✨｜sdxl

[wet nacelle]

Holy shit

great work

[native knot]

😉

[copper kraken]

k nevermind i think this is actually something here

[wet nacelle]

👍🏻

[copper kraken]

shit i didn't realize it wasn't saving but yeah anyway, batch of 4, three had LILLY, all had VILLAGE, one had LIY VILLAGE, and then there's this one

chose that cuz there's a lot of opportunity for it to f it up

with counting letter parts that iterate

[copper kraken]

k holy shit

a billboard that says "IlEMWUNHS3"

first try

lost the U and that's it

[copper kraken]

S stayed an S, MW didn't merge

3 didn't turn into an S

[copper kraken]

hell yes everyone's discovering ipadapter+ now

it's insane

[copper kraken]

needs shark teeth lol

[native knot]

Ok...somehow this combination of settings gave me skater nature lego man?

[gloomy lark]

wow that's neat

[native knot]

I mean...all the elements are there, but wild combination

[gloomy lark]

@copper kraken

[copper kraken]

yeah this is pretty great for getting letters down

sometimes you need a few generations or to drop down to exponential... font is something to consider to stop merging but yeah

looking good

[gloomy lark]

definitely

[native knot]

Ok...this result is epic...

[native knot]

🤣

[gloomy lark]

hah it'll put your eye out kid

[native knot]

fk'n great

[copper kraken]

new ipadapter is god

don't even get me started on the style transfer weights

input, input, output

[burnt shadow]

Haha, gonna test IPAdapter+ today in case I get it going in Comfy 🙂

[copper kraken]

you should be able to just drag one of my workflows in

here's a style one

composition and style, combining two images for each

[burnt shadow]

yes, thank you, then I first have to copy them over into the according folder (?)

[copper kraken]

just save the images, then drag them into comfyui

[burnt shadow]

I would still need the models in Comfy is what I ment

[copper kraken]

just go ahead and drag those workflows in, if you don't have ipadapter they nodes will be red

i'll help you get the models etc

[burnt shadow]

Allight ty 🙂

[copper kraken]

np

[burnt shadow]

I openend it in Comfy and like expected it shows a few reds 🙂

[burnt shadow]

So which models would I have to get from that long list at github?

[copper kraken]

alright, do you have comfyui manager? got a manager button?

[burnt shadow]

yep

[copper kraken]

k do the whole install missing custom nodes thing

[burnt shadow]

Installing 🙂

[copper kraken]

https://huggingface.co/h94/IP-Adapter/resolve/main/sdxl_models/ip-adapter-plus_sdxl_vit-h.safetensors get this one first

https://huggingface.co/h94/IP-Adapter/resolve/main/models/image_encoder/model.safetensors and this, rename it "CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors" and put in /ComfyUI/models/clip_vision

the other goes in /ComfyUI/models/ipadapter

https://huggingface.co/h94/IP-Adapter/resolve/main/sdxl_models/ip-adapter_sdxl_vit-h.safetensors i'd get this second, it can be handy when it's overwhelming the image in some way and dialing down the weight isn't fixing it

works differently, it's more subtle and doesn't hammer the composition as much

the rest... kinda just depends if you want to do all the other stuff with sd1.5 and faces etc

[burnt shadow]

I would like to mainly use SDXL

[copper kraken]

sounds good, then get those three things and you're set

really only need the first two

[burnt shadow]

first the basics

yes, loading, (takes a moment 🙂 )

[copper kraken]

cool

[burnt shadow]

Where do I put this one (almost finished loading)?

[copper kraken]

/ComfyUI/models/ipadapter

[burnt shadow]

allight, ty

[copper kraken]

once you have that you should be able to run the workflows i dropped in those images

np

[burnt shadow]

btw for anyone into anything alike 🙂 https://youtube.com/watch?v=bKmQrrIjGYo

loading the CLIP model now

15 mins 😄

[copper kraken]

ahh crap

i hate how it blocks you from doing anything while it downloads

[burnt shadow]

nah, it´s fine 🙂

[copper kraken]

good time to grab some snacks i guess

it's pretty hard to get up and deal with bodily functions etc once you discover this

[burnt shadow]

I still can generate 🙂

[burnt shadow]

😄

[copper kraken]

you laugh now. you'll be using a water bottle to go in three hours lol

[burnt shadow]

😄

[copper kraken]

if any of you notice anything really interesting in particular with the weights, how to combine images with the composition weights above all else... def share

and uhhhh

embeds scaling - i haven't been systematic but i'm not aware of any reason to use V-only vs any other option... i've just been using the one at the end of the list with C penalty etc

[burnt shadow]

@copper kraken even though it shows this now in the manager, the IPAdapter nodes are still red (after restarting)

[copper kraken]

i remember i had some weird issue with that too

i forget what the exact solution was but i feel like i needed to restart comfyui twice for some dumb reason or something like that...

first thing to check though is if any of those new nodes show up in your node list

in the ipadapter folder

[native knot]

Plus, don't forget to download the models.

That won't cause the red, though.

[burnt shadow]

the models are in place already 🙂

[copper kraken]

yeah there was something weird before

try reloading the image too

the workflow that is

[burnt shadow]

yes did it twice (after it didn´t work with the first one)

possibly completely rebooting could help?

[native knot]

Or look at the console for why those nodes may not have loaded.

[copper kraken]

yeah copy paste everything into notepad and search for ipadapter

i wish i could remember what the issue was but i remember having tihs issue for sure when i updated

what i do remember is it either fixed itself invisibly or it was easy once i spotted the problem

do you see these?

[coarse creek]

how do i fix this img2img? i am using aihorde so sdxl (what generated this) doesnt have img2img, so i have to use sd1.5

[burnt shadow]

No, and I got this showing at the console @native knot

Couldn´t load the custom nodes for whatsoever reason

ImportError: cannot import name 'clip_preprocess' from 'comfy.clip_vision' (F:\AI-Art\comfyUI\ComfyUI_windows_portable\ComfyUI\comfy\clip_vision.py)

Cannot import F:\AI-Art\comfyUI\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_IPAdapter_plus module for custom nodes: cannot import name 'clip_preprocess' from 'comfy.clip_vision' (F:\AI-Art\comfyUI\ComfyUI_windows_portable\ComfyUI\comfy\clip_vision.py)

[copper kraken]

AH! i remember now

[native knot]

clip vision

[copper kraken]

it was my version of pytorch i think

[burnt shadow]

Do you know which one is needed?

[copper kraken]

what version do you have

i'll check and see what i have

[burnt shadow]

how can I check? Been a while? 😄

[copper kraken]

search the output for pytorch

[burnt shadow]

the console output?

[copper kraken]

yeah

[burnt shadow]

Doesn´t show any 😄

[copper kraken]

k one sec lemme reboot comfy

managed to scroll all that stuff off the top here by now lol

delicious

[native knot]

When was the last time you updated ComfyUI itself?

I think your version is too old for the new IPA nodes.

That was in your screenshot.

[burnt shadow]

End of Octobre (already pressed "Update all") 😄

[copper kraken]

oh holy shit

[native knot]

Yeah...WAY too old, bruh

[burnt shadow]

😄

[copper kraken]

that's like realizing you haven't updated since the carter administration 😄

in AI years

shit, a month is a decade in AI years

[burnt shadow]

I thought it might update manually on start 🙂

[native knot]

Nope

[copper kraken]

nah, gotta hit the button

[native knot]

You have to do it.

[copper kraken]

i check the github pretty much daily to see if there's updates that i'm interested in

[native knot]

Comfy is kind of an "update early/update often" thing. lol

[burnt shadow]

is there a yaml file I could edit for automatic update checking/installation like I could with other UI?

[copper kraken]

idk

[native knot]

Current version:

[burnt shadow]

yes thank you

still updating 🙂

[native knot]

You were 484 revisions behind.

😉

[burnt shadow]

how big is an update?

[burnt shadow]

😄

[copper kraken]

don't feel bad, i did this too a while ago

now i update almost daily lol

[native knot]

Most of the time they're tiny updates...but the new IPAdapter+ stuff was a bigger change. It changed the fundamental structure and required a ComfyUI update in order to run the new nodes.

[burnt shadow]

I bascially installed comfy end of October, did a few things, and simply dug it out again a few days ago

[copper kraken]

ipadapter plus will ensure that you open it daily for a looong time 😄

[burnt shadow]

😄

I´m currently more like in a softer AI phase, still generating basically daily, yet not as intense like during the 18 months before 😄

[copper kraken]

ha nice

i had a few lulls, then comfyui clicked

80k+ images and 3 monthsn later here i am

[native knot]

Shit...I have no idea how many images anymore.

In fact, I tossed my archive.

[burnt shadow]

me neither, something like 350k?

[copper kraken]

yeah eventually i'll be doing that

i've looked back at what i was making a couple months ago and i'm like dayum, this is trash

[native knot]

I was barely going back to it for anything.

[burnt shadow]

though only the ones saved, generated quite a bit more (via Wombo´s Dream) at the time

[copper kraken]

there are some things i really want to salvage though

and digging through all that shit would take days

it's cheaper in terms of time to just buy another bigass HDD and put a full one on the shelf lol

[burnt shadow]

wait for AI-image-recognition-search-tool, like where you can prompt your search 🙂

[copper kraken]

yes that would be fabulous

mostly i want to salvage styles

made lots of weird stuff like this with sd15

where i just love the style

and haven't ever been able to replicate it except with that one specific workflow and the silicon sd1.5 checkpoint

[burnt shadow]

well IP Adapter could help in that regard, no?

[copper kraken]

the style ipadapter is bringing these back to life in a new way

[copper kraken]

yep

just don't want to lose any of it

[burnt shadow]

I keep the stuff as well, alone for "nostalgy" 😄

[native knot]

I've got all of my nostalgia fix from my gaming collection.

[burnt shadow]

ah yes, it´s nostalgia in English 😄

[burnt shadow]

yeah, well, the steam library is basically enough here and I rarely to never replay games I played through

[copper kraken]

man you don't want to see my garage

i've already accepted i'll never be able to park in there again for the rest of my life

think my workflows get chaotic? it's like 3D abstract art man

i have three cats that all want to get in there so i can't take the time to get in and out carefully

[burnt shadow]

Would you have a CBS Coleco Vision in there?

[copper kraken]

the process of storing stuff now literally consists of look both ways, rip the door open and heave and slam it fast as i can as all three try to dive through the gap

[native knot]

I was never terribly into the Colecovision or the Intellivision, though I did play them.

[burnt shadow]

Liked it a lot at the times 😄

[native knot]

I was more interested in my Commodore64 and Odyssey2 at the time those things were around.

[burnt shadow]

yes, yet those were slightly later (only had C64 + Amiga)

[native knot]

The C64 and Colecovision both launched in the same month.

Odyssey2 came out about a year before the Intellivision.

But they were close enough.

[burnt shadow]

got a 64 in 84 while a coleco in 82 and Amiga in 87

[native knot]

The Odyssey2 was my first system, but the first one I really loved was the C64 followed soon by the NES. I think the NES is probably my favorite right now, but it sometimes flips to the SNES.

[burnt shadow]

Amiga in 89 I feel

[burnt shadow]

played zelda and secrets of mana on SNES and a bit of Donkey Kong Country

[native knot]

I ❤️ my copies of Zelda.

I have a boxed copy and a 2nd cart-only copy for playing for both the original Zelda and for Link's Awakening. TLOZ:LA is the best Zelda, don't @ me.

[burnt shadow]

so is it enough to update comfy only or would I habe to click update all?

[copper kraken]

it's usually a good idea to update everything too

[native knot]

You should update everything.

[copper kraken]

being 484 revisions behind, updating comfy might break a node

yeah

[native knot]

Gets the other add-ins up to date, too.

[burnt shadow]

yes, ty 🤭

[copper kraken]

masks with composition are really weird

[burnt shadow]

@copper kraken @native knot first IPA+ image generated 🙂

[native knot]

What are the source images?

Or source image and text?

[burnt shadow]

these are VQGAN images that work very well for input:

all I did was loading new images

[native knot]

nice

[burnt shadow]

@native knot @copper kraken

[copper kraken]

awesome

yeah we def need to get a lil shared collection of good style images started

i've been trying to set aside some good ones

[burnt shadow]

@native knot These work particularly well because they are similar to a diffused image, they are generated with the VQGAN AI (I use it via Wombo yet would like to use it locally as well, haven´t checked whether it´s that easily possible)

[copper kraken]

Interesting

[burnt shadow]

@native knot

[copper kraken]

so what weights are you using with these?

are you combining embeds?

[burnt shadow]

If you are referring to me I simply loaded the images in, just started dabbling with the weights

[copper kraken]

loaded them into ipadapter standard?

[burnt shadow]

btw why do you send the upscaled image again to a sampler before saving?

[copper kraken]

cuz i'm resampling it to add detail

[burnt shadow]

I simply loaded the images into the for "slots"

[copper kraken]

the straight up upscale basically just expands the pixels... bigger image, but no more detail, just like hitting ctl-+ on your broswer

[burnt shadow]

so with the sampler it gets AI-processed once again?

[copper kraken]

yup

have to be pretty careful with the settings for that second stage

i usually use exponential as the scheduler andcap the denoise at 0.5

otherwise it changes too much, which would be great, except usually it looks stupid cuz it's not at sdxl's trained resolution anymore

i find tiled resampling is better, but this is a lot faster

if you're wanting to add NEW details, def want to use tiled

[burnt shadow]

In Easy Diffusion I usually simply upscale via prompt strength 0 and steps 1, that way not alterations occur (currently testing with your setup in Comfy)

[copper kraken]

there's shit i can do to make it change even less

here i'm looking to add subtle detail

[burnt shadow]

yes, for those cases it´s different

[burnt shadow]

that´s already pretty different though 😄

[copper kraken]

that's a diff image completely lol

pre upscale , post upscale

[burnt shadow]

ah, I see 😄

[burnt shadow]

yes, improved

@copper kraken so well, thank you very much for you help @native knot

@copper kraken so how would I automatically adjust the image size to let´s say the first of the IPA-images without having to do it manually? When using inputimages it naturally choses the given dimensions, yet how can I route the image dimensions to the actual image generation when using IPA?

[copper kraken]

You mean read in the dimensions?

[burnt shadow]

yes

I already changed the empty latent image´s width and height to input, yet don´t know how to get the value output of the loaded image

@copper kraken and habe you already checked on the VQGAN images along IPA?

[copper kraken]

the node you need is in this image

it's in impactpack but i stuck it in here so you can just do install missing custom if you don't have it 🙂

[copper kraken]

going to in a sec here 🙂

[burnt shadow]

Looking forward

[copper kraken]

so what's the idea with the vqgan images and IPadapter - that they're similar to something that's partially denoised? curious about the theory behind it

just kinda muddy and weird with weights = linear, but really interesting with some of the other ones

reverse in/out

that's with the latent image being automatically set to the size of the image

[burnt shadow]

Well, they aren´t actually partially denoised but a fully generated image (that´s what the VQGAN-images look like), just saying, even though those are fully rendered they look and sort of behave similar to diffused images

[copper kraken]

ahh gotcha

yeah idk anything about vqgan

got into this stuff in the first week or two of december

[burnt shadow]

me neither except for using it via Wombo

[copper kraken]

gotcha

[burnt shadow]

it´s been the model they used prior to DiscoDiffusion (they used prior to SD)

[burnt shadow]

Looks like it´s as well doable locally (also like I remembered it was)

https://youtube.com/watch?v=_Br5WRwUz_U

[burnt shadow]

Interesting 🙂

@copper kraken This is what I get when installing missing nodes. It sepaks of renaming the node yet I wouldn´t know where exactly

[copper kraken]

i think that probably means within the workflow itself

but i'm not sure

i'll be honest, i've ignored all of those messages

every time

probably not very wise, but yeah

[burnt shadow]

😄

Yes, after two restarts that message dissappeared

[copper kraken]

lol

[gloomy lark]

From emad on sd3: Should be sometime this month per CTO comment, expect API shortly

Model went to all the chip makers for optimisation too and the distilled versions all done

[copper kraken]

good. praying that all goes through

some really cool results with timestepping the ipadapter

[burnt shadow]

@copper kraken so I got everything working except for these:

[gloomy lark]

Jeez, reading the Reddit thread, various things are not so great. 77 token effective limit still in place. Context window is 512 tokens, but the model is only trained on images that had a 77 token limit. Accusations that beta testers were booted who called out shortcomings. Oh well, better than what we have.

[copper kraken]

those aren't essential for this workflow

[burnt shadow]

and it says this :

[copper kraken]

you really want to get the one on the left eventually though... unsampling is extremely useful

[burnt shadow]

ah, so I can simply ignore them?

[copper kraken]

wow

yeah, for this workflow

where are they getting that info? was there an announcement, or is this a leak from beta testers?

that certainly would explain the very vanilla tests

i guess the best case scenario is the main issues come from incomplete training as a result of their financial problems

[gloomy lark]

I'm scared to post it here lest something bad happen to my beta access. Saying they have access to beta testers and their experience is that it's barley better than sdxl

[copper kraken]

wait you have access to sd3 now?

[gloomy lark]

People on Reddit

O I don't want to copy paste what they're saying.

I'll get the link.

[copper kraken]

k

[gloomy lark]

https://www.reddit.com/r/StableDiffusion/s/ce3wAe70LW

[copper kraken]

here's what i really want... something more trainable, better controlnet/ipadapter support, some upgrades for comfyui, etc...

[gloomy lark]

So scroll down in the comments. A couple people who claim to know things expand on what they've seen

[copper kraken]

text would really be nice and actions would be really nice but if we can train that or controlnet it then we can work around it

[gloomy lark]

I really wanted more than 77 tokens. My local llms have such a hard time string within that limit

[copper kraken]

yeah it sure would be nice

guess we need to get serious about thinking about ways to prompt with images more effectively

[gloomy lark]

Well, my hardware will get better, so the llms will get better. 🙂

[copper kraken]

true that

but yeah, i really think that's the secret in the end

generate components with just enough information on what the other components will look like style-wise that they can be integrated

then bring everytihng together somehow

what you said about regional prompting with one subject per region was pretty interesting on that note

the other thing i wonder about is this

" Even today some claim that SDXL is no better than SD1.5, so I am not surprised that some people say that about SD3 vs SDXL. People look for different things in a model, and I would not be suprised that a fine-tuned SDXL model can produce better looking images today compared to SD3. "

and seeing some flashes of genuine greatness from cascade

the biggest thing that stood out with cascade tbh was what it does with the texture of water during rough seas

yeah, the details are shitty and fucked up, but they are also at the same time fundamentally much better

the structure of the lighting is so much better

" They should have never announced and let us enjoy Cascade " <-- very true

people would be way more invested in getting whatever they could out of it if they hadn't followed up a week or two later with a way too early SD3 announcement

[burnt shadow]

So I checked the file and there isn´t any specific node from what I can tell, instead it uses the Empty Latent Image Node as well (?)

[copper kraken]

hmm well it's in the ImpactPack

[burnt shadow]

and when generating it ignores the size of the input image but uses the one from mentioned node

[copper kraken]

it's called "image info"

[burnt shadow]

cannot find it 😄

[burnt shadow]

Have you actually used it in that workflow?

[copper kraken]

https://github.com/ltdrdata/ComfyUI-Impact-Pack

yeah, it works

i've used it before

[burnt shadow]

when going through the categories of Impact I couldn´t find any image info in the shown nodes (quite a few). It´s installed though

[copper kraken]

Hmm

So you have it nowV

[burnt shadow]

when going via gitclone it says the directory Impact Pack already exists. Gonna check on it eventually, currently I just don´t feel like deleting the folder and reinstalling after all the shizzle already, prefer actually testing IPA for now 😄

[burnt shadow]

Like said, it appears within the node list (Impact Pack) yet within its sub-categories I cannot find an image info node and also not via the search

[copper kraken]

yeah it doesn't really matter too much anyway, just a minor convenience

[copper kraken]

Love all those

[burnt shadow]

thank you 🙂 Thanks to your help 🙂

[humble cape]

I use sdxl img2img and is use theses keywords in the starting of my prompt : poor quality photo , photo taken by an amateur

For my pfp is sdxl turbo i2i

[clear spade]

Hi Everyone,
Hope you all are doing good.
Is there any way to merge two models trained on different concepts. Like in one model we trained with cats and other we trained with dogs. Need to merge both these models keeping both dogs and cats without affecting output of each based on prompts we provide.

[noble shoal]

My unprofessional opinion on that is, hence cat and dog are 2 different tokens/concepts, you should be able to just merge the models together and keep both. As I said, unprofessional. You can just merge them on the fly with comfyui and check for yourself.

[clear spade]

Thanks for suggestion. I'm using automatic1111. Is there any option on automatic1111 to do so?

[noble shoal]

To be honest, I have no idea. I've lost track of A1111 since I use comfyui.

[clear spade]

Oh okay. No problem
Thanks

[copper kraken]

On the fly, no

That's only available in comfyui

You have to dump it out to disk and reload it

Also, use forge, it's much faster and better with vram than a1111

[clear spade]

Thanks for your suggestion. will try it

[copper kraken]

Made a huge diff for me on my 3080 12gb. I couldn't do batch sizes over 3 with SDXL without oom. Then for whatever reason it was down to one

And sometimes it went oom anyway

Forge I could do eight and it was twice as fast

[clear spade]

Sure, will try it

[native moon]

wow this looks cool!

[glass forge]

So where is it and is it that good?

I did apply to test it but I guess I'm not artist enough for it.

Or i gave a wrong discord name idk

whatever

It better not make inappropriate number of fingers.

[gloomy lark]

Nowhere yet. It'll certainly be mentioned here if it ever comes out in a month

[hollow echo]

Hello everyone, I need to do a small task of merging two images. I have to merge the image of a baby with an image of their grandparents. Is this possible? How do I do it? Thank you in advance, best regards.

[minor pecan]

https://tenor.com/view/ok-ojsimpson-stab-stabbing-standby-gif-20273350

[glass forge]

😄

[final ibex]

Is it safe to make the statement that SDXL is "too" real?

I haven't don't a lot with SD, but one of things I work on from time-to-time is an angel in a nice landscape setting. I have some solid prompt and setting variations that are very workable for me. The biggest trouble was getting the wingspan in the frame, plus some longshot distance around the subject.

I read that XL solves the framing distance / cropping problem. My "angel" is not meant to be completely realistic from the beginning, and is to be refined once some other things are corrected.

So the framing issue is better, but I cannot get any creativity out of the angels dress or the landscape surrounding. It looks like pictures of Xena princess Warrior during a filming break in a boring back lot, worn and tired. All the charm and fantasy is gone. Even the angel wings are like neglected and withering houseplants. In 1.5 I was getting less perfect skin and facial features and more interesting armor and robes, capes, boots with all sorts of grandiose design, clothings and accessories, gauntlets, etc. Now it's just blah.

[glass forge]

Pffff too real.... I wish, we're not there yet!

Pretty good tho, at first glance.

Avoid the eyes...

And fingers.

if you want epic fantasy weird whatever awesome results try the midjourney_xl lora

or the sd_xl_dpo lora

=0

together they work wonders

[final ibex]

@glass forge I'll check those out; thank you!

[glass forge]

😏

[gloomy lark]

Can you link that midjourney Lora you like? I know of one called midjourney mimic on civitai.

[glass forge]

https://civitai.com/models/251417/midjourney-mimic

thats the one

0.5-0.8 strength is bets in my experience

[gloomy lark]

Ok cool thanks. Yeah it really makes stuff look great. I was playing around with merging it with my favorite model and I found that it brought down prompt adherence for the sake of centering single subjects quite a bit. Something a lot of Lora's suffer from I think.

[glass forge]

yeah things do get smuddgy if you use a lot of loras with high settings

its almost like cooking

you need to be carefu with how much you add of each

like seasonings in food

https://civitai.com/models/242825/dpo-direct-preference-optimization-lora-for-xl-and-15-openrail

[gloomy lark]

For what I was doing, when you reach the strength that makes the image special, is also the point where it's not what you asked for anymore. I'll have to keep playing with it though.

[glass forge]

dpo lora does brign adherence of promt up a lot tho

but again exprient with strenght

[gloomy lark]

Cool I'll have a look

[glass forge]

👍

[burnt shadow]

Thank you, those were basically some of the very first images generated with IPAdapter+ after @copper kraken helped a lot on installation and also workflow 🙂

[potent ledge]

1girl

[crystal igloo]

generate a bee

[copper kraken]

Here is the image you requested.

[crystal igloo]

how do i gen stuff?

[frail cave]

Where can we generate pictures?

[slender coral]

Any suggestions on how to get the Ren and Stimpy cartoon style?

I can't seem to get it 😄

Ren and Stimpy, The Ren & Stimpy Show theme, beach background, hd, 4k, detailed.```

[frail cave]

How generate image ?

[flat acorn]

album cover I made

prev

first cover

[spiral bridge]

What’s a great sdxl model for realism I wanna recreate irl images with just prompts as a challenge

[solid adder]

hi

[nova oxide]

Galacticgears V10

[sterile rain]

Generate cartoon pictures. In a warm town, there lived a cute little cat. The kitten has big eyes and striped black and white hair. There was also a little blue bird, a little blue bird, full of feathers, and big eyes.
The picture shows a cat lounging in a cozy bed, outside the sun is shining, but the cat is indifferent, next to the alarm clock indicating the time to be late. His mother looked anxiously at the watch at the door of the room, her expression anxious. By contrasting the warmth of the interior with the vitality of the exterior, this picture vividly depicts the dragging habits of the kitten.
The kitten's face is beaming with contentment and pride as his mother cat and little bluebird cheer him on. The picture not only shows the success of the little cat, but also conveys the positive results brought by advance preparation. Return seed value

[gloomy lark]

Hey there, which model are you using for these? Thanks

[livid perch]

Hi!
https://civitai.com/models/139562

[wet nacelle]

@native knot Do you know where I could find the model for clip?

[native knot]

All those are in the list of models to install under the Install Models button in ComfyUI Manager.

[wet nacelle]

Which model is it?

[native knot]

There's several. Just grab all the ones for ipadapter plus.

[stray warren]

Here's the picture you didn't request

[wet nacelle]

@native knotI'm still confused. I downloaded these models from the manager and I'm still running into the error.

[copper kraken]

those aren't clip

[wet nacelle]

This is why I'm lost. I have no idea what to download

[copper kraken]

/ComfyUI/models/clip_vision

CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors, download and rename

https://github.com/cubiq/ComfyUI_IPAdapter_plus/tree/main

grab that one

it'll be called model.safetensor so you have to rename it appropriately

[wet nacelle]

Thanks

[copper kraken]

np

[copper kraken]

might need to restart comfyui and/or hit the refresh button

[copper kraken]

make sure you hit update comfyui and update all as well

[wet nacelle]

Sorry. What am I renaming it to?

[copper kraken]

CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors

[wet nacelle]

Ah I understand now. Thanks.

[native knot]

The clip models are available in the Manager, too though.

[wet nacelle]

I had trouble finding it. I have stupid.

[copper kraken]

main reason i download things manually is just cuz my connection sucks and it locks up comfyui till it's done

but yes, def good advice for finding stuff if you're on a reasonable connection 🙂

[copper kraken]

is ipadapter working now?L

[wet nacelle]

I just gave up on it as soon as I started using it. I thought I was going to be able to transfer style from one image to the other but that was a different workflow I can't copy from.

[copper kraken]

you can totally use it to transfer style

drop whatever workflow you have that gives you something you like in here

i'll pop in the IPA for ya 🙂

[elder harbor]

https://tenor.com/bFCiF.gif

[elder harbor]

https://tenor.com/c7MC3oKgOLk.gif

[elder harbor]

https://tenor.com/bc3HD.gif

[molten gull]

thats a cool car, too 🙂

[molten gull]

the last one i like a lot

how'd you make it ?

[rich olive]

SVD

[molten gull]

whats that and how does it work ?

[rich olive]

Can't really explain, it's a txt or img to video model by stability ai, can create stunning animations, I'm using it through comfyui

[molten gull]

can i just throw that thing you made there into my comfyui like i can throw in images ?

[rich olive]

I'm using img to video, just download the model and get a workflow

[vocal stream]

nice, animateddiff?

[rich olive]

SVD

[glass forge]

so made a lora

is 160 steps enough for 25 images

i mena it works.

i try to train juts like clothing items

whatever consitency is not there 100% even with the loras

everythign works "at first glance"

but dont look too close

[stone fossil]

Perfection. 😛

xd

[wet nacelle]

@copper kraken You made the original right? Here you go.

[wet nacelle]

\

[copper kraken]

yup that was me

[snow fable]

Hi everyone, I'm new here, I would like to install SDXL on my computer locally, I've been reviewing the minimum requirements and I think I meet them well, but I don't know where to start, any advice, where should I start?

[vocal stream]

type that same question into google and click on the link which is most recent and follow whatever steps it gives you

[analog tapir]

cheers

I would like to genereate the cover of a porno mag, as part of an image

[gloomy lark]

https://github.com/AUTOMATIC1111/stable-diffusion-webui Go to the Installation and Running section.

[analog tapir]

is my prompt just too long at 61 tks?

[gloomy lark]

really depends.

sdxl can only handle 1 to 2 subjects before it starts losing things

[copper kraken]

ipa style?

[native knot]

Nope

[copper kraken]

really cool regardless

[minor pecan]

mysterious kitty meat

[native knot]

Nice...that last one. ❤️

[gloomy lark]

pretty much the statue of liberty right there

[ember tusk]

一只白色的兔子

#🏞｜general-with-images 一只白色的兔子

[noble shoal]

Here is the image you requested

[verbal ivy]

||1girl, Chinese, dress, (beauty)||

1girl, Chinese, dress, (beauty)

[copper kraken]

Here is the image you requested.

[shy kelp]

Lightning and black jungle and red sky

[copper kraken]

Here is the image you requested.

[glass forge]

Heaven.

[uncut gull]

Does anyone know how to load a "controllite" controlnet in Comfy? I can load normal controlnets and controlnet loras using the loader and loader(diff) connected to the apply node or apply advanced node, but neither of those works for "controllite" models.
These models work normally in A1111 with no special settings, so I have no clues about what makes them different.
https://huggingface.co/bdsqlsz/qinglong_controlnet-lllite/tree/main

Nevermind found it https://github.com/kohya-ss/ControlNet-LLLite-ComfyUI

[mortal heart]

street view of the front of an adirondack sytle coffee shop in summer

[wet nacelle]

Do it yo self

[minor pecan]

Your image is now loading...

[mortal heart]

been a while since I have been on here, I forgot the command to generate 😄

[minor pecan]

#1047610792226340935

[minor pecan]

Here are the images you requested.

[mortal heart]

Awesome, thanks Doc!

[quartz warren]

ah new inpaint method with object detection https://www.youtube.com/watch?v=X89IQop_0dM

[glass forge]

Do one with Winnie the pooh holding the us flag.

[lusty dagger]

https://discordapp.com/channels/1002292111942635562/1004159122335354970 a pop girl

[topaz glacier]

手机简笔画

[bitter juniper]

I can finally run Deforum + SDXL thanks to Forge UI !

https://tenor.com/view/waynes-world-schwing-party-on-bow-were-not-worthy-gif-16269849

no idea who are the dev of Forge Ui but thanks for that gift !!!!

[copper kraken]

thank @visual glade for the backend 🙂

[verbal ivy]

||a woman in a kim dress sitting on a bench,||

A man, wearing a suit

[verbal ivy]

||A man, wearing a suit||

message:A man, wearing a suit

[royal karma]

draw chinchilla with the text happy birthday

[glass forge]

This is we gonna make Murica great again.

[fast eagle]

s

[uncut gull]

Testing my app and it's looking so artworthy. 🙂

[copper kraken]

great work so far for real

having done a fair amount of testing on this, anyone with an android tablet is going to be real happy 🙂

[meager canopy]

This is not what I meant by "suit up!"

[tranquil tulip]

@meager canopy a girl

[meager canopy]

How original

[still zealot]

A man, wearing a suit

[uncut gull]

Will push a commit today with a new app name and icon (and a bunch of fixes / improvements). I've been prepping material to explain the app better. Alpha version will be done today. 😊

[copper kraken]

i'm excited to see what it looks like!

[stone fossil]

Can I request 24 and a half girl?

[elder shoal]

Guys i have a question
what lora should i use to get this style of line espetialy?

[pallid path]

something about these images is so damn beautiful

stuff like "strong sunlight" really help

[crisp owl]

This is actually really great 😂

[native knot]

He's a Yakuzard.

[tranquil tulip]

@meager canopy A man ran out of a piece of 4a paper

[torpid garnet]

Does anyone know if @primal vault is still developing anything for SDXL? He hasn't updated his page in almost 5 months but I remember him being in this channel quite a bit. I hope I didn't tag the wrong person. Sry if I did.

https://civitai.com/models/111463/searge-sdxl-evolved-v432-optimized-workflow-for-comfyui-2023-11-13-txt2img-img2img-inpaint-revision-controlnet-loras-freeu-v1-and-v2

Does anyone have a good ComfyUI workflow alternative to Searge? Or has everyone moved on to something different from ComfyUI? Its been a while since I've been on the channel.

[tawdry current]

Just make your own workflow that has the parts you need? It's not like his templates are some holy grail or anything. What do you specifically need for your workflow?

[torpid garnet]

I need mostly the top portion of his workflow. Prompt areas, customizing image settings and, a place for multiple loras, upscaling and fixing.

I'm not very good at messing with all the noodles and boxes.

[torpid garnet]

If you have a good workflow that you can share I would appreciate trying it out.

[tawdry current]

AI is a really fast moving field, it might help you out a ton more in the long run if you take some time now to learn piecing workflows together for yourself. There are tons of guides and videos about it all that range from super elementary to extremely advanced

Once you can read noodle, you can make spaghetti

[torpid garnet]

lol... that is fair.

[tawdry current]

IDK maybe I'm just biased though since I spend a lot of time developing with ue4 and ue5 lol

I've been a pastafarian for a long time due to blueprints

[torpid garnet]

I'll give it a try but I'm far from a developer. I am a medical receptionist. This is all greek to me.

[native knot]

Sadly, I believe Michael may have passed away. 😦
Not only did he have great work here, but he was the creator of MCP (Minecraft Coder Pack/Mod Coder Pack) and even ended up working for Mojang AB. The last time there was any activity across literally every account I know of for him was back on December 8th.

[copper kraken]

wow 😦

[tawdry current]

You'd be surprised what any average or greater intelligence person can learn, given some time and mild dedication. Just start out simple and small and work your way up. Learn the terminology and try to grasp what each step is doing and why you're doing it.

You'll screw up a ton, but that helps you learn

[hoary saddle]

when github fork? 🙂

[primal vault]

"The reports of my death are greatly exaggerated." - I'm just busy IRL and can't spend much time on SD/SDXL stuff right now.

[nocturne fox]

A dog is running towards an old man in Memphis style with 4k image quality

[copper kraken]

Here is the image you requested.

[nocturne fox]

Chinese girls wearing bikinis, big eyes, small face, on the beach

[copper kraken]

Here is the image you requested.

[visual glade]

https://huggingface.co/stabilityai/cosxl

[copper kraken]

awesome, thanks

[nocturne fox]

傻逼，放的什么国旗？

[copper kraken]

我是一个机器人，只能把中国真正统治者的旗帜放在中国的长城边。

[nocturne fox]

Chinese opera characters
extremely minimalism portrait
geometric shapes
cartoonish lithographs

[copper kraken]

Here is the image you requested.

[nocturne fox]

傻逼玩意

滚蛋

[uncut gull]

Posted alpha last night! https://github.com/QuintessentialForms/ParrotLUX

[uncut gull]

Amazing! Instruct demo before SD3. Wonder if Lightning merge would work.

[meager canopy]

Why are you just copying another post here? Stop being a fool and leave if you have nothing to contribute.

[noble shoal]

Ok, the new Edit model is kinda cool

[meager canopy]

I tried it, but don't think it's something I'd use. Couldn't figure out what the non edit model was for.

[noble shoal]

The non edit model is a riddle for me too

[meager canopy]

It doesn't merge and doesn't create good images 🤷🏻‍♂️

[native knot]

Bro, I can't say how relieved I am to see this post. So glad that the rumor can be busted from the source itself.

[noble shoal]

I defiantly have use-cases for the edit model though

[meager canopy]

Cool! Was that just using the demo workflow?

[noble shoal]

Yes

And if you hook up a styler you don't even have to prompt

Use the styler output as prompt and it will just change the style of the image. Done

[valid niche]

The architectural design of the Ark at Sea presents a magnificent ocean painting. On the vast sea surface,the ark stands tall and towering,blending with the shimmering sea water. The architectural design style is unique,combining modern simplicity and ocean elements,with smooth and powerful lines. The appearance of the ark adopts a deep blue color scheme,symbolizing the depth and vastness of the ocean. The details are even more exquisite and subtle,such as the sail shaped building structure,cleverly converting wind power into clean energy. At night,the ark is brightly lit,like a beacon on the sea,guiding the lost navigator. This picture not only showcases the harmonious coexistence between human wisdom and nature,but also outlines a grand blueprint for a future ocean city

The architectural design of the Ark at Sea presents a magnificent ocean painting. On the vast sea surface,the ark stands tall and towering,blending with the shimmering sea water. The architectural design style is unique,combining modern simplicity and ocean elements,with smooth and powerful lines. The appearance of the ark adopts a deep blue color scheme,symbolizing the depth and vastness of the ocean. The details are even more exquisite and subtle,such as the sail shaped building structure,cleverly converting wind power into clean energy. At night,the ark is brightly lit,like a beacon on the sea,guiding the lost navigator. This picture not only showcases the harmonious coexistence between human wisdom and nature,but also outlines a grand blueprint for a future ocean city

[meager canopy]

Not sure what happened to mine 😦

[noble shoal]

Let me change my WF so it outputs png images again

[meager canopy]

I only added the styler

[noble shoal]

And changed the sampler

[meager canopy]

Yes, and changing the CFG to try and balance it better.

Getting there

They all look washed out

Better

[noble shoal]

Still weird artifacts going on

[meager canopy]

Yes, when I change the style, it looks bad again

[meager canopy]

This is with the default workflow, just adding the prompt a pencil line drawing of a law enforcement robot

[noble shoal]

Don't bother to mention the robots "Pencil drawing"

Default workflow

[meager canopy]

This isn't using the styler though

[noble shoal]

I just fired up the default workflow to reproduce

[meager canopy]

How did it know what style to change to?

Oh! I misread your earlier message

[noble shoal]

The workflow is in my image. I havent changed anything

[noble shoal]

Oh, ok

[meager canopy]

Nope...still sh!t. Something is very wrong here.
This is using your image workflow

[noble shoal]

You sure? Why is your image called "image.png"?

[meager canopy]

Because I just copied it out of the save window

[noble shoal]

Ok, that makes sense

Have you updated comfyui lately?

[meager canopy]

Every time it starts 😄

[noble shoal]

Can you send me the last image with metadata so i can reproduce it on my machine?

[meager canopy]

It is literally what you are using

[noble shoal]

oh

[uncut steeple]

Please use #1092446741984444416 to advertise your project

[meager canopy]

I dropped your image into my UI and ran it

[noble shoal]

Then there is really something odd going on on your side.

[meager canopy]

Yeh, maybe I should try the portable version

Perhaps all of my gens would have been better 🤣

[slate birch]

#🌠｜show-and-tell

[sick fulcrum]

Can sdxl run on colab?

[meager canopy]

@noble shoal First run on windows portable version... 🙂

[noble shoal]

Now that waaaaaaay better

[meager canopy]

Yes, really odd, but I'm happier now 🙂

[livid pond]

hello, i watch this video to generate sdxl image,https://www.youtube.com/watch?v=9k-yb83ZHfc&ab_channel=MattWolfe, why my image is pure blue color image. vae:https://huggingface.co/stabilityai/sdxl-vae/tree/main, base:https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/tree/main,refiner:https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0/tree/main, my comfyui log is :got prompt
model_type EPS
Using split attention in VAE
Using split attention in VAE
clip missing: ['clip_l.logit_scale', 'clip_l.transformer.text_projection.weight']
Requested to load SDXLClipModel
Loading 1 new model
Requested to load SDXL
Loading 1 new model
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:08<00:00, 2.34it/s]
Requested to load AutoencoderKL
Loading 1 new model
Prompt executed in 18.44 seconds

thank you

[meager canopy]

I don't believe this! The image quality has gone back to how it was before I installed the portable version. Maybe it's reletaed to some custom nodes I install 🤷🏻‍♂️

[hoary saddle]

this is great, looking for a way to add a remote comfyui ip address/port 🙂

[uncut gull]

Yeah that's been on my todo list for ages... Do you have a cloud comfy api running somewhere?

[hoary saddle]

i run it on a local pc with a 4090 i use for testing, have it setup so i can reach it from a public IP

mass replaced all the 8188's with my port number and the IP address as the host

no luck yet tho, still playin around, great app so far tho!

[uncut gull]

You install/run the github repo on the same pc as comfy, then you access your public IP on port 6789 should work?

[hoary saddle]

just modifying Comfy-SC.json right now to see if i can get it to connect 🙂

yeah, running great on my cloud VPS server for the frontend

just sits here while trying to reach back to my comfy server via those changes i made in the comfy json file

[uncut gull]

Yeah, it won't work I think. Let me recheck the code to confirm though.

[hoary saddle]

my doodle version from months ago

mine was all python so it kinda sucked lol

[uncut gull]

Huh. Actually it should work. In Comfy-SC.json you would change "host" to the ip address and port to the port. I've never tested it, but the code handling it looks right at a glance.

[hoary saddle]

massed replaced all the x.x.x.x with my external IP, nothing in console, comfy didn't throw any info either on the other server
will keep tinkering, i enjoy this stuff 🙂

i know my external is working tho, my other app is still connecting via IP/port

[uncut gull]

Yep. Looks right then. Can't debug till after work. If there's a bug in my code (sounds likely), it's in server.js somewhere after line 177.
Be warned though: I have an implementation of formdata there that looks intimidating. 🙂 But comfy uses formdata, so there it is.

Easiest fix is just to run it in your local PC, but I'll debug it tonight.

[hoary saddle]

looks right, and claudeAI thinks it looks right 🙂

[hoary saddle]

no worries 🙂 don't get in trouble at the office

[uncut gull]

LOL found it. Line 295 and line 338 request.port ? ... should say requestOptions.port ? ... Don't think I can push a change from here tho.

[elder plume]

have you explored adding support and widgets to a pre-existing application?

[uncut gull]

Not sure I follow. What did you have in mind?
(This is a webapp, so technically it could be a node in Comfy or a tab in A1111.)

[elder plume]

i am saying that there is a lot of demand for high quality openpose posing inside e.g. photoshop, which could be connected with, whatever

or if there were something that had a hope and a prayer of supporting extensions on ipad for drawing purposes

but then you wouldn't get mired in the product development of writing a paint application

there are a lot of existing ones

[hoary saddle]

changed both lines, gonna restart node

[elder plume]

have you tried any of the photoshop plugins?

[uncut gull]

Make sure you change both POST and GET lines.

[hoary saddle]

i have not

[uncut gull]

I have never used photoshop, and I'm only personally interested in a tablet UI, something like Procreate (also never used that though.)

[elder plume]

which drawing tools have you used on a tablet?

[uncut gull]

There are zero open-source apps with a tablet-based UI like Procreate / InfinitePainter that I could integrate this into. It's a very deep integration you know. Have you tried the Krita plugin? That thing is clumsy and tiny. 😕

[uncut gull]

Infinite Painter on Android. That's where my UI inspiration came from.

[uncut gull]

Any success??? 😮

[hoary saddle]

not yet, gonna add some console logging to see where it's hanging

[uncut gull]

Also, if you have an ipad, please test my app. 🙏I'm really curious to know how far it gets through loading in Safari.

[elder plume]

i see. so what is your goal?

everything makes sense to me

[uncut gull]

Goal is to be able to use AI while I'm making art, but forget about the tech. Seamless experience so I don't break my flow / lose sight of my art. 🙂

[elder plume]

for example, "compile krita to wasm, write a shim for the pen attributes of the pointer / touch events" has a 100%, known, defined goal that requires basically no product development decisions. then, you can make a plugin that "just" works inside krita

it's a ton of work

but suppose you had unlimited manic energy. lots of people would use that

that would also solve the issue of using krita on an ipad

[uncut gull]

Krita runs on Android, I've used it for animation. I always go back to InfinitePainter because the UI just "disappears", so I really feel immersed in my art.
My goal here is about the experience, not about the exact tech, if that makes sense.

[elder plume]

gotchya

[hoary saddle]

i wanted to recreate the SD doodler thing, need to get back to working on it, so many side projects lol

[elder plume]

but you can modify krita to make the UI better, no?

and if you did that

[uncut gull]

Yeah sketch -> render has been the main way I've found myself using my app so far. Works so smooth. 🙂

[elder plume]

then, you would get everything else they've figured out

[uncut gull]

No, definitely can't. Would take years to rewrite the codebase.
(This uses CSS, the most powerful UI spec out there.)

[elder plume]

hmm who said anything about rewriting the codebase?

i mean surely making UI elements in it is pretty straightforward

[hoary saddle]

guess i need to set mine up with a cascade workflow, would probably get better results, might try that after dinner tonight

[elder plume]

this is just food for thought

[elder plume]

there is only a canny controlnet for cascade

no idea how img2img would work with it for this problem

[hoary saddle]

ah yeah, true

[elder plume]

@uncut gull like if you have unlimited manic energy for programming tasks, why not focus on krita?

[uncut gull]

That is absurdly horrifying. Seriously. UI coding is the hardest part of any software. It's 80% of the time investment, easily.
And you can't make this UI with Krita's toolkit. It's just not compatible.

[elder plume]

And you can't make this UI with Krita's toolkit. It's just not compatible.
this is something you determined from reading their code?

i know you mean to use the word "impracticable" and not "can't"

but i am aware there are already krita plugins for the automatic web ui API

and that has custom UI

and yes, maybe krita doesn't have a robust way to define new widgets / manipulators, but i don't know. i guess the important thing is that answering such a question requires zero product decisions, it is only an engineering question

[uncut gull]

The UI touches every single tool in the codebase. It's the deepest part of any integration. You have to change literally every piece of code there is. Sorry, but I don't think you've tried something like this before.

[elder plume]

so i am hearing you have not read their codebase

which is okay

or at least explored this

that's why we're talking about it

i haven't either

[uncut gull]

I have hacked on Krita before. You're talking about an 80% rewrite.

[elder plume]

what was your goal with krita hacking?

[uncut gull]

Debugging the SD plugin back when it was new.

[elder plume]

okay, and you didn't find any pre-existing browser based drawing tool that you could modify? i am sincerely curious which ones you've looked at and why you didn't think they were useful to work with

and finally, you are 100% certain you don't want to build a plugin for photoshop instead, which has a really robust and thriving plugin ecosystem? or something else on the PC / macOS device that has first-party remote pen / tablet support, such as photoshop?

i wouldn't want to work on krita either, don't misunderstand me

here's an example of a product decision: maybe there is a reason there aren't widgets for making pose vectors. presumably it would make more sense that the user draws a pose, maybe with a stick figure, and then something else turns that into the pose vector for controlnets

because moving around pose widgets is kind of tedious

i don' tknow though.

[uncut gull]

Because, again, you have to integrate this into the UI. Plugins let you add tools at best. You realize my app had an open-ended UI called APIFlows? That means whatever your AI tool input needs, you can write 1 json file and have a seamless art experience. I had zero prior art to draw on building this. Just the idea of connecting layers to an open-ended tool UI as node inputs is completely novel across all art apps. This cannot be done without rewriting an app's UI, and that's 80% of any art app, and there are zero tablet art apps with even basic features.

[elder plume]

okay i really respect what you are working on don't misunderstand me

it would be really helpful to answer the question

even if it is tempting to give me a bunch of opinions

[uncut gull]

This would require training a new AI model, but that dataset doesn't exist.

[elder plume]

i kind of want to know what you already looked at

and either the answer is "nothing" or a list of pre-existing web apps

[uncut gull]

I tried to improve the Krita plugin, then went looking for open source art apps on github, was shocked by their lack of features, wondered if I should fork something, then realized the UI code was something never done before. It couldn't be patched in. 😕 So here I am 2 months later with a new app.

[elder plume]

photoshop already has pretty robust generative tooling, and it will ostensibly get better and better. it is arguable if controlnets even make sense, in the long term, there's a reason dall-e3, imagen/google, ideogram, midjourney do not have them

[elder plume]

okay, so you're not aware of any pen based open source web apps?

you haven't looked at any

simple yes or no correct?

[uncut gull]

I am trying very hard not to be offended, so I'll sign off here. Have a nice day.

[elder plume]

i am sorry, i am not trying to offend you. i am not talking about krita, just other applications that you may or may not have looked at

i found two that might interest you - https://github.com/LHRUN/paint-board and https://github.com/bitbof/klecks

@uncut gull both are very high quality and have a lot of great product development in them

@hoary saddle this lhrun paint board is really nice and has pre-existing, super slick google "ai doodle" integration

like it's very simple

[uncut gull]

LOL yep those are both shockingly bad apps I saw. Here's some more. 😛 https://opensource.com/life/16/5/open-source-drawing-applications-android

[hoary saddle]

hah, i was just reading that github

[elder plume]

this thing you linked me does not have either of these projects in it

it sounds like the answer was no

[hoary saddle]

i like yours better if i can get it to run on my setup 🙂 gonna finish up some work at the office and tinker more tonight

[elder plume]

yeah i think it's all good

i am just wondering

there is so much effort going into so many different, uncoordinated places

[elder plume]

and before i am painted as the bad guy here, you know, it is pretty bold to say those really high quality, well polished, maintained apps with audiences and some adoption are "shockingly bad"

i don't think a paint app needs to have adoption or be bug free or have an audience or whatever to be good. there is no "right" app or opinion. but something something, log in your own eye first

[uncut gull]

It's about art. Krita vs. Infinite Painter: Krita has more features, fewer bugs, a larger community, and is free, and has a SD plugin. So, why don't I use Krita on my tablet? It's something about the "immersiveness" of Infinite Painter (which was basically copying Procreate). It just feels natural. You forget about the tech and lose yourself in the art, just like with physical media.
That's the goal of my app. That flow-like experience. Nothing else. 🙂

[copper kraken]

yeah, this is a big deal imo

there's a big gap right now... if you want to use drawing in SD, you have to transfer stuff back and forth and break your flow with too much time on the tech stuff, as much as i enjoy that

[elder plume]

I understand you are very opinionated about the UX, and I agree with you

It’s just the wisdom of writing a layered paint tool from scratch

When that LHRUN repo has literally three onscreen UI elements

The journey to removing or adapting them to suit your needs is small

That’s an objective fact

[uncut gull]

Then you are free to try, friend. I've never written typescript before.
(Did you know that LHRUN app you just mentioned doesn't have pinch to zoom-pan-rotate like Procreate does? That's because in order to code that functionality you need to understand matrix inversion. Instead, they have a "hold two fingers" to activate a pan and zoom mode, no rotation. Easy vectors. What else. They don't have layers, much less layer groups, so obviously no layer blending. They aren't GPU accelerated; they're limited to SVG drawing, so you can't paint at all. They don't have undo/redo. Do I really need to go on? Do you get what I mean by "shockingly bad"? I'm not offended though. You just didn't take time to do research, and you also wouldn't have my dev background to guide you if you tried to research. Thanks for your interest in my app, and if you do get to test it on ipad, I'd love to know how it goes. :-))

[elder plume]

it is true, i am not really paying close attention. i suppose kleks is pretty close

it has layers

i mean you had to write the layers from scratch no? might as well have written them from scratch for LHRUN

pinch to zoom? same thing

if you feel strongly about it, implement it maybe for this existing thing

i agree with you that this stuff is challenging

and i am not minimizing what you are trying to accomplish

I've never written typescript before
you should try it soon

[uncut gull]

What, exactly, would I be reusing from LHRUN? Because bear in mind, I would need to spend a month at least learning its codebase. Three months for Kleks, and a year for Krita (if I'm even capable of learning a codebase that big).
I started this app around February 22nd, and I've already surpassed both those apps, and I know my own codebase inside and out, so I have unlimited, fast iterating power on it.

[elder plume]

i did a very light search on github. maybe 5 minutes. it's your time, not mine

I started this app around February 22nd, and I've already surpassed both those apps
hmm

[uncut gull]

I strongly oppose typed languages. I'm in the camp that you waste 50% of your time debugging typedefs and only save maybe 10% preventing type-bugs.

[elder plume]

all i advocating for is to focus your manic energy

[uncut gull]

LOL yeah with caveats. Kleks probably has a couple things I don't. But I could never rewrite Klek's UI. Not my code.

[elder plume]

to just consider it

because it can yield great things

it's like a water cannon

you choose where to point it

while in principle you will learn a lot by building a layer best editing tool, even a high quality one

like it's good to be very opinionated about the UX, but delivering those opinions into applications people actually use, like Photoshop, it will be very high yield

i mean nobody uses krita either

if you feel strongly about tablet based tools

it's a lot of work. ther'es a reason there isn't a good github project for native ipad drawing

it's a ton of work and in a darwinian way, all the people who tried to do it for free have failed

i'm sure people have tried

[uncut gull]

For whatever reason, all my YT artists use Krita. I'm really like zero-exposure on PS.