI CEO or something like that 😛

[wispy nest]

lol

[pallid ruin]

In my opinion is a very good model, I hope it would be a way to train it for more than 4 steps

[languid pebble]

You can use more steps to get more ... stuff like my heart queen above ...

[languid pebble]

😛

[stoic lark]

Sooo... Is A11 officially dead? Baby don't like change and still hasn't learnt comfy.

[languid pebble]

There's a ReForge ... using old Forge + A1111 updates ...

Worked fine here in my few tests ...

[stoic lark]

Everything intact? Regional, tilevae, tiledif, etc.?

[languid pebble]

Haven't tested everything but should be worth a try ...

[wispy nest]

I learnt comfy in the last week from scratch
I would recommend it, it was not hard

[languid pebble]

Yes ... Comfy is something to learn ... you will understand A.I. much more ... if you are younger than me 😄

[wispy nest]

the big thing is you can just copy people's workflows from their images

[stoic lark]

Yoo kan hwat?

Really?

[wispy nest]

yeah if they made their image with comfy

you can drag it into comfy

and you get their workflow

[languid pebble]

Workflow included in every picture, if not removed ... load it into your webbrowser ans save it from there

[stoic lark]

Goddammit, fine. To the comfymobile.

[wispy nest]

I went through like 200 workflows at the start of the week
took from reddit, discord and youtube

what I learnt is that everyone kinda does the same thing

there are not many people doing massive scary workflows

[languid pebble]

And many forget to add a save picture node 😄

[wispy nest]

generally people do something like this:
PAG, FreeU, Deepshrink, with an IP adapter and canny and depth control net, into a tiled upscale

and that's about as complex as it gets

[stoic lark]

Comfy was a hella lot easier to install than A11. I hope that's an indication of its ease of use.

[stoic lark]

https://tenor.com/view/anchorman-baxter-spanish-willferrell-will-gif-4923325

[languid pebble]

It will be hard cause it's a complete different work than with A1111 ...

[stoic lark]

I have a bg in modular synthesis, so I keep thinking I should be good. Just haven't taken the plunge, yet.

[wispy nest]

funnily enough it kinda looks like a synth

[languid pebble]

I still don't know how to add a good upscale to others workflows ...

[stoic lark]

Ewph. How do I specify a custom models folder. Don't wanna make dups of my A11 collection.

[wispy nest]

upscale is tricky because there are different options

[stoic lark]

I figured that out last time I used it.

[wispy nest]

I like this upscaler a lot https://openmodeldb.info/models/4x-RealWebPhoto-v4-dat2

but a lot of people use this https://github.com/ssitu/ComfyUI_UltimateSDUpscale

[languid pebble]

There's a extra_model_paths.yaml.example in your Comfy Folder

[stoic lark]

Muchos grazi, senpai.

[languid pebble]

Remove the .example after you are done ...

[wispy nest]

is that AI or is that a real photo

[stoic lark]

nahanahanahanah nah nah nahhhh nahanahnah nah...nah.. SPEGHETTINAUT!

[pallid ruin]

Its the DMD2 model on top of leosam´shelloworld sdxl model

[stoic lark]

No Sabbath fans?

[wispy nest]

helloworld is great

[pallid ruin]

F its still overtrained in a few things

If you write "astronaut on the moon" it will try to do that photo in the right

But I think that happens to most models

[jovial tiger]

wow that's pretty quick and looks good too. using mobius' clip

[languid pebble]

I think it's been LeoB who found it ...

[pallid ruin]

yee

[languid pebble]

Didn't want to steel the honor for it 🙂

[wispy nest]

took this workflow from a youtuber
its noise injection

[languid pebble]

WTH ... something happening to my X ^^

[wispy nest]

wow nice

I like the details like the neck

[languid pebble]

Needs protection I think 🙂

Human-Machine ... one of my topics ...

Where does humanity end and machine start? Is there still humanity? Or does it just look like it?

It's to late for me now. Thanks for all your support and the fun time we share ❤️ Good night!

[wispy nest]

❤️

[unique condor]

we human beings cant prove we aren't already machines. if put to the turing test, on a long enough time line, a pattern might emerge. that would mean you're already a machine with determinism. when you break us down to the lowest levels, neurons by themselves have no intelligence. it's only when they form extremely complex and dense networks that a sort of intelligence can arise. sounds extremely similar to certain things we've been doing with AI... granted, it's not a true 1:1 analog, still a whole ton of other differences like our brains don't train the networks using back propagation and the data isn't incrementally frozen, etc etc

[languid pebble]

When I started to work with A.I. I have been thinking about it. I have made a picture about that topic for my exhibition. THB I think the name A.I. is pretty stupid. Cause A.I. is pretty stupid. It can only do one thing and you have to combine them to do more. As long as the A.I. can't have a body like us and make the experiences ... it will be stupid ...

[unique condor]

well all our current "AIs" are right now are just glorified versions of google search, so i wouldn't worry about skynet yet. AGI, however, is a whole different story

like even with image diffusion, shits just averaging a bunch of stuff from the datasets that happen to be associated with concepts. it doesn't think, it doesn't know what it's doing, it doesn't know what it's making, it's just a glorified spreadsheet and really good at mixing values together

[languid pebble]

A.I. is still just a marketing gag ...

We will put different A.I.s together to get a "better" one ... but it still doesn't know what reality is ...

[hearty violet]

hi, what are the best way to transform it

on it?

[languid pebble]

Contro9lNet + Prompt? No idea ...

[hearty violet]

yeah rs

I thought in I2T + controlnet

I think I have to try, right rs

ok let me try it, thanks mojo 🙂

IP-Adapter too

[languid pebble]

I'm not really a pro using it ... but sounds you are getting better ideas than me.

[split elk]

👋 .

[hearty violet]

thanks, I will try and show results

[languid pebble]

Looking forward to them! Good luck!

Good nite!

[hearty violet]

good night : )

[clever oar]

🖐️

[hearty violet]

its working, Omost helpinga lot, but this green vibe...

[wispy nest]

SDXL has a bias towards yellow if I remember rightly

[hearty violet]

man, its 1.5... I will do with SDXL! : D

see the prompt... I used chatgpt to see the image and creat a promp to ask Omost to creat the prompt to SD

with kaggle free I cant run with XL is so tight the space

I will try on colab

[hearty violet]

I love how Omost construct the promp, I intende construct my prompts like that

[wispy nest]

isnt that python?

[hearty violet]

who teach me it help me a lot, is very importnt to me that ability to direction the atention

I think yes

[wispy nest]

yeah python is really nice

works as a prompt too

[hearty violet]

man, this is so cool... I'm learning to much with SD and python : D

[hearty violet]

better, I think is the VAE, was none I put automatically

so cool ^^

my SDXL rsss idk why

[hearty violet]

It's possible on A1111 I use more
Then one inpaint with others controlnet in one shot?

For example, Inpaint for the threes and controlnet scribble, while generate the image with controlnet Lineart and depth, on I2I... Make sense?

Using this image

And other Inpaint with other controlnet for example, all of this in one shoot

[blissful oasis]

What part of "simple white background" do you not understand 😂 the white background part, that is!
Guess it's because people use transparents to train some Lora models and knowing how sometimes the transparent displays a green background on desktop...

[languid pebble]

Good morning 🙂

[robust path]

Morning, segmentation is fun 🙂

[languid pebble]

Pretty cool!

[robust path]

Thanks! It's based on this extremely simple drawing I made. Picking the right colors was the hardest part. 😁

[languid pebble]

wOw! Good work! Far away from what I am doing but maybe I should have a look on that!

What model did you use?

[robust path]

Btw it's not regional prompting, it's the prompting woth the colors which are defining the subjects. Forgot what it's called, have to look it up.
Model is Dreamshaper V8 (1.5)

Oh wait, I did mention the right name, segmentation LOL

[languid pebble]

Ahh... thanks. I'd bet it's to much to prompt it with SD3 ...

[robust path]

Not possible, you need the color tables and the segmentation lora

[languid pebble]

Cat, Box, Dog and background should be possible but I think the rest would be to much...

[robust path]

@languid pebble https://github.com/Mikubill/sd-webui-controlnet/discussions/503

[languid pebble]

Thank you!

[robust path]

I meant the controlnet, not the lora

[unkempt spindle]

https://www.youtube.com/watch?v=iqBa7V4nkPE Lumalabs+Wiggle+SD

[languid pebble]

I love the crazy and creative things we can do with A.I.! ❤️

[celest sigil]

You may not want to eat some of these plants 😄

[languid pebble]

None of em ... 😂

[vagrant dust]

simple eat or be eaten

[radiant saddle]

Before the storm

[clever oar]

lost tail 😃

[languid pebble]

Not his lucky day 😄

[clever oar]

sacrifices for career success😎

[hearty violet]

my god so good!!! 😍 🤩 how did you do this? I'm turn on automatic 1111 on colab to se something with PNG info

[hearty violet]

I can't see nothing on PGN info ; (

[wispy nest]

the guy in blue and orange is amazing

is this kolors

I recognise a little bit the "palace" or "mansion" token or something similar

[willow chasm]

Create a vibrant and detailed festival poster with the following elements:

1. Background- Dominant color: Deep red with a subtle pattern resembling woven straw or traditional textile. Large bold text at the top: "FESTIVAL" in dark red with a slight shadow effect for depth.
   2.Main Character- Central figure: A smiling farmer wearing a traditional Vietnamese conical hat (nón lá) and an orange-brown traditional outfit.

   • Position: Slightly tilted forward with arms crossed, holding a bundle of golden rice stalks. 3. Supporting Elements:- Surrounding the main character: Abundant golden rice stalks with detailed grains and leaves.
   • Green swallows with dynamic wings, positioned to the left and right of the character.- Geometric shapes: A bar chart with upward trending green bars and a cogwheel, both integrated into the design around the character’s feet. 4. Text and Labels-
     5Details and Style:
   • Art style: Vibrant and semi-realistic with a touch of traditional Vietnamese art.
   • Colors: Rich, warm tones with bright highlights to emphasize the festive and agricultural theme.
   • Lighting: Soft, natural lighting to enhance the realism of the illustration.
   • Texture: Slight grain to simulate a printed poster.

2. Additional Elements:

   • Behind the character: A stylized mountain range in green and yellow hues, blending into the background.
   • Foreground elements: Parts of a gear and a foot wearing a traditional Vietnamese slipper, partially covered by rice stalks.

Message# Create a vibrant and detailed festival poster with the following elements:

1. Background- Dominant color: Deep red with a subtle pattern resembling woven straw or traditional textile. Large bold text at the top: "FESTIVAL" in dark red with a slight shadow effect for depth.
   2.Main Character- Central figure: A smiling farmer wearing a traditional Vietnamese conical hat (nón lá) and an orange-brown traditional outfit.

   • Position: Slightly tilted forward with arms crossed, holding a bundle of golden rice stalks. 3. Supporting Elements:- Surrounding the main character: Abundant golden rice stalks with detailed grains and leaves.
   • Green swallows with dynamic wings, positioned to the left and right of the character.- Geometric shapes: A bar chart with upward trending green bars and a cogwheel, both integrated into the design around the character’s feet. 4. Text and Labels-
     5Details and Style:
   • Art style: Vibrant and semi-realistic with a touch of traditional Vietnamese art.
   • Colors: Rich, warm tones with bright highlights to emphasize the festive and agricultural theme.
   • Lighting: Soft, natural lighting to enhance the realism of the illustration.
   • Texture: Slight grain to simulate a printed poster.

2. Additional Elements:

   • Behind the character: A stylized mountain range in green and yellow hues, blending into the background.
   • Foreground elements: Parts of a gear and a foot wearing a traditional Vietnamese slipper, partially covered by rice stalks.

[jovial tiger]

Yeah it's kolors. Apparently there's a Chinese website that shows off all the style prompts that work with it. This one was: haute couture, a man stuffing his face with Cheetos , dynamic, dramatic, haute couture, elegant, gorgeous clothing, haute couture

[robust path]

What's the url?

Kolors

[jovial tiger]

https://modelscope-cn.translate.goog/datasets/modelscope/Kolors_awesome_prompts/dataPeview?_x_tr_sl=zh-CN&_x_tr_tl=en&_x_tr_hl=en&_x_tr_pto=sc&_x_tr_hist=true

[robust path]

Thanks! 👍

[eager wasp]

Wow, that's great! Prob works best when the style/prompt for these is in chinese 😓 Wonder how it'll look in english....

[sleek burrow]

Hi guys, would anyone be able to tell me how these types of images can be generated? any advice would help.

[pallid ruin]

Epic

[pallid ruin]

It works in both languages, but in Chinese it's better, at least in the comparisons I saw

[hearty violet]

checkpoints and loras with the syle, a properly point, controlnet can be used too

[eager wasp]

Often it depends, sometimes a prompt works better in english (not that often) sometimes Chinese, and for added fun it depends on what you use to translate with 🤡

After looking at the dataset, these are just a LOT of sdxl(??) templates tried in kolors, not kolors specific ways to prompt, doesn't make it any less of a nice resource. (and you can even prompt the style in chinese and the subject in english, lol)

[languid pebble]

An easy pro solution! Thanks for showing!

Bad luck @hearty violet it's comfy 😄

[hearty violet]

bad luky? this r_Mada is my savior!

I make 2 course of comfy but only don't practice yet

[languid pebble]

Click his picture ... open in browser and save it ... you should be able to drop it into your comfy afterwards ...

[hearty violet]

@gentle blaze thanks a lot! I will try replicate your work. Can I ask wich checkpoint did you used? I see that you trained your own checpoints, do you suggetion one of then to architecure?

[hearty violet]

really? I tried on A11 but dont show anything, I will do it thanks Mojo!

[languid pebble]

it should load the workflow ... but show/save from browser and not discord

[hearty violet]

ok I understand 👌

[hearty violet]

Mojo can you help me? I salve from browser (edge) but when I put on comfy dont show anything

[vagrant dust]

You could try to use chatgpt to get prompt which are close to these images and then find a good anime model. Without an anime model you get images dominates by the main colors:

[vagrant dust]

Because there is not workflow within the image. But the image shows the Workflow nearly in total. But it requires a few additional nodes and models

[languid pebble]

Looks like it doesn't work here either ...

Looks like you have to rebuild ...

[hearty violet]

ok I if is it I think I can replicate it, I will try now and post here

[languid pebble]

Good luck!

[hearty violet]

thanks friend 🙂

[crisp stream]

no input image

[robust path]

looks cool, but wdym?

[languid pebble]

And thanks for sharing your problem. Learned a lot!

[crisp stream]

the relation to your drawing

[hearty violet]

you are welcome 🙂 I kow I learn a lot in this way to, so it's is because I like to share, be welcome

VAE and CLIP SKIP is according of the model? like how the model was trained?

[robust path]

sorry, still don't get it 😅
what did you do?

[languid pebble]

If there's not specific information in the model card I'd use Automatic and Clip skip 1

[crisp stream]

was simply an image generated before going to bed. Resembling a bit the one you posted strutcture-wise with more plain colours

moe like composition-wise

a bit 😉

[languid pebble]

🥳 🤣

[robust path]

Oh I see. My image was meant as a template for Segmentation controlnet, the colours define the subjects, so you have to use exactly the right ones (hex codes). That's why I mentioned it was the hardest part to do 😁
Cool that is inspired you though ❤️

[vagrant dust]

So far so good. Biggest problem for the "simple" workflow is the trees. Both Canny Edge or Lineart Controlnets are sure the lines within the trees need to stay there...

[robust path]

I think I'm in love with shoebills 😊

The fantasy kind

[hearty violet]

I think is canny that do this.... I'm using lineart and deptt (leres+) and dont get this risks. But no like r_Mada yes rss

[hearty violet]

better

[vagrant dust]

Guess my first Style input was not great at all. To colorful, needed a more neutral style image as input

[sterile kiln]

faster stronger

[hearty violet]

maybe low CFG weight? look so dark

[sterile kiln]

https://www.youtube.com/watch?v=gAjR4_CbPpQ

[hearty violet]

man I cant download one SDXL to use, I'm on Kaggle and it broke with my civit ai api and idk why to fix it.... I wanna try this with XL so much

[robust path]

Bauhaus Batman

[crisp stream]

yes, I know 🙂 Then it at least was generated before your post, still made a conncection 🙂

[eager ice]

An AR work with a specific scene in a museum with an escalator extending from the ground to the roof, photo grade, UHD

[pallid ruin]

Not perfect but here it is on sd3, I gotta try in sdxl+dmd2

[languid pebble]

Found that under your sofa?

[sterile kiln]

does anyone know of a recent openpose model for SDXL that's more efficient than what we've known since?

[fleet swift]

#Gyomei Himejima dnd style half-orc in anime style blade cutting demons. High resolution.

fights with two axes and a chain. there are flames around

[nimble mason]

threw that into a small model i'm working on that operates in SD3 latent space

[pallid ruin]

Nice! 🤗 🥺

[nimble mason]

thx!

this is with barely any training at all... right now it's just crappy code in a jupyter notebook

gotta clean this up but i'll release it at some point

[pallid ruin]

I´ll be waiting for it!

[nimble mason]

lighting on that last one is espec great, reminds me of the excitement of an incoming wedge tornado as a kid

[nimble mason]

wow

[nimble mason]

lol that tail 😄

[pallid ruin]

btw, do you have the prompt for this image? ngl its one of the best ones I´ve ever seen using open source AIs

[unique condor]

how i fix hair right??? evry time it no me hears when i says curly bangs? wut i do fix it???? plz respon

[robust path]

Haha you noticed! 😁

[jovial tiger]

Sure, it's just kolors without any other refiner. Prompt is: haute couture, a man holding an assault rifle, dynamic, dramatic, haute couture, elegant, gorgeous clothing, haute couture

[rain gazelle]

I like this one 🙂

[nimble mason]

wonder if anyone else has also trained a model that runs in sd3 latent space without sd3

this is working pretty well

left is an old sdxl image of mine, right is it refined being passed through this model... takes about 2-3 seconds

[next hull]

\ astronaut in alien planet

[gentle blaze]

DMD2

[robust path]

Thanks 🙂

[languid pebble]

Final result:

[unique condor]

this failure really caught me off guard, but i still like it for the absurdity

[hearty violet]

thanks r_Mada it will help me a lot! 😍 🤗

[wispy nest]

Can someone please help me with this error: "Torch is not able to use GPU"
I have an amd gpu (rx 6800)
Ive folowed this tutorial but my problem still isnt fixed
https://www.stablediffusiontutorials.com/2024/03/stable-diffusion-error-amd.html

[hearty violet]

where areyou tryng use it, locally?

[wispy nest]

ive put the folder in documents

[hearty violet]

are you running SD locally so?

[wispy nest]

I think so

[hearty violet]

do you have nvidia or amd gpu?

[wispy nest]

amd 6000s series

[hearty violet]

must be it

idk so much about it but wait

[wispy nest]

okay

[hearty violet]

maybe it help?
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs

are you tryng A11 or comfy?

[wispy nest]

a1111

[hearty violet]

see if it help you 🙂

[wispy nest]

ill look into it
thx m8

[clever oar]

hazard where you egg icon

[hearty violet]

I changed it 🔃

XD

[clever oar]

the brain doesn't like change😎

[hearty violet]

haushuashuah

[clever oar]

back it

i like stability

😠

[hearty violet]

https://tenor.com/view/marcos-bomba-humo-magic-disappear-gif-11061911

[hearty violet]

put a icon for you, I wanna see

@gentle blaze the first shoot with your workflow more simple and adapt to A11, better image until now 😍 🤩

I'm using your checkpoint RMSDXL Hybrid Turbo

it's not possible I use the bursh to inpaint a are to controlnet use only this area?

I used brush here with scribble

but don't work

[clever oar]

dont understan you change it back?

[clever oar]

what you wanna do with this image?

[hearty violet]

no I'm asking you to change yours

change the discord logo

[clever oar]

i dont want change beacouse its stablitity

[hearty violet]

on controlnet scribble I use inpaint thinking that controlnet will use only the inpaint area, but don't work

[hearty violet]

stabilityAI? XD

[clever oar]

yep))

bad eng

[hearty violet]

sure, I know

[clever oar]

not my lang

[hearty violet]

you are a moderator

not mine too

where are you from?

[clever oar]

russia

[clever oar]

so you can't do it because you changed the icon

you know that a111 link to discord?

if you change something in discord you change in a1111

ai know if you do something wrongXD

[hearty violet]

I tried use mask but don't work too

[hearty violet]

cool, I'm from Brasil

I wanna realistic trees not this circles, I will try inpaint it

[clever oar]

girls in bikini mm

[hearty violet]

I like too

[clever oar]

what do scrible?

[hearty violet]

the russian are beautiful too

scrible is more creativity

[clever oar]

you can change cn on canny?

[hearty violet]

canny is more fidelity

was to scrieal make a three with more creativity

[clever oar]

with promt?

[hearty violet]

yes

[clever oar]

hm i try for you something

but i not have scrible

[hearty violet]

will be so good if I can use mask and prompt with each controlnet separarely

[clever oar]

its work without inpaint?

[hearty violet]

wai

sure

[clever oar]

can you show inpaint settings

[hearty violet]

#1080668565876785256 message

take it here

I'm on I2I but I wil l show you

[clever oar]

cant see

[hearty violet]

https://github.com/lllyasviel/stable-diffusion-webui-forge/wiki/ControlNet-model-download

[clever oar]

its inpaint sketch i not use it

[hearty violet]

is writed forge but have all model there

[hearty violet]

but you asked it XD

[clever oar]

oh

sorry

i dont use sketch XD

[hearty violet]

I'm on Imag2Img I will show you

[clever oar]

inpaint sketch i use

[cunning onyx]

Delightful snippets of information https://www.youtube.com/channel/UCdnPl9sin9oshcLnxdtLYuA?sub_confirmation=1

[clever oar]

its work like that without brush?

[hearty violet]

sure, It's i2i not inpaint

you work with image there

[clever oar]

give me promt

[hearty violet]

I used Omost

and edit it

https://huggingface.co/spaces/lllyasviel/Omost

did you understand?

[clever oar]

no i think)

i know for inpaint need brush

and you select area

[hearty violet]

yes rss

[clever oar]

on your sreen without brush

[hearty violet]

its because it's not inpaint is i2i, use the image totally

it's like I inpaint all

[clever oar]

ok

you spent so much effort on this, are you sure there are no other ways?

[hearty violet]

I think is the better way, It's not difficult it's pleaure to me do it

help me a lot, I'm architect and it's a powerfull tool

[radiant saddle]

Thanks, yeah took some tries to get that tone before it rains look

[clever oar]

cant check it normal i have 4 gb vram

[nimble mason]

i love surreal lighting like that

[nimble mason]

we need to somehow get you a real card so you can generate frog beasts at 4k resolution

[hearty violet]

me too, I use Kaggle and Colab

[wispy nest]

you can rent 24GB vram GPUs for sometimes as low as 0.20 dollars

for an hour

[hearty violet]

kaggle gives 32vram 30hrs by week free

[wispy nest]

you're not actually meant to use those for stable diffusion really

[hearty violet]

why?

[wispy nest]

the reason Google provides the free tiers of Kaggle and Colab
is so that people who don't have GPUs can still do machine learning training and learning

[clever oar]

your promt in line art

dont know it change something XD

[hearty violet]

are you kidding me?

[wispy nest]

the guy who runs it is on reddit you can go see his comments if you want

[hearty violet]

I ask on kaggle server and they told me that is for it, I can

[wispy nest]

was this Google staff or just discord users?

[hearty violet]

user I think

well I do the possible to use Kaggle and not be banned before starting use SD, make question on server, on kaggle forum and see the politics

its very powerfull plataform, and I intend traine models there in the future

[clever oar]

hm with your image my lineart dont give effect like before

night in desert😎

[gentle blaze]

dmd2 + sdxl

[languid pebble]

Looks like it's good for grumpy grandmas 😄

[gentle blaze]

That combination only makes grumpy grandmothers. The prompt is: happy Hello Kitty in a ball pit

[candid surge]

Hi chat

[candid surge]

"happy Hello Kitty in a ball pit"

[pallid ruin]

😁 dmd2 + cinematicredmond (same prompt)

Not bad, just added "iphone photo"

[languid pebble]

Looks like sugar wonderland

[pallid ruin]

idk how it works that good when its just 4 steps

I wish it could work with non lcm at 20 steps

[languid pebble]

Shut up, A.I. You are drunk ^^

Biden is out?

[candid surge]

can we prompt that?

wait, better not share well known people I guess

[languid pebble]

first ultrapixel test ...

[languid pebble]

second ...

[languid pebble]

Got a problem with your hand? I bet these are not SD3 images 😄

[nocturne oak]

Well...after the illness and the hand fell off, I had to get a replacement.

[languid pebble]

SD3 would have given you a replacement with 7 fingers. Pretty good for playing piano! 😄

ultrapixel last try 🙂

Pretty cool for no upscaling ^^

[languid pebble]

Didn't Barbarossa have a replacement hand? So you are in a "good" companie ...

[nocturne oak]

He might've still had his hand if he'd have been more aware of the rules guidelines.

[languid pebble]

I'm stupid ... so I'm that much into that history ...

[nimble mason]

still workin on my 16 chan enhancer

[wispy nest]

getting at least some of the 16 channel VAE benefits on older non-SD3 models would be great yeah

even though the model has issues I almost exclusively use SD3 now because of the amazing VAE

[languid pebble]

Playing Pokemon Go?

[clever oar]

😃 yep

time to ramen

[languid pebble]

Yummy!

[candid surge]

It's Caitlin.

[clever idol]

https://civitai.com/models/593072
Here is my LORA for Ly.

[main tendon]

De stijil,chibi style,white background,bold lines

[nimble mason]

amazing per usual

[wispy nest]

how did you get the lion so detailed

[languid pebble]

Zavychroma XL Model and Highres Fix ... nothing special in this picture

It's more been a ReForge test 😄

[wispy nest]

ah ok I should try zavy more

[languid pebble]

It's pretty nice! Using it a while now ...

[wispy nest]

I use jugg and realvis

but they are not as contrasty

for cinematic, zavy might be better choice I think

[languid pebble]

I like Juggernaut, too.

[wispy nest]

colourful as well is a nice one

[languid pebble]

I'm waiting for the new SD3 ... even though I haven't found a good workflow yet. GLIF generations are better than my local ones ...

[wispy nest]

yeah the SD3 VAE is so good

its already the best looking images

[languid pebble]

Leonardo Phoenix is pretty good, too.

A bit what I thought to get with SD3 🙂

[wispy nest]

ah yeah I haven't tried leonardo yet

I like the letter A in watch

its cool font

[languid pebble]

I'm not really a fan of closed systems ... But Leo Phoenix can handle text and prompts pretty well. They are giving away free tokens every day ... so worth a try

[wispy nest]

yeah I don't like the closed ones either, I will try it though

[wispy nest]

I guess that was SD3?

really good fonts

[languid pebble]

That's Leonardo Phoenix ... inspired by Portal computer game ...

[wispy nest]

oh wow

ok so more than just SD3 can do good text

[languid pebble]

They also have kinda prompt improver. It can kill you prompt but more often adds cool stuff ...

Improved prompt: A distressed, retro-futuristic warning wallpaper with a worn, metallic sheen, featuring bold, red, sans-serif font that reads, "Don't listen to your future you!" in a stylized, curved line, reminiscent of the test chamber signage from Portal 2, set against a dark, gradient blue background with a subtle, circuit board-inspired pattern, evoking a sense of industrial decay and ominous warning, with faint, flickering, neon-like highlights that accentuate the text, and a subtle, high-contrast, film-grain texture that adds to the overall sense of retro-futuristic unease.

Glif SD3 picture with that prompt:

[wispy nest]

oh yeah I never like prompt improvers

[languid pebble]

This is the first one that kinda work for me ...

[wispy nest]

before noise injection:

after noise injection:

[languid pebble]

pretty cool!

[wispy nest]

thanks I love this technique

credit to Treeshark for the workflow

[languid pebble]

we will never stop learning 🙂

[wispy nest]

another cool thing is you can prompt the noise differently
to add a different theme like steampunk but only to the fine details

the elf and her clothes are the same but it swapped leaves for twisted metal and gears

[languid pebble]

I will have a closer look later. Work is calling me now! Have a good one!

[wispy nest]

bye 🙂

[celest sigil]

https://tenor.com/view/popcorn-3d-couch-potato-chill-gif-5159897

[sterile kiln]

The faces of my friend and me, in this cult scene from the film “Bernie”.

[wispy nest]

if you put underwater it turns plants into kelp and corals LOL

[desert loom]

Can we have a softer version :-)or slight noise?

[rain anchor]

/prompt black girl 14, long afro hair, in the river, with little yorkshire terrier floppy ears

[robust warren]

😆

[wispy nest]

ok this is my attempt at a softer one

[languid pebble]

Pretty complex workflow ... took me 20 minutes to get there's a second one 🤣

[lavish garden]

(Soon...)

[steep sleet]

Fooocus using Mobius Checkpoint and Ethereal Grace LoRA

[dire sun]

Atomic orbitals made of flesh

[languid pebble]

Dang third leg ...

[languid pebble]

@nimble mason Have you tried the Ultrapixel workflow?

[nimble mason]

i still haven't, incredibly... my 4090 has been completely occupied with training projects

wish like hell the new cards would come out already, i'm not gonna buy anything till those price drops hit with used stock at a minimum, but damn i need more shit lol

[languid pebble]

I was thinking how to change the cascade models in the standard workflow ...

50s series first rumors are out ...

I'll be a guest of ASUS at the GamesCon ... hope to get some information ...

[nimble mason]

yeah, who knows... maybe late this fall, maybe spring?

awesome

[languid pebble]

I'd think fall ...

[nimble mason]

either way i figure that even if the 5090 is a 28gb joke there'll be some fresh used stock of 90 series shit to buy

yeah

i'd really like to get some basement boxes set up with dual 3090s at a minimum

whatever ends up making sense with the price to processing ratio

[languid pebble]

German but you will understand the important parts: https://www.pcgameshardware.de/Grafikkarten-Grafikkarte-97980/News/Geforce-RTX-5090-mit-500-Watt-1451654/#Echobox=1720995745

[nimble mason]

lol

yeah the 28gb vram rumors... seriously, if nvidia pulls that crap after years and years of 24gb, the only difference between the 4090 and an a100 (practically speaking) being vram and an insane price gap

they really deserve to get slammed for market manipulation bs/antitrust

https://www.cnbc.com/2024/06/06/us-regulators-to-open-antitrust-probes-into-nvidia-microsoft-and-openai.html
https://www.reuters.com/technology/french-antitrust-regulators-preparing-nvidia-charges-sources-say-2024-07-01/

[languid pebble]

We can only hope the best ... but for sure ... AMD lost this point and NVIDIA wants to make money

[nimble mason]

yeah it's past the point of fair profit imo... it's pretty destructive now

setting everything back by years

[languid pebble]

But it could happen that the community leaves NVIDIA ...

[nimble mason]

want to screw me over and ask for $3k for an 80gb version of my $1700 4090? jfc, but okay. $30k though? wowww

yeah, sure hope so

it's driving me nuts with these training projects... this 4090 is easily fast enough to do anything i want to do here, but i'm always slamming into OOM problems

[languid pebble]

I haven't followed that topic much ... but it sounds they are working on solutions for AMD ...

[nimble mason]

it's like if ford motor co was pretty much the only one selling driveable cars, so they made the version with air conditioning cost twice as much lol

yeah

they need their own cuda pronto

[languid pebble]

NVIDIA has problems, too. If I understand right the last Torch comes without flash, latest Xformers don't work with the previous system ... no idea how to handle it best ...

[nimble mason]

lol

yikes

[languid pebble]

2 downgrades? There's one model I can't use cause it takes 60 Minutes/Image with my 4090

[nimble mason]

wtf, what model is that

that's insane lol

[languid pebble]

I think the model size was 12 GB ... I have to many installed ... dream-project? No idea ...

Shouldn't try all the alphas and betas ... getting confused 😄

I think it's Aura flow ....

[nimble mason]

oh, yeah, that one is kinda slow but it shouldn't be THAT slow lol

[languid pebble]

I think it's cause of missing flash attention and XFormers

[languid pebble]

@nimble mason Thanks for being here. I am always learning a lot!

[languid pebble]

Time for bed?

[clever oar]

no)

[nimble mason]

same here... my god there is an infinite amount of shit to learn lol

[languid pebble]

Neverending story .. lalala lala la lalala ... 😄

[clever oar]

time to eat

I cooked kupaty sausages

[nimble mason]

soeunds obscene

[languid pebble]

🙂

[clever oar]

like this

[languid pebble]

Enjoy!

[clever oar]

thx

[clever oar]

I'm sorry that your feelings were hurt🙁

[nimble mason]

ooo that actually looks good

espec the tomatoes

[clever oar]

i have chili papers

I still need to make some ramen

[languid pebble]

Should be easy

[clever oar]

I had to tinker with one ingredient 😎

not was easy

[languid pebble]

😂

[nocturne oak]

Well now. I think I'm done eating ramen for a bit.

[clever oar]

you're right, we need to cook normal soup

[languid pebble]

Good morning!

[hearty violet]

why my outpaint don't work?

[steep sleet]

Fooocus with Mobius Checkpoint

[stoic lark]

I give you, the robosexual pride flag. Designed in collaboration with GPT 4-o, Gemini, Llama 34b, and GGPT. Realized in collaboration with SDXL and Dall-e 4.

[clever oar]

if i start automatic it takes 1 gb vram after 1 generate it takes 3 gb and dont go back to 1 gb

how clean it

[languid pebble]

Is it A1111?

[clever oar]

yep

i can generate one picture

after that cuda of memory

[languid pebble]

There a new Fork called ReForge you might wanna try that ...

[crisp stream]

WDYM? 😄

[clever oar]

low vram help

[languid pebble]

Try https://github.com/Panchovix/stable-diffusion-webui-reForge

[clever oar]

i lazy for install new web ui)

[hearty violet]

did you see this?

https://www.youtube.com/watch?v=5qjR9aFRg1A

you upload a drawing, it does the reverse engineering and shows you step by step how to draw it, very good for studying drawing

[heady crane]

Guys try this model I don't have a1111

[royal monolith]

you should be able to run it in comfy just as easily as auto1111

[vivid shore]

1

[celest sigil]

Everyone just sitting around genning

[languid pebble]

Good morning coffee 🙂

[dark wharf]

Create a skeletons fossilized of a fierce battle between a Tyrannosaurus Rex and a Spinosaurus. The Spinosaurus and Tyrannosaurus skeleton models are generated based on museum-level fossil skeletons. The Spinosaurus fossil is based on the fossil skeleton specimen unearthed in Morocco in 2014. The complete Spinosaurus skeleton fossil model numbered FSAC-KK11888 highlights the long spinal spines on the back of the Spinosaurus skeleton, the elongated mouth, conical teeth, and high nostrils on the head. The T-Rex should be in an attacking pose, with its jaws wide open, The Spinosaurus is depicted with its crocodile-like mouth open and claws outstretched, ready to strike back. both captured mid-motion in a dynamic scene. The fossil should be embedded in a rugged stone slab with a natural, weathered texture, resembling an ancient discovery. The fossil should evoke a sense of prehistoric drama and tension.

[hearty gate]

GM

[frank drift]

Finally getting the hang of things! Starting with a simple celebrity for practice.

[candid surge]

wut? who's that?

[jovial tiger]

POV of man wearing beige boots on top of cliff, overlooking ancient ruins in grand canyon, shot with Sony Alpha A7 III, wide-angle. hyper realistic, cinematic photography, top view

[wide grove]

"Why isn't my hole being dug?"

[jovial tiger]

hows your day goin'? ah, it's been ok. rebels keep shooting at us.

[wispy nest]

wow I love these. which model is this ?

it knows star wars well

[jovial tiger]

kolors

[wispy nest]

ooh

I need to get

[jovial tiger]

POV of man wearing white storm trooper boots on top of tie fighter, overlooking death star battle in spaceship launching port, shot with Sony Alpha A7 III, wide-angle. hyper realistic, cinematic photography, top view

[wispy nest]

I love the way kolors looks

[languid pebble]

Good morning coffee 😄

[fossil mason]

The Last.fm user Kamgarodrigue17 has no listening history on their profile yet.

Just created your Last.fm account? Make sure you set it to track your music app.

Using Spotify? You can link that here. This can take a few minutes to start working.

Please note that .fmbot is not affiliated with Last.fm or Spotify.

The Last.fm user Kamgarodrigue17 has no listening history on their profile yet.

Just created your Last.fm account? Make sure you set it to track your music app.

Using Spotify? You can link that here. This can take a few minutes to start working.

Please note that .fmbot is not affiliated with Last.fm or Spotify.

[vast basalt]

any lora similar to this style?

[jaunty prawn]

I very much regret opening this channel 😐

[languid pebble]

😂

[celest sigil]

https://civitai.com/models/31977/children-book-illustration-style

Try searching children's cartoons 🤣

[jaunty prawn]

Is this Scott Pilgrem ?

https://tenor.com/view/stunned-scott-pilgrim-scott-pilgrim-takes-off-speechless-shocked-gif-5839745274967432864

[jaunty prawn]

https://civitai.com/models/425090/scott-pilgrim-takes-off-style-pack

makes me think of this~

[royal monolith]

@wispy nest

[hearty violet]

it was very good

[royal monolith]

thanks

[hearty violet]

@royal monolith I have to agree with this things to use sd3?

[royal monolith]

yeah, you do. and so did everyone else

[hearty violet]

ok only asking

wich model is the correct? I dont recognize this incl_clip archive and idk what s fp8, fp16

[royal monolith]

download all of the safetensors models. are you using comfy or auto1111 or something else?

[hearty violet]

a11

[royal monolith]

you're also going to want to go to the text encoders folder and download those as well. i'm not familiar with where auto1111 wants models put, however. that's a question for @dry crow

[hearty violet]

hmm looks differente use sd3 for sd1.5 and xl... I never have to take encoders idk where I have to put it... but I know the forlder hierarchy from a11

@dry crow can you help me? 🙂

looks simple, I ony have to put on chechkpoint folder

[hearty violet]

well a lot of things that idk

[dry crow]

You have to wait for the next stable auto1111 update to use SD3 with It.
In the current version its not supported.
You then only need the 5gb model of SD3 without the clip text encoders

[glass walrus]

Does this mean the installer is done?

[celest sigil]

Repost 😄

#🏞｜general-with-images message

Must've been good though, I remembered it 🙂

[languid pebble]

Yeah ... to many social medias ... this s*** happens ... sorry for that!

Sometimes I plan a post on X (Twitter) forget it ... and post a second "Good morning Coffee" picture ^^

[celest sigil]

I just delete all my images after posting anywhere.

[languid pebble]

Once I told an artist at my exhibition that I've deleted about 3k pictures that I've created to learn A.I. She almost fainted... 😄

[celest sigil]

I've deleted 10's of thousands

Definately more than 100k anyway

[languid pebble]

Sometimes it takes a lot of trials to get the picture you want. A friend once asked for an A.I. picture of her and her dog ... I never wanna do something like that again 😄

[languid pebble]

One topic I like: The lonely traveller ...

[unique condor]

Wait you guys save them? Sheeeeeit... I save maybe 1/100 I make lol

[celest sigil]

No, I don't save any

[unique condor]

Lol

[celest sigil]

That's how many I've possibly deleted along the way, not in one go 😄

[unique condor]

Oh I see

Well I just don't really get attached to anything I make with AI, but I do actual art. Most of the ahit I save are just memes and goofy stuff

I just view AI stuff as crap I do with my second monitor while I'm busy with other stuff. Helps with ADHD

[celest sigil]

Ah, so that's why I do it while I'm working!! 🤣

[unique condor]

yeah it really does help. ive been finding weird ways to cope with my ADHD for decades because i hate the stimulant based meds and they're the only ones that work for me. can't stand having my resting heart rate go from 60 to like 100bpm

[languid pebble]

That's why so many workflows don't save the images 😄

[hearty violet]

sd3 dont work yet for a11111?

[languid pebble]

I think original A1111 it works ... but not sure how good.

And maybe I'm on a beta ...

[hearty violet]

maybe other versin on hugginface... I will check, thanks

[languid pebble]

I got the idea: I don't wanna work with that at the moment any longer ...

[dry crow]

Not yet on stable. Only on the dev branch

[hearty violet]

so I can get that unstable version? sorry I don't understad XD

[dry crow]

You would have to use git checkout dev

But the dev version is buggy, would recommend to just wait for the stable release

[languid pebble]

Dev = Developer = Beta

[hearty violet]

Ok I will wait, thanks guys

hmm I'm new on comfy, how can I install manager extension on colab? is it used to install thers extensions, don't?

@languid pebble do you use comfy, right? can you help me?

[languid pebble]

The manager can install missing costum nodes and models, yes. But I have no idea how it works on colab.

[hearty violet]

exist a civiti ai browser plus to comfy?

[languid pebble]

And model manager:

CivitAI plugin for Comfy I have no idea ...

[hearty violet]

finally I'm on comfy * . *

[hearty violet]

how I do it?

[languid pebble]

It's a plugin you have to install ... but no idea how that works on colab

[hearty violet]

my god let me check it... haha ^^

[languid pebble]

I only used colabs for DeForum Videos about a 1.5 years ago ...

When installed it should look like this:

[gentle blaze]

I've had Photopea for a while and I didn't know about this wonder....

[sterile kiln]

There is a plugin like this in Photopea??

[gentle blaze]

It is the Magic Replace lasso tool. I suppose it will have some usage limitations...

[sterile kiln]

Stable.Art was the 1st to offer this (on Photoshop), and to my mind it's still the most effective.
There are a few minor problems, however, since the plugin has been discontinued, but it still works very well.

[gentle blaze]

I doubt I will use it much, but I was surprised that it has that tool with these models and parameters

[sterile kiln]

It's very useful for very localized inpainting. If it's like Stable Art, you can lasso areas and use SD with it, it's incredible.
Added to this, the Photoshop tools are very useful. It saves an enormous amount of time, and your work is much cleaner.

If you've got a near-perfect generated image, but one flaw is bothering you, it works miracles.

[robust path]

Upscaled with Upscayl and ultramix_balanced

[frank drift]

The one needed cropping, but I made it my pfp anyway for now

[opaque steeple]

a boy playing socoor.

[pure nymph]

for an Illusion Museum design an Entrance staircase with the main door at the top.that has some interactive illusion hidden in the stairs. Realistic, Exciting, attatactive, inviting.

[clever oar]

🖐️

[languid pebble]

👋

[clever oar]

i was on lake today

[marsh shale]

Was this made with stable diffusion u guys think?

[languid pebble]

I have been at an art exhibition ...

[wispy nest]

this is chinese sora clone

[smoky vigil]

Turbodong, is that you?

[clever oar]

yep

[languid pebble]

Cause there's a cam doing square pictures ... it's called Polaroid ...

[warm prawn]

hi everyone! today could you ask about the best practicies for positive prompt, i´ve found how do you write in the prompt for not cloned face or duplicate characters, but no work. this is my prompt
A image showing Taylor swift and Lana del Rey deep in conversation, surrounded by advisors and campaign staff. The setting is a modern, high-energy office space, with strong and dynamic lighting. Taylor swift is gesturing emphatically, with a confident and focused expression, while Lana listens attentively, also looking determined. The atmosphere is one of collaboration and strategic planning, with a sense of purpose and unity. The image should be highly realistic, bright, natural lighting.

negative_prompt: "naked, nude, deformed iris, deformed pupils, sketch, drawing, CGI, watermark, anime, comic, painting, abstract, surreal, fantasy, sci-fi, alien, mythical, low poly, pixel art, 3D render, vector, blurred background, text, close-up, cropped, out of frame, worst quality, jpeg artifacts, ugly, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, deformed, blurry, dehydrated, bad proportions, extra limbs, cloned face, disfigured, malformed limbs, missing arms, missing legs, fused fingers, long neck, low quality, lowres, mutation, signature, masked faces, face covered, cut-off face, trimmed face, obscured face, distorted faces, unnatural expressions, overly bright, overly dark, harsh shadows, excessive reflections, cluttered background, overly simplistic, flat colors, unrealistic colors, closed eyes, sleepy eyes, squinting eyes, duplicated people, repeated individuals",

can you help me?

[languid pebble]

Kill the negative Prompt ...

Model is SD3? You don't wanna use that ... use a finetuned SDXL ...

[warm prawn]

the base model is SDXL 1.0

[languid pebble]

Don't use base models ... they are unfinished ...

Use finetuned ones from civitai.com

For SDXL use a resolution about 1 Megapixel ... something around 1024x1024 ...

[warm prawn]

actually I have integrations with their models, but it doesn't work, is the same , i can see the most of the models work for one person, but not for two.
this image has the resolution 1024x1024

[languid pebble]

You could try: https://glif.app/@Oliveira/glifs/clw44qfbl0000m0zztwqk2tnf free creations after registration ...

SD3 so more complex prompts possible ... but hands, etc might be bad ..

[celest lance]

Why do my images come out like this sometimes? I notice when playinga round with the VAE and scheduler it will make it look better but it still doesnt look close to the prompt I copied from civitai

i got this after messing with the scheduler whats the importance of vae/scheduler?

[languid pebble]

At the models descriptions on civitai there are often important informations about the setting after the "more..." link. Otherwise you can look at the example pictures for the settings ...

[celest lance]

I did change my stuff to hsi settings which si here

But the settings only has the sampler not sure what to set the scheduler nad vae too

[languid pebble]

Wrong VAE or Shedulers can mess around a lot ..

[celest lance]

Ok so u can only use certain vaes with certain models right? I think

they gotta be compatible

[languid pebble]

Yes ... there are VAE's for SD 1.5, SDXL, SD3 and so on ....

[celest lance]

Ok thanks I need to start naming my models better i think

[languid pebble]

It's a point I only believe ... many models include VAE ... so you can use automatic mode ...

Depending on your UserInterface ...

[celest lance]

this is my workflow what vae u think they used for that image?

[languid pebble]

Steps 100is pretty high ... CFG 10 is pretty high, too ...

[celest lance]

yeah it says 100 steps on civitai ur right tho i notice i get better results with less

do people be capping

on their prompts

got this image pretty cool

[languid pebble]

CFG around 5-8 might be better, too.

[celest lance]

put steps to 80 and cfg to 4.5 on those images

[languid pebble]

Haters say: You only enter text and get good results ... but that's wrong .. there are so many settings to manage ...

[pallid ruin]

Hello! Do anyone have a kolors workflow?

[languid pebble]

No idea whee 🙂

SAhould be included there:

[celest lance]

@languid pebble where would i plugin Lora Loader?

[languid pebble]

I am not that good to teach others ....

[wispy nest]

lora loader has model as output

if I remember rightly

so it plugs into the model dot on ksampler

[hearty violet]

I got Manager 😄

[wispy nest]

nice

[hearty violet]

how can I install checkpoints there via web?

[celest lance]

Can I use two loras at one time?

Like how would I use a naruto LORA for characters and then a studio ghibli to make the naruto characters into studio ghibli style

[wispy nest]

yeah you can lora stack

[hearty violet]

how do I install missing nodes? It's not only click there? it's showing me this message, I thought it was automatically

[royal monolith]

you click on the button labeled install for the ones that you're missing - which is all of the ones that show up when you go into the install missing nodes function in manager. that popup you've got wants you to go to github, find the node's code page, copy the URL to the main page, and paste it into that box

[wispy nest]

it can be a bit finnicky

there's also terminal

and worse case scenario make a new custom node on a fresh github and paste the code in

as a temporary fix

[pallid ruin]

Aaaaa I still can´t make kolors work 😭

[wispy nest]

go on L2 discord they have workflows

[pallid ruin]

Nono, I got the workflow but I got an error

[hearty violet]

it's working

https://tenor.com/view/sway-arms-wave-swinging-gif-16544984

[pallid ruin]

Kolors finally woooooorked

[nimble mason]

has anyone here used "UltraPixel" for stable cascade?

cuz this is some real sketchy shit

[wispy nest]

not yet

looks amazing though

[nimble mason]

the ultrapixel_t2i.safetensors is not a safetensors file

it's a zip archive with a safetensors extension that contains a pickle

[wispy nest]

oh no

thanks for letting me know

[nimble mason]

yeah i don't like the deception at all

@grand walrus @proud dagger

https://github.com/2kpr/ComfyUI-UltraPixel/tree/main that's the node pack - i haven't noticed anything funny here though it's just been a quick cursory look...
https://github.com/catcathh/UltraPixel this is what the repo (and the official repo from the authors of the paper) links to for grabbing the model. that links to:
https://huggingface.co/roubaofeipi/UltraPixel/blob/main/ultrapixel_t2i.safetensors

looks like a normal pytorch save as best as i can tell but jeez, why call it safetensors then?

[nimble mason]

regardless of whether there's malicious code in there, that kind of deception is inexcusable

[jaunty prawn]

too bad this doesn't work 😂

[languid pebble]

It works so you can run away ....

... but bad luck we have Daleks Evolution ...

[gray canyon]

hii guys

[jaunty prawn]

Ive watched enough dr who to know they just FLY up the stairwell 🥲

[languid pebble]

Awwww... never really been a fan ...

[jaunty prawn]

😂

[languid pebble]

There's an old and a new series, isn't it?

[gray canyon]

hey what is the name of that discord server is which of ai, binaco , bonoco or something?

[languid pebble]

banodoco?

[gray canyon]

idk proper name my old account was in that its was amazing server for stabel deffusion and animatediff the logo was yellow or gold something