#✨|sdxl
1 messages · Page 132 of 1
hi!! rookie here!
what kind of magic is this? how can this be done? stable diffusion api?
I made a little magic machine
input images, output new image
myyyyy gawd! you are a wizard. nice job!! it's impressive
is it possible to play with it? 🙂
if you download the right user interface
bit of a learning curve if you're new to all of this. but most of the images I post have the metadata in them
well a lot of them do
so it'll load my setup, or if you like something someone else posts same thing
so you use 3x ip adapter? or do you mix ip adapter and revision/img2img?
(not at home so i can't take a look at your workflow)
well 2 ipadapters, but one only run model data through and the second only use the clipvision
and then a second clip g clipvision
and then a controlnet canny
it can work out a lot differently depending on how you adjust things, but prompts are pretty much automated unless you choose to take control of it for whatever reason
so I can just plug random images in and start rendering
ok, sounds too complicated to understand without looking at the workflow - especially the part with the model data and clipvision. will check it when i'm back home.
SDXL AUTOMATIC1111 DynaVision checkpoint
so cute

hey, what CLIP vision are you using for ip-adapter_sdxl_vit-h.bin?
@hardy cipher ? I'm messing with image blend again and there is a new IPA SDXL model? I'm asking you because I know you use that stuff
lol, sometimes the sampler can make a huge difference 😮
but which one is the tastiest
Ouch if I slightly close my eyes the eyes look in the right place.
That "yarn" tho 😆
Elements are the most intresting (to me) to add to the models LoRa wise.
Persons and objects fun also ofc, however it is style that gives the most freedom.
Its looks good I like it. 🙂
Yes, he is wearing a meat hoodie...
Once again here is the link to the LoRa to do it: https://civitai.com/models/146424
Last time posting it as I dont like marketing a lot but I think it is cute as w00lz.
Creasy face
lol, xD great minds think alike? (it's on purpose they aren't that close)
Found a new (to me) interesting keyword: "{subject} info graphic" (if you check my metadata, yes i wrote it wrong in one prompt)
ooh, i think you might've just helped me
I knew it!
venussaur, blastoise, ghastly... uhhh
charizard, rattata, sandslash
goddammit, i really have to do the whole list, don't i?
lol, started a generation for all 150 pokemon in that style
every pokemon after 150 is non-existant in my mind
So I was asking myself, how do you cook eggs soft cooked but make they not hard to peel?
Throw them in ice water. Well, it depends how soft they are.
Noice. ❤️
when peeling, remove the bottom(air pocket) and make sure the membrane is broken, submerge in water, now crack and roll the shell, the water will get between membrane and egg making seperation easy
Now these are some egg tips, thanks lads. 🙂
I think Eface is actually an LLM
i am?!
dreamcatcher sleep well 🙂
He is not, is is more ugly in real life. 😉
Are you conscious?
Runs.
no, i'm highly sophisticated
and at times drunk
but conscious? naahh... i don't do that
Eface is one of the best people I know hehe niggah got me into SD.
somehow, her hair just not being there after the crown looks uhh... i dunno i feel hair should be where the crown is too
I am learning about hands at the moment. Interesting
bulbasaur, ivysaur, venusaur
charmander, charmeleon, charizard (no wings lol)
squirtle, wartortle, blastoise
caterpie, metapod, butterfree
weedle, kakuna, beedrill... uhh...i think some tokens also have other uses here
same seed, so the 3 bird evolutions are pretty much the same (pidgeot series)
rattata and ratticate
spearow and fearow
ekans and arbok
pikachu and raichu, they got buff 😮
sandshrew and sandslash, very cool
clefairy and clefable. holy shit lol
vulpix and ninetails
jigglypuff and wigglytuff... rest of generations not done yet, these are amazing 😮
oh wait, there's more. didn't refresh yet. zubat and golbat
oddish, gloom, and vileplume. dang, proper horror show with these
paras and parasect
venonat and venomoth
digglet and dugtrio. lol
meowth (damn this one is cool) persian failed, probably a lot of overlap with other tokens
psyduck and golduck. what an evolution!
mankey and primeape
growlith and arcanine, very cool
poliwag, poliwhirl, poliwrath keeping it real
abra, cadabra, alakazam
machop, machoke, machamp, unfortunately no 4 arms 😦
Nice, those are neat
bellsprout, weepinbell and victreebell keeping the horror show going
tentacool and tentacruel. hentai series incoming!
geodude, graveler and golem, golem probably not ast strong again, overlapping tokens
ponyta and rapidash
slowpoke and slowbro
looks like bathroom tiles but not sure how i did it.
magnemite and magneton
farfetch'd, doduo, dodrio... yeah i figured these would not go very right
magneton we used to call cassette players 🙂
seel and dewgong
grimer and muk, majestic as fuck
shellder and cloister
gastly, haunter and gengar
+10 points for that gengar tho, he looks very ready to go
onyx -> overlap again, it would dud
drowzee and hypno
krabby and kingler
and now i finally caught up to generations XD
gonna take a small break fom posting(not generation) and post the rest later
is this a lora you've trained or just prompting?
just prompt 🙂
was doing a creature concept art prompt and one of them got this cool result
so naturally i fugred i need to put it through all the pokemon now 😛
Wow those are awesome. What is the prompt/Lora did you use please?
(Sandslash:0.5) creature concept art by Utagawa Toyoharu, surreal absurdism, 2d info graphic, painting Negative prompt: pokemon
no loras
voltorb and elctrode. yeah, in game they don't differ that much either
exeggcute and exeggutor, about what you'd expect
cubone and marowak
hitmonlee and hitmonchan
lickitung just being plain distrubting
koffing and weezing
ryhorn and rhydon
chansey, just as ugly as those other pink fcks
tangela
kangaskhan, horsea, seadra
goldeen and seaking... meh.
staryu and starmie. that's a black swan story right there.
mr mime here to deliver nightmares
scyther, jynx, electabuzz, magmar
picnir, tauros
magikarp and gyaridos. you vs the guy she told you not to worry about
lapras, ditto
evee, vaporeon, jolteon, flareon. these are very cool 😮
porygon lol
omanyte and omastar
kabuto and kabutops
aerodactyl, snorlax
articuno, zapdos, moltres, again very cool trio
dratini, dragonair, dragonite
mewtoo and mew
that's it ^^
all done
nice color scheme. your images are my favorite - have downloaded quite a few of them as input for ip adapter & co.
😄
now going all randoms with also random artists/movements 😄 i'm sure more cool combos will happen!
Amogus
(I figured out how to improve IPA's blend thingy while implementing AIT =])
Collaboration could be the key
prompt?
w00len, a very dark hyper realistic macro close up shot of ((Bambi made out of wool and yarn)) bokeh forrest backdrop
Using my lora.
Can you imagine that somebody has so much time, that he creates a @hardy cipher LoRA?
Ok....it was me
@indigo carbon can I DM?
yeah
Thanks
This LoRA is absolutely cursed. Prompt was basically "a man sitting on a kitchen table"
better model, more steps, more confidence
What would the confidence be in this case?
if you give contradicting elements in your prompt, it can greatly reduce image quality in your image output
i call it low confidence -> the model doesn't really know how to handle the contradiction
i'm pretty sure it'll be documented somewhere and have a real name, but that's just from me fucking around in SD
more real pokemon ^^
rapidash, cool ^^
lol, got spearow and fearow again ^^
machop and slowpoke
psyduck army lol
articuno hella awesome
moltres 😮
sorry, just saw this. currently using clip g and the ipadapter "pytorch_model" clipvision model in the workflow I'm messing with. I'm still in the testing phase with this stuff and haven't worked out precisely what the difference is in their behavior as far as the render is concerned. I've got clip h to work but it's tricky and had to switch things up. don't really see what advantage it serves other than using slightly fewer resources
is she about to throw stars around?!
she's about to spit some fire and ice
it's fine. I figured it out and even managed to keep AIT
your approach is definitely more precise
the setup I have can really take on different forms depending on the parameter settings
I do have the contrlonet on the last image so it can take control if I want, but it's not necessary. third image is really superfluous and optional
nice, so you're getting the hang of the placement of things too. it's weird how ti works out
I aimed for it to be a universal ImageA+imageB thing while maintaining AIT. I'm kinda confident it's atleast as good as MJ blend is?
perhaps. never used mj. but sort of got mocked when I suggested that I could blend images alright 😞
not ideal imo, CNET makes you loose AIT- which is not something I'm cool with
but yeah, I think this is the way. and over time you'll be able to refine it more. I'm trying to get into the data a bit more and see how it's processed
well if I could figure out how to run a second ipadapter in the same model flow I'd probably just do that tbh. but need to understand it better before I'll be able to figure that out
my main thing was input 3 images and have them all blend. plus it kind of covers all the bases of what a person could want. but you could definitely turn everything down on the third image. even have it set up to input an empty latent rather than the image
the workflow I made seems to be really consistent at blending images while achieving qualities really close to pure SDXL txt2img. this takes about 20-ish seconds per 4k image like this due to AIT. also, can do batch1-4
yeah, that's definitely a cleaner approach
also seems to keep SDXL's quality usually. I took inspiration from your BLIP idea, and it's a key component of that workflow as of now
also the VIT-H IPA seems better than the normal one. let alone it's way more efficient
well I had errors with it at first
I may end up overhauling my approach soon. just have to figure out how and why first, lol
idk, I won't be changing my approach if that means giving up on AIT
I will release this once I'm 100% sure it's exceptionally refined though
well I wasn't planning on making anything other people would use. it just kind of worked out and so I showed my approach to a few people
normally I build things until I don't like them anymore and knock them down like legos
also this approach allows for a styles feature like on my normal txt2img AIT workflow
by the way, not 20 seconds; 16 seconds*
damn. not bad
the thing I made is bulky. but then I wasn't going for efficiency per se. I was just trying to see how all the tools would work together at a certain point. because why not?
really just want to focus on the ipadapter though, at least as the my primary focal point.
fun to mess with something so new
reminds me of monster hunter
these are exquisite
Basically yours
lol, I'm flattered. it all started with those rainbow goblins I think
the OG goblins
Oh yeah, the OG goblins. I think they are also in the dataset
those were some of my first sdxl images
Peak PoP style
all growed up
It happens so fast
you also got the analog film thing going there. that's what I go for a lot
Hey guys I'm kind of at a loss with comfy, I have no idea why I'm getting the AI look, could someone reccomend some adjustments?
It took it somehow from your pictures
that's the video game look
they're on point. how many did you use?
I don't specifically know what that means, as it's not the style or look I'm going for.
25 😅
well you should probably make it not do that
Great idea is there a online discord I could ask such a question to people in the hopes they would suggesst something.
sweet. now my images can be bred in latent space much more effectively. 
Your generation resolution is way too high at 1920x1080, try 1344x768 and then highres fix to get it towards 1920x1080
Also don't use words like Photoreal
testing.
Could I ask why?
photorealistic describes an art style
Things that are tagged like "Photorealistic" tend to be things that aren't real, but are in a "realistic" style
use terms like "photograph of" you can name lens, camera type, etc
"professional photographer"
etc
Movie Still works well too, but then you only get certain types of images
Oh, they will
I just can't stand bokeh

well, it's not like I even notice it in movies and other things. just don't like it for myself
blended duck and ball, then made a space thingy, than blended both. this is so much fun lol
nice
Still getting the AI look, do you see any other issues?
lol
that's isometric
do you have an isometric setting of some kind? style or lora?
Its in the prompt
Not sure what you mean by "AI Look"
You are ignoring me!!!!! Send you a DM btw 😅
I want isometric, I'm asking about the actual look the AI look of the image.
@indigo carbon Hey which IPAdapter node set are you using? There's a couple now.
Almost swirly and little detail and the odd lines:
You tend to get stuff like that with isometric things. Try different settings, different prompts. See what looks best.
I'd suggest messing with the negative prompt more as well
this is what I'm currently working with. I think this beats MJ's blend?
it also maintains AIT, ~16s per image
Ah ok, supposedly this one uses less memory - https://github.com/cubiq/ComfyUI_IPAdapter_plus
Just about to try it out
I don't like this one, it's not compatible with AIT. I did try it, but the chinese one takes the cake imo
My problem is when I use AIT it just pushes it over 10GB VRAM, so it's slower than running it normally.
AIT lowers VRAM, odd.
Seems to make it use more when coupled with IPAdapter for me
than you didn't infer it right. the way I inferred it allows it to blend images while having a styles feature and maintaining AIT
You can with the correct files
I'll try it again, pretty sure I was using it the correct way because it did make it faster with just 1 IPAdapter.
I can give you a pre release of what I'm working with if you want, it works really well. (you can check out the pics I posted here recently)
Yeah I don't mind taking a look, will be interested how similar it is to mine.
I'll DM it to you; people here might try to use it then complain it spits errors
ok
This is what it does when I use it btw, maybe I'm starting it too early in the chain.
Greetings you busy bodies...
Hello
i wish the weight slider lora worked for sdxl, great lora
is this the G and L prompting stuff?
yea
XL has two text processors, CLIP-G and CLIP-L
some people do different prompts in each but usually you just put the same in both
negative prompts
negative prompts
What I used?
It's IPA VIT-H implement with AIT and BLIP
concept art {prompt}. digital artwork,illustrative,painterly,matte painting,highly detailed,by Matt Bors,by Skottie Young
very cool! will you release this at some point?
I will but after I'm 100% sure this allows SDXL to dominate MJ's blend like SDXL destroys MJ in txt2img
So far it's a yes imo, but I'll let it sit for a few days
cool - looking forward to try it out!
The way I implemented here also allows for a styles feature, just like with the normal AIT workflow
I did all the reading of how IPA is meant to be inferred, than executed it in ComfyUI as just a presampler to just make a vague latent that's precisely the 2 images combined, than passes to BLIP to turn into text and the normal Ksampler with AIT, while using BLIP's answer and the unfinished latent with SDXL loaded with AIT on the normal Ksampler. This allows for image blending that so far seems to outdo MJ in some cases, while taking ~16 seconds per image
I love how it just placed the cow on the bench here hahaha
very interesting! nice
It should theoretically work with even batch size, but I haven't tried that yet
I think we could refine the BLIP question though. Currently I have the BLIP question set to: what does this image consist of?
I only worked with BLIP in auto mode
Man. It was asking about the background before. I didn't really know what to put
Default was background question
Certainly room for improvement
if only the hands weren't fucked
We need to come up with an ultimate universal question
Could also use more than one and ask more than one question
question, so does the sdxl model have more refined models? like one for anime, one for this and that?
Not really, we don't want to overdo BLIP, it's a simple model that gets the job done. We just need a good, simple question
I think what I have it set to makes promising results (checkout recent stuff I've posted), but with stuff like BLIP there should be an ultimate question possible
That works too. I'd thought about other models as well, but want something efficient
Like, look at this, man
when i try using the sdxl model on the web ui, it just switches back to the model i had selected before
I'll check out your workflow in a bit and see how it works for me
Easily beats MJ in some cases
Well it shouldn't. They're weird
Curious how mj does it
it shows up here, i click it, goes back to OTO which was last selected
Probably like this, MJ isn't pixel diffusion. the reason this beats MJ sometimes is probably because SDXL is just a better model.
I read that it was pixel diffusion, but I actually don't know anything about it. So thst might be incorrect
Not pixel diffusion, there are latent previews. You can easily tell
Not sure, man. That's an sdxl model though?
that looks great!
Maybe what they used way back was some kind of pixel diffusion, but now it's just paid latent diffusion
Yeah, that's a great example of SDXL doing a better job at blending than MJ
Pixel diffusion does allow for image blending without the need for something like IPA, but it makes a whole lot of new issues, so not ideal.
oto is not, but when i try to use sdxl base it wont let me
Not only pixel diffusion performance is poor, so is the quality. DALL-E is a pixel diffusion, and it doesn't stand a chance against SDXL
what's your console saying?
size mismatch i guess
looks like you are loading a SDXL model as SD1.x
Also this, but not only it's a LoRA, it's also a 1.5 LoRA
yeah looks like you need to update
Also hey Comfy, I figured out IPA with AIT
Just seems so computationally intensive
did you have to build a copy of the model with AIT support?
I mean a copy with IPA?
Nope, just implement it in a different way and made a new workflow for it. So far it seems to outdo MJ blend in most cases
I won't use the word "dumb" but I kinda want to say pixel diffusion seems to be that usually
pixel diffusion is just not efficient
And not as good. It's old tech
it makes more sense to compress the images and do diffusion on that compressed version
Only good thing about pixel diffusion is it would be easier to blend images with it; but I seem to have found a way to blend images using SDXL
Also, I did try that at first, but the first step for my solution was to switch to IPA VIT-H, and that's compatible with AIT and better than the normal one. Then made a workflow that uses that in a smart way (see stuff I posted here) and that's pretty much it
Takes about 16s to do this, and the results usually beat MJ's blend
Also allows for a styles feature like I did on the normal AIT workflow
But honestly, I will need to take a damn break before I start working with OneFlow; haven't touched that yet, and I don't really want to. Speed is already more than enough with AIT, I have a feeling it will be a nightmare to do with OneFlow
nice work! love it
I think you need to install python 🐍🤷♂️
I know a lot of you have helped me out with stuff and I shared recently that I had a tagging project I was working on over the weekend. Well, it worked out well enough that I wanted to share the ComfyUI workflow with more people on the off chance that it's useful to someone. So I created my first non-image post on Civit with the documented workflow where anyone can pick it up:
https://civitai.com/models/147873?modelVersionId=164971
Thanks for checking it out if you happen to be the one other person in the world that might use it other than me. 🙂 Also, thanks to many of you from this channel for your help and discussion as I figure all this out.
thanks for sharing it, soul! looks like really interesting workflow tools / components. nice work!
If it helps someone in some way other than myself, then I'll call that a win.
share workflow dood 🙂
any tips on generating seamless backgrounds like this
like simple mountains or clouds or far away trees with sky etc
there's no chance anyone has already build a patch to fix this, right? 😉
any good model to generate fantasy seamless backgrounds?
@twilit nexus there is in A1111 way to do seamless tiles, but in all directions, so mountains would cause trouble
well horizontal would work
like sideways
but need a good model for generating fantasy bgs
like the far away stuff
snowy desert etc
need for my game 😄
you will see. Only works all side tiles, probably you can do similar renders, and paint in some painting program inbetweens?
keep in mind: https://www.reddit.com/r/aigamedev/comments/142j3yt/valve_is_not_willing_to_publish_games_with_ai/
meh hobby games not going steam
SDXL Artist Style Studies: https://sdxl.parrotzone.art/
Ha. That just popped up in my news feed like an hour ago
more stripes = more adidas 🙂
Catching proper ones
Bookmarked.
Bookmarked as well, thank you!
last i know. Bad is it is sorted by given name 🙂 Why! 🙂
this is for 1, but i believe tags are same, yes but results are different i forgot, anyway.
https://proximacentaurib.notion.site/2b07d3195d5948c6a7e5836f9d535592?v=b5b75a67cc52483c9965cfc141f6f582
For information about this project see: About: Image Synthesis Studies Database 🦜
Navigation:
- use Ctrl+Shift+L/ Command+Shift+L for dark mode
- type a tag into the searchbar to see all entries associated with it
if you want to recommend something you can use this form: 🎱 Modifier Recommendations
You can find backups of this database in this...
this is as well for sd, but artist do not change styles, shouldnt.
https://www.urania.ai/top-sd-artists
A comparison of all 1,833 artists in Stable Diffusion. Each artist is tagged and 4 generated images per artist are shown.
nice - thanks for sharing. bookmarked :). many SD1 styles/artists tokens are also giving good results with SDXL
found this artist there
A lot of these sites make the same prompting mistake, e.g. look at Scott Adams. That's a bunch of portraits of the artist Scott Adams but not even in his style! He's a cartoonist. When you push the model to perform a certain task that doesn't align, you can blow right through learned traits and get vanilla behavior. This is especially true if the training is weak.
i know how aply artist, how to apply medium?
they're not training it on these artists are they? they're simply using, I"m assuming, a prompt template with the artist's name inserted into it
just talk to the model :). painted on, made with etc
Quilling for me as English itself is very challenging, difficult but i think i got it
this isn't a new issue, and I don't think it's indicative of weak training. it just means the ai is misinterpreting what is wanted and creating an image of the artist themselves instead of their style. could be indicative of a suboptimal prompt template though. but I'm sure they considered such things
yeah, you're absolutely right. this was always happening and most sites have images like this. I browse it for some inspiration and go from there but I always keep that in mind.
Weird wonderful did okay!
true
for example Jakub Schikaneder is well known for painting old prague only. In buildings it is something he wouldnt ever paint, but some sort of his style is there
"art by" gives good results for many, others lack coherence and need additional help. That's how I got my "weak training" assumption.
also @cursive warren did an amazing artist study for SD2 back then on this server: https://discord.com/channels/1002292111942635562/1047177083601813504
I think just by can be even stronger with a lot of tokens
I ran into that issue alot with 1.5. lovely image of bob ross, or andy warhol, or whoever else
I've been experimenting with weakened prompts that encourage the model to take a garden path for a particular artist, e.g. landscape, character, cityscape, abstract, animals, machinery, etc.
Promising.
art by, design by, directed by or you can get even more specific like clothing design by, set design by, lighting by etc... it all does have different effects. and sometimes just by... is having an even stronger effect - especially with SDXL
yeah prompt engineering is on a whole new level with SDXL. the storytelling capabilities are so much stronger now
I finally got around integrating prompt editing in ComfyUI. one of the features I really missed from a1111. I used it a lot for TI embeds with SD2
i do artist art by somebody.
What i dont like on most those pages about author and styles is small detail, it is sorted by given name, would prefer surname 🙂 just little bit unusual
yeah. most would really profit from a tagged database with sorting and filtering options
mona tiger warhol!
and i see there A. Hupburn 🙂
indigenous lisa?
@high skiff did you make that SD1.5 funko pop lora? i see your name, so assuming, when sdxl version 😉
Yes, I did, and I have no idea, I'm taking a full break from AI at the moment. I have hit a sort of awkward wall stuck in between 1.5 and SDXL, and I'm not expecting to return particularly soon
Greetings, Sytan 🙂 nice to see you
got some MJ maxi talking shit about how much better it is than SD
I got extremely discouraged after not finding a way to fix the issues with my high res fix V2. It still offers the best upscale quality I have seen for SDXL, but it has a lot of caveats
For some styles, it's issues are actually beneficial. It is insanely good at upscaling water color. Can easily go to 4096x with it, and it looks stunning
Well it has advantages and also disadvantages over SD
true, but i still wanna argue with him lol
dem hands
perfection
What makes MJ much better than SD according to him?
it's great for casual folk. but it's like having someone else drive you somewhere. you have no idea what's going on under the hood
or how it gets where it's going
here is an example of how amazing my high res fix V2 can be when it works. left is the low res, middle is my old high res fix 1.0 and right is 2.0
This new high res fix is insanely good at adding new texture information, but for some images, that's the opposite of what you want
Casual folk or actual artists for one obvious reason but for the latter one Firefly is the absolute number 1
And not MJ
Has Firefly gotten less shit since I last saw it? Cause it was horrifically bad when it first launched lol
some of the adobe ai stuff is very helpful, but firefly is so restrictive
I have yet to see anything good come out of firefly that doesn't look like Dalle mini could have done it lol
It never was bad tbh
What
The watermark was annoying tho
give me 15 seconds, I can prove you wrong lol
But it was during beta only
the firefly discord did not like me naysaying the watermark, lol
As soon as Firefly was integrated into Photoshop and Illustrator that was it
some of the actual promotional images for firefly when it was first released in beta
they also made sure to tell me that it's also in the metadata
looks like dog shit lol
Also, wasn't there that whole thing about Adobe stealing art from peoples cloud's to train firefly without their permission?
No
I mean thats what some folks claimed
about 2-3 weeks ago
What was trained on was their Stock and non copyrighted images
There is no world in which Adobe is above stealing from their users. They do it all the time lol
hell, they stole from me back when I used them years ago
well they tell them that's what they're doing
They didnt steal from you, it was your mistake what you did
wait a minute, I just recognized you. You're the guy who I got into this with last time. The troll
The guy who thinks that Adobe charging me for 6 months after my account was terminated was MY fault, and not theirs
You subscribed to their product and wanted your money back after the deadline, wasnt that the case?
I've seen several reformed trolls around these parts. don't recall infidelis trolling though
no, I had my account deleted on purpose cause they kept double charging me, they "reimbursed me" with free sub months, of which I never actually saw. Then once I deleted my account entirely (can't even sign into it anymore), they kept charging me for 6 months, to which they then told me to basically come for them in court
I had to get my bank to sever the transactions, or they likely would have continued
never got my money back, so fuck adobe IMO
and I am not the only person who has had similar situations with them
I used them religiously for like 3 years, even got people to join, so its not like I am some hater, I used them daily for a good while.
Then they started fucking me over, I deleted my account, and they continued to fuck me over
this looks so generic
firefly was 🤮 when it was in beta lol
How could they double charge you
I'd hope its a little less shit now at least, I have not seen it
they would charge me once, and then... charge me a second time. Happened 3 times. They then told me they gave me 3 free months back as reimbursement, of which I never actually got, which lead to my deleting my account
its prob like apple there are other models out there that can do the same thing but ppl will continue to pay for it because it has adobe brand
for sure
No, there are multiple reasons for why people stick to it
I am just glad most people are off their firefly high horses praising Adobe for being "ethical" with their training data lol
substance products will come with creative cloud one day 
Man, I remember how much I loved substance, then Adobe ruined that too
No, that would mean they would have to higher up the prices
not for just a little
They raised them now with FF out of beta
For me not so much
And im not affected by the credit system for generations
The free users are dead
ok, they raised the monthly cost by 10%, starting November 1st
Depending on subscription
For me it should be about 5$
Not sure tho bcs i pay in€
its 10% for all plans
Oh, this is SDXL channel
starting November 1st, all Adobe plans are 10% more per month
Didnt do the math, but eventually
they announced it themselves
Right now im at 60€/month
its because they wanna help poor artists like me
lol, as if
Well their tools are oriented towards professionals tbh
Adobe is now $660/yr for prepaid annual, fucking hell
i should see if i can get my grandfathered plan back which was like $30 something a month
not likely
professional tools and dont even have alternatives to controlnet or animation models
they will probably tell you to go off yourself with a smile or something lol
idk, ive had it for a few years now for around $30 a month
Wdym
i use adobe products heavily
exactly what he said
never tried firefly or w/e their ai thing is called though, no reason to when there is SD and MJ
I used to as well. So glad I don't anymore, switched to much more reasonably priced options for myself. I know thats not the case for all people, but its been great for me
For 2D animating Animate and After Effects are the way to go from their products, for 3D its someone else
For me Max as generalist tool
My friend works as an AI dev for Unity, and they have some new tools coming out for AI that are insane. Looking forward to when I can share more info really
i started 3d with maya but then realized that i have licensing issues if i don't pay for maya so went to blender
Dont have to as they have different target audiences anyway
too bad almost everyone hates unity atm
I don't even like blender, but I would still say its one of the best free alternatives to paid software right now
Yeaaahhhh
My friend was blindsided by that as well
Unity is fcked up atm
He handles AI dev for Unity, not pricing or anything like that
And they did it themselves
yuppp
yeah my brother doesn't use SD or MJ and thought FF was amazing and was showing me some stuff one day, he was showing me inpainting with it and i show him some SD stuff and he's just like, wtf???
For a lot of folks Blender is enough anyway
yeah, agreed
I use to be big in Blender, but now I don't like the direction they have been taking it. Just cause I don't really support blender anymore doesn't mean its not a good alternative tho
(TELL YOUR FRIEND TO LOOK AT GODOT!
)
you can do some amazing things in it
Unless GODOT is looking to poach him, I think he will stay at Unity lol
blender is best NLE, but hard to learn. Also check blender support, now it is supported so many big names.
should show him outpainting,ip adapter,openpose ,canny or lineart should be enough for him to ditch firefly alltogether
yeah i hear there is some big tech hiring freeze or something atm anyway
FF is by far the best for my workflow, for others that doesnt work out well just as SDXL doesnt work out for me and MJ either. People have different workflows and what they need
Oh, why is that?
I have my own issues with blender, specifically with the new features they are adding, but it is still cool to see, regardless
im sure he's be on board but he doesn't have time for it
he'd gladly tell me to do it for him though 
i was showing him stable audio yesterday and i felt bad because he makes music with FL
Not really depending on what he does etc
wastin money is one of them
I mean lets be real, Firefly is NOT oriented towards prompters and "AI artists" just as SDXL isnt oriented towards artists
I was huge in the blender community for a few years. Like top 10 users in the biggest blender server on discord big, but I started to notice that blender is not focused on fixing things, just adding new and sketchy features not many people ask for
Like, blender has horrific optimization issues, and yet they are adding in all these new systems rather than fixing the old ones. All in a race to be the "first" at something, while doing it generally sub par, in order to get more grant money from deep pocketed investors looking to paddle their GPU's for rendering (NVIDIA specifically)
Its just a big sprawling bunch of roots that don't mesh or have consistency or reliability between them. Simulations are miles behind geo-nodes, but you alsos till can't do proper bevels or chamfers in blender without its poly engine absolutely shitting itself.
I just dedicated so much time to trying to find community solutions for issues that should not be a thing in blender, and got tired of having my bug reports dismissed as "we know its an issue, we won't fix it"
sounds likes its oriented towards ppl who know nothing about sd capabilities and just wanna generate some stuff on top of their art
Whats NLE
non linear editing of video
Oh, thats something their fanboys wont admit and will defend Blender by any means necessary
I mean dont get me wrong, Blender is amazing
But those people laughing at 3ds Max and making Blender look godlike compared to it is pathetic
I mean those fanboys
Not the devs
brilliant
those fine lines in the eyes 
Or they simply dont need those capabilities at all
ofc they dont, they are using firefly
The one that was the nail in the coffin of me using blender was me, and various other people working on a bug report for blender (the guy who made blender Super Image Denoiser is my friend and helped us with this)
I spent weeks finguring out an Albedo denoise pass issue for image denoising, tracked it down to an integer flip point where blender switches from casting rays through transmissive materials, and reflecting them. Below a threshold I was able to find down to the 9th decimal point, the denoise would fail. increasing it by any amount would allow it to work
Turns out it was an issue they had known of for basically a decade, when cycles original dropped. They informed us that they knew it was there, and why it was there, but that they would never fix it, even though it ruins peoples renders all the damn time.
We spent so much time trying to find the data, explain what was wrong, worked together as a community on it for a long time just to hear "Even though we are remaking the entire render engine pipeline from the ground up, we have no interest in fixing one of the biggest rendering bugs in Blender/Cycles history" and then close it with no additional comment
@cinder aspenmeant to reply
I can only imagine how frustrating that felt
a pikachu cat will make it all better
my specialty in blender was fluid and glass renders, so low roughness denoise bugs were my bane
I found the example image
you can see at 0.273861, its broken and noisy, but above, it fixes
bugged denoise albedo
bugged denoise normal vector
Why was that their response?
I mean why wouldnt you want to fix that
no idea, they just said they didn't care to fix it, then closed it
and looking through reports, many people had reported it in some capacity
I learned of ways to cheat it by render pass swapping, but it requires rendering every frame 2 times, and is a pain in the ass to automate
Lack of support from Blender stuff is also a thing due to its nature, one of reasons why others will prefer Autodesk, Maxon or Adobe products
Especially companies
hm speed, for glass can be luxcore renderer or raytracer nice, cant be? @high skiff
I can call at Autodesk whenever i want if i need something for my 3ds Max
I used to dabble with luxcore, but it had a lot of issues back when I did this as well, this was like 3-4 years ago by now
I used blender a few days after 2.8 dropped, and now we are on like 3.6-3.7 now I think
Ofc on the other side Blender has a wider community of hobbyists
And Blender is free
Max isnt
Let alone addons
Nice 😄
what means corporate gold?
https://fund.blender.org/
Adobe is gold 🙂
thats pitifully small for Adobe 😅
There was kinda a panic within the Blender community when Adobe funded Blender
As if they would acquire Blender
yeah, it was unfounded, the creators of blender kinda hate adobe lol
i wonder why
It wouldnt make sense for Blender to do thst
Adobe being Corporate gold is literally less than 1 minimum wage salary lol
But there is a joke around where Autodesk acquires Blender
And the logo changes
Autodesk is Adobe of the 3D world
if Autodesk doesn't start adapting AI features very quickly now they will get into trouble not too far in the future
They already do
Wait what
Didnt you see whats coming for Maya
NVIDIA doesn't fund blender anymore? lmfao
they will prob attempt to buy some of them and boom they are back
bro, NVIDIA dropped their blender support
guess I missed that siggraph talk 😄
Plus partnership with Wonder Dynamics
I wish we could see how much each company donates
For whatever reason they implement those in Maya but Max isnt talked about for now
Adobe is 30k a year, which is basically nothing for a company their size
these are the real blender supporters
those are all at least 120k+
so who knows, could be 120k, could be several million. We will never know lol
yes i believe they are giving more
I just find it hilarious that Adobe would be throwing out pocket change to blender
Like, I think such a low donation amount makes them look worse than just not supporting them
V13 of my workflow - different node set for IP-Adapter, updated instructions and links, various smaller improvements:
https://github.com/JPS-GER/JPS-ComfyUI-Workflows
imagine giving a company that likely leads to a TON of substance sales and other subscription sales just 0.000017% of your yearly income lmfao
I guess something to do with Substance
But who knows
Blender isnt someone who they have any ties to
this random guy donates a minimum of 4x what ADOBE donates every year lol
I cannot tell you how many people I know who end up paying out the ass for substance cause they use blender
I can name 7 people off the top of my head
They dont earn as much money from the Blender community as they do from the others i think
and then after effects goes very hand in hand with blender, and most creative individuals looking to do renders retouch in photoshop and lightroom. I am willing to bet that blender is a HUGE marketed for adobe
I mean the more professional artists for sure use Photoshop and Substance
within Blender community
But they also use alternatives to Substance eventually
as long as I have been in the 3D scene (about 5 years now) I have yet to see any real alternatives, unfortunately
maybe mixle or whatever its called, I forgot
I mean alternatives, if those are better is a different question
Substance is king if you ask me for texturing
Mari for movies
Ah, I merged the name lmao
Substance for everything else
Quixel Mixer
or why not Affinity?
I love the creator of affinity
that man is a king
Affinity's HDR tonemapping functionality is god tier
Not enough for many pros
and its auto panos are too
Especially when it comes to AI features
And they also eventually dont even have to pay for PS anyway
Their company pays
Unless they are freelancers
I have been using kirta this last week for digital art, and I feel like its actually a much nicer experience for me for drawing, feels a lot more purpose built, and it has a lot more granular control. I was able to control all of the sliders on my color picker, add as many as I want with as many values and math function as I could ever want. Was able to build my perfect color picker from scratch
also VR, there is sculpting in VR, adobe acquire medium, so i dont update it, and using very rarely shapelab. One must try it to believe it 🙂
Krita is more streamlined for art than Photoshop but that makes sense
I use adobe Medium all the time. Its the only adobe product I still touch, cause it has none of their stink on it. Its free, works great, and is reliable, the opposite of a lot of adobe acquisitions lol
i have painter and rebelle. Rebelle is superb.
Also, my bad, there IS an alternative to substance designer. Its much better and more poerful
but it costs even more lmao
oh, looks like prices actually came down on it
Marmoset Toolbag. Its the big brother to substance
Im glad Substance got an update on baking
Overhaul
Otherwise i would have to use Marmoset for baking
Affinity did made really great progress. if you work with print Adobe still offers the safest and most efficient workflow and is still the way to go - color profiles, pre-press, pdf compatibility, spot colors etc.
I had to streamline my pipeline at this point, i use way too many softwares lol
Too bad you can't use pantone with Adobe anymore 
also there's no substitute for InDesign
RIP reliable industry standard colors
You can...with a big catch to it
you still can
yeah, an additional monthly fee and a limited color selection
I am shocked a company as predatory as pantone would have cut ties for most users for pantone, but then I also realize its a smart predatory move to get more money lol
I don't like it as well but I have clients that have pantone colors so I gotta pay for it anyway.
For professionals its not as a big deal as it is for hobbyists
of course it sucks that I can't open my projects if I don't pay for a subscription. that's not the best feeling in the world
Pros easily earn money back from their investment in Adobe and Pantone
I as a hobbyist dont earn money yet from artworks etc
yeah. If I can't pay for my yearly Adobe cloud subscription I better look for a new job heh
Thats the Adobe Experience ™️
Luckily i have a full time job that allows me to spend a lot of money on creative stuff
compared to the costs of the old Adobe Creative Suite, the subscription is cheaper
rich ppl get richer by throwin away money
I currently really need a job lol
wise words - doesn't work sadly for me 😄
Yes and no
Werent you close to get into AI job?
just get sdxl to make you pantone colours. they are super accurate 😄
Cuz when you bought the old sweet, you had it forever, you instantly lose access to all things Adobe the second you stop paying, so by that comparison, Adobe is infinitely more expensive now
Subscriptions usually cost more...long term
*old suite
$2.5k for the creative suite and than $700-800 for each update
Also, more times than not, they force you to bundle in programs that you likely don't use in order to make it seem like a better value, when in reality it would be cheaper to just bundle the few that you want
yeah. that's their trick 😉
Im at 60€/month now just for Creative Cloud, plus 19,19€ for Substance texturing bundle
yeah substance should really be a part of the cloud sub, but now they also want firefly credits heh
Let me tell you tho, 3ds Max costs me by far the most due to all the addons i pay
Money money money
yeah
is affinity publisher so bad? Or just colors accuracy?
I'm not doing a lot of 3D but I've seen the prices
I think Auto desk is the only company that gives Adobe a run for their money when it comes to predatory and overpriced behavior
I have indie license for Max, not standard tho
it's pink 😄 we need SD-RAW-HDR-CMYK
I pay 350€/year for Max which isnt expensive comparewise...but then the programs around it and addons come into the game
yeah if you work with that or are an enthusiast that can be acceptable
As long as i can afford it, yes
subscription services are not my favorite concept - but for many things especially in Adobe tools there are no substitutes because they have the monopoly
I mean i could have gone with Blender as well, i did for a very short period of time
Yes, and tbh it makes it affordable for me
Imagine if i had to pay all of the softwares and addons at once
Full price
My as* could i have em all then
I would have to sell my mother, my soul and my left kidney for that
yeah and you get updates and new features without extra costs. at least that is one positive aspect of a subscription
Yes
Ideally you also get your money back with those products
Which is what professionals do
I mean - I was totally against it and still am 😄 but now I'm logged into the Adobe ecosystem. I will never be able to work on my projects if I don't pay them for the rest of my life
Im in their ecosystem as well including generative AI
You may also understand eventually why SDXL is a horrible pipeline tool in my case judging by my tools above
Its not about the image quality because SDXL can produce good images
Yeah as said - there's no real alternative for a lot of stuff - especially if you work in professional print, image processing. affinity really improved so I have to check it out again, but all those tools have to integrate so many stuff from the last 25 - 30 years.
GIMP may be great but I can absolutely not work with it because how it processes colors and files. you can't output an industry standard file with it.
now print it perfectly, imposible 😄 @west breach
Yes, although for a lot of folks they can get away with Affinity or even GIMP easily
But at some point you dont come around PS
I'll be sticking to Krita, never gonna use PS again heh
Krita was my bae
yeah, of course. I have set up GIMP for others so they can make their social media images and teasers. but it's too limited and many stuff is just integrated in ways that don't make sense to me.
I even installed it once on phone but didnt turn out well lol
yeah krita looks really nice too
gimp is counterintuitive. But surprised by inkscape, how it works. It is not so bad. And i have last version of non subscription based illustrator.
I dont like its interface
Ahhh yeah, I run it on my samsung tablet when on the go. Its super cool to have full desktop krita on a mobile pen tablet
yup same 😄 I go crazy in GIMP
I have an ultra budget high quality drawing setup, and I love it
For tablets its a different thing, yeah
I have a samsung TabS6, which has the full Wacom digitizer and functions. And then I use an app called Super Display, and for $10, it can turn your tablet into a super low latency high quality external display with pressure sensitivity pass through and many other features
https://cdn.discordapp.com/attachments/427011414851321866/1151326565569937478/20230912_181817.mp4
Here is a video of my uncle trying it on my Aunts ancient Samsung tablet. The slight choppiness is cause the tablet is over a decade old now lol
nice! yeah super display is great
ithas macros, touch pass through, functions, screen cropping, all sorts of stuff. Its so cool
for $10, its a must have. Worth every penny
it can do wireless as well, but I have never tried it
with a drawin tablet + SD u can truly create anything u want/have in mind
I have this @high skiff https://www.amazon.com/XP-PEN-Artist15-6-Pro-15-6-Full-Laminated/dp/B07M5X7MH1
very nice, very nice
thats the one i have i was gonna buy a wacom but then realized wacom is like the apple/adobe of the drawin tablet world 😅
it is for its price doing what it should o.k. Wacom is Apple between tablets
😄
didnt see what are you writing @glad grove
My tablet is Wacom, but I have had Huion and Xp pen, and they really ar eno different lol
Pro is pro, iam amateur 😄
Glow seems familiar, but supposing all brands got same
I use it wireless and it works great :). PC via cable, tablet on wifi
Here is mine
The latency is genuinely impressive, for being on a tablet though a USB. Both input and output latency is solid
nice. Laminated?
USB-C, my tablet is powered only by usb-c and i am on edge with psu, 3070 + 3700X + all other things, and i forgeting plug it in wall, but it works, just powered by only PC throught usb-c
Mine is an android tablet
corel painter?
Krita
its a samsung TabS6
I am constantly shocked at the latency
it can do up to 4k120fps too
is it amoled display?🤩
yes. right. I was bit confused. Too much layout presets and they look very similar
my tablet runs 1440p@60hz
yessir
Its
to draw on
196% sRGB gamut
tho I run it in accurate mode, not saturated mode
and the induction mesh and pen are both made by wacom. 4096 pressure sensitivity levels, velocity support, tilt support, hover support. Its awesome haha
i know workers in soviet union worked on those numbers fullfill 5 years plan 😄
and its so cool, cause it runs android
pantone has an interesting effect when you add it to prompts
So I have an installation of full desktop krita I can also run on the tablet itself on the go
krita porting their desktop app to run natively on android is such a great idea. It has all the functionality of desktop krita, in a tablet form factor. It makes apps like procreate, clipstudio, and the others look like a joke when it comes to functionality and capbility
Even if it is just as clunky as a desktop app lol
pantone road train 😄
THE PANTRAIN
Microsoft Edge just added rounded borders to how it displays all web pages 😵💫
is that lil pump?
mona lisa/ gang member
pretty much the same thing
let see what a picture of lil pump will do
did you like the picture I posted earlier?
idk lil pump/mona lisa
@hardy cipher I have to find it I posted and left,do you find the frogs?
the first one looks like Deborah Harry
the lego ones cool too
Do you describe the pictures in the T2i system or leave it blank?
with the ones I've been making lately?
like in general ,you think it makes a difference
I have a blip model set up to interrogate them. send one to clip g, one to clip l. then send them through styler nodes
yes, it does make a difference
I have the styler nodes on a separate system, the one Ive been using most is MiDaS Node System
ahh. I was having issus with that recently. probably could have solved it but just got impatient and did other things
poor runner 🙂
bigfoot vs smallfoot
@hardy cipher MiDaS node was broken they just fixed it with new update
this could be interesting image, but what a arms 😦
ahh. well that makes sense
I should make a midas node. probably isn't that hard
or even better, some other kind of depth map that doesn't get used. and probably sucks for controlnet
ckassy lad
Putem in a blender
like that classic eve 6 song
This one was weird :Pepsi /David Lee Roth
K
this is large nobody 🙂 so large nobody is somebody.



