#๐๏ฝgeneral-with-images
1 messages ยท Page 68 of 1
are those grilled cheese?
toasted sandwich yep
hoooly crap. I accidentally came across an AOC lora and boy
didnt think I liked this chick until I saw her potential ๐
AOC?
i cant link it cuz ima git banned
as in alexandria aoc?
yh lol
๐
the pepperonis look like little bowls
not getting the cheese pull though ๐ฆ
thats the least appetizing pizza ive seen
Hamburger pizza
dont get me wrong, I love eating high calorie sludge no matter what shape or form. but that's just amorphous
The FBI agent watching your internet search history
and I cant decide if those are leaves or bellpepper slices
no it is simply hamburger without top
did you know that a hotdog is technically a sandwich by law in an american state?
and so is pizza
and tomato is fruit
a hotdog maybe but not za
with this lora i might be able to do cars at night properly now
yea it just works
now Im curious to see how many other good loras ive been missing out on
if this works in conjunction with the darkness lora ima nutt o:
Detail Tweaker
that one is only for anime isnt it?
no
no goes amazing on everything
o:
I've merged it into most of my mixes, they went up in quality immediately
i like movie grain, just not sure any slightly compression do not smears it out.
Someone needs to make a filters script to add stuff like film grain, or other effects to pictures after the generation
you were warned
I love what it did to the b ๐ bie textures tho
now to play around with these and mix them up o:
wym by mix them up?
oh thats what you mean
these are now useful tools to further make my already good prompts even more insane ๐คช
im already mixing the detail and lowra
the result is this
daaaamn
i wanna lick the chocolate rn
the cookie looks good although the chocolate looks too waxy
when it's completely dry maybe
Freshly baked chocolate chip cookie, close-up, gooey chocolate, golden edges, soft center, tempting, natural lighting, dark background
Lowered McLaren P1 coming through!
damn...
thats the best spider-woman i think ive ever gotten
and another damn to that car
i want one
it seems people here cant make up their minds as to what spiderwoman is supposed to look like
why?
i mean all the spider-women i have done look different from each other
ill send a few
not the best selection i could give
why i got explicit message, when it is just comics like picture?
but im not going to go through 3000+ images to find the best :-)
my point is I guess there is no consensus on what spiderwoman looks like
We all know people don't care, in their minds the only thing that matters is if that she's hot. LMAO
it is spider woman generated by stable diffusion, no piece of skin on here.... ๐ฆ
oh we cant send that sort of stuff here
i am telling you, he is covered from top to bottom, no face nothing. Why why why
yes
Making me wanna go play modded Marvel's Spider-Man...
man im really really really really loving this detail lora
u can mod that?
ik it just works so well
on pc yeah
now im going back and redoing all my old prompts lol...
Suits, animations, characters, combat, you name it.
is it worth buying it
cuz with this gpu I know it will run good
what gpu you got?
3090
Very worth it if you want to mod. I turned the whole game into an Andrew Garfield and Tobey Maguire experience.
that game uses a ton of vram
By the way, have a Dodge Viper.
?
oh right
yeah as i said
discord is just weird
hmmm i wonder if the detail tweaker thing can add detail to stuff like burgers and other food stuff
what I wonder is if the detail loras can be used in conjunction with CTU
cuz if it can, I think we just reached endgame
would like to have focus bit more up.
there are caveats tho ):
I noticed sometimes if I use them, I get a completely different pose or even person
try it in the after detailer prompt instead
what is that??
nonestly just looking at all those settings is driving me away
fuck am I gonna spend 2 hrs twiddling with all those nobs

even has a hand detector
a more perfect balance
adding the grain lora on top made her look more like uma thurman
tired as balls rn but tomorrow I will try using the loras only during the upscale in the CTU prompt o:
I wish you could switch models to an inpainting model before doing the inpainting with after detailer
something went wrong, no tiles, just ordinary render
Not sure if box ring or coronation place ๐
still looks good somehow
yes @earnest dune
Zoo garden with the animals and another cross creature. Zoo is in english supposing Zoo?
this is right
golden nintees twenties ๐
yes nice, i just used
ektachrome photograph and it added some tone
@earnest dune having weird results. but put more weight on creature
i got only on first line
Daisies ((creature))
what about model? And DDIM but it shouldnt matter
there are not always creatures, some sort of seed fishing
You dont need the other lines as deep depth of field, or Ectachrome photograph, but i like them
also i noticed probably smaller number of iteration.
with model?
using correct VAE?
what are you trying to get?
it is probably the best option.
like a daisysaur
so... you got cutesty creatures bernix
i maybe know reason
i am not using xformers
does it affect the imgs
no it doesnt
but using --opt-sdp-attention
instead as i have 3070 and was suggested for 3000 4000 cards.
@hearty karma are you sure?
yes
if its something to do with the seed then its not going to act the same on different gpus
it impacts image determinism
also i know very much depends on resolution, sizes of images and layouts
xformers will likely result in similar enough first picture but the 2nd through 4th in same batch are increasingly disconnected
torch compile does the same thing
flower woman :-)
now its just a woman with a wreath sort of thing on her head
Hey, i've been messing around with controlNet - was just wondering where I should go to ask a quick question about the weird ouputs I'm getting.
Didn't want to put it here because it doesnt seem like the right place but i hope yall can point me in the right direction for community based help for SD
thanks ๐
#๐ค๏ฝtech-support probably
AI
I wonder if the alternating keyword thing would help mix daisy and creature together
[daisy|creature]
what is your gpu? @earnest dune
o.k. so not that xformers alternative
or try interrogate that image, what you like and see what it will produce
i dont know but flavors.txt is sort of words used for interrogation.
Where do you have it?
o.k. i will try it then
a small bird sitting on top of a rock surrounded by daisies and daisies flowers in a field, Chris LaBrooy, behance hd, an ambient occlusion render, photorealism
but there should be as well negative prompt which is empty
no when generating positive prompt in A1111 it doesnt generate negative prompt. Main is:
extra limb, extra fingers, extra leg, extra arm, poor quality, bad quality, low quality, extra horn
o.k.
But i think negative matter same as positive
some things simply doesnt work. Creature from LEGO is working from Nodlles is working but not from flowers very...
try
creature by Arcimboldo ๐
yes he mades portraits in 16th century out of foods and things
Having now better results putting in negative prompt daisies on ground
you must use negative prompt as well to get similar result. Try it
o.k.
It is good for not creatures, but on result i believe negative prompt has effect as positive prompt, in 2.1 maybe even more
Save the yellow bunny 
yes
you are using an SD 2.1 model with SD 1.5 embeds somewhere most likely
it's not easy tbh the model creators have to tag it properly
model usualy 768 or 20 21 15 14
i think model what you running is from its name
that you downloaded is 21
i never touched embeedings, so i dont know
stable-diffusion-webui\embeddings
i have it empty
so shouldnt be problem
are you trying to use Preview?
Live preview? when it generate image?
yeah, any of the generation preview windows tend to not work with 2.1
i think he means, that when picture is generated that preview. Or not @oak osprey iam confused.
that is very cool ๐
yellow pasta creatures are working still...
give me all of your creatures
STOP. i believe you heard, give me a lot of your creatures. what i said was, give me, ALL, of your creatures.
20 and 21 need probably more gpu VRAM. Maybe back to 15?
will perform today probably.
Lora, will never tried it probably ๐
i mean 1.5 models SD 1.5
no need ๐ SD2.x is fine
is that 2.1? it seems like it, due to its coherent understanding of prompts
i have to throw keyword after keyword with attention++++++ modifiers+++++++++++ galore to get it to do certain things in 1.5
i think so? idk which sd im using and where to look at it
loll
if disabling a LoRA helped, chances are you're on SD2.x now and you had a 1.5 LoRA enabled
yep probably that
the second looks like one con in James Bond
if detail tweaker is an 1.5 lora
@earnest dune try "Matthew Broderick" prompt and if he looks like the real deal it's 1.5 and if it looks like a Wish.com knockoff version it's 2.x
such a good way to understand which sd youre using
i bet you have artius in top right corner ๐
who me
yes i love it
and it is 21
you would love junglerally's digital-diffusion more
realism engine is also great but requires more prompting work than digital-diffusion does, to produce the same 'magic'
is it from hugging face
artius and DD trade blows on which produces better output from same prompt/seed/keys
civitai is easier. And prefer safetensors
@earnest dune yeah junglerally/digital-diffusion
CivitAI isn't easier. Transformers AutoModel class just loads this stuff directly from HF
can you tell me the difference between .ckpt and .safetensors
ckpt is a pickle file and safetensors is a diffusers specific format that lets you ensure it won't run unsafe code
thank you
sorry that's a members only Medium link, this might be better
adorable ๐
fruits works
my prompt was stable diffusion as a person and i guess SD has horse legs
it's kind of incredible that pickle was ever used for publicly sharing models
what a bunch of idiots
and conjoined with a horse
blueberries
There are as well software on github for checking if ckpt or pt are safe
daisy hamster!!
๐
still life lemon?
looks like it
i am getting darker lemon, ok i will try more light
i just got that result from typing lemonade world lmfao
very inspiring words
AI
what a refreshment
wait what did you mean by still life, the art style or the one that is used by making lemonade
got this from a songs lyrics btw
art style. painting still life theme
it is not style. can be cubism, can be anything
A video changing tshirt colour while walking in a city roads
Currently, there is no public bot on the server that generates images. There is an experimental bot available for early server members & Stable Society. You can obtain the Stable Society role by winning our weekly events #โญ๏ฝpow-info #1087493421209485393! However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610โ for more details! You can also stop by โ #1025467151206854736 for any issues you experience while using DreamStudio or โ #๐ค๏ฝtech-support for any problems you encounter while installing it locally!
weird
wow
AI
i really want to eat this
yes nice
smurf's ice cream ๐
in shine of reflectors ๐
It is photo like from movie making
LowRA at -1 makes things extremely bright ๐
nice abstract
Crystal Caves++. --Exploring ancient underground passage with shimmering stalactites and glowing crystals--
Looks good
kamikadze snowmen ๐
Soooo, the more I crank up "black" in the negatives, the more of the image ends up black... any idea how to communicate that I don't want 90% of the image to be the color #000000 to the AI?
i think you are using too much (()) and [[]] can imagine can causing issues
man, i found an issue in my code 
i wasn't putting the positive prompt properly to the tile
now it looks so much better
you was right, too much (()) caused it too be like too much CFG, even though I thought it was on benign tags
yeah you uh, need to learn how prompt attention works ๐
better
tried to get a melting snowman, smiling deliriously, on a hot beach, and it does this to me
low sun angle of winter, snowy background
๐ซ
any controlnet experts ?
control_sd15_canny i have this model
now after updating controlnet i see these other models
way smaller
the models i have downloaded are 5 gb
what version do i use ?
the 1.45 one from the github ?
or the old one from controlnet 1.0 which is 5gb
also i try to update but it wont work
can i just use git pull on the extension folder and update it like that ?
ok so these are outdated then
they are from before 1.1
just making sure i can delete them because they are like 50gb total
https://huggingface.co/lllyasviel/ControlNet/tree/main/models
they are from here
fug ill just try it and see if there is a difference
I like it very
they are making angels in sand ๐
i love this radio show
do prompt editing
try just doing a pile of snow melting on a beach
after it has the basic thing add the prompt for a snowman
like in automatic you can replace parts of a prompt after some iterations, or add aprts
is this o.k.?
i don't use sd-webui
i guess you use DF right
i use my own discord bot code
I have just switched back to auto because anapnoe doesn't work with my symlinks
not sure why, and tired of each update forcing me to lose my style as it overwrites styles.css
chattr +a styles.css
you're welcome
deLoRAean
wtf is chattr? I am on Windows not Linux
there's your problem
I have to stash or delete it. No, same fucking shit on Ubuntu.
I dual boot and it is him making changes to the .css
one for pre-pull to git stash and another post-pull to stash apply
What is a real screw me is how it can't see my models I have symlinked buto auto does.
since a fork I don't get it.
or, the preferable approach is to fork it on github and use git config -e to edit your git config, and point it to your repo instead of his.
then, remote add upstream https://github.com/whoever/whatever
git pull upstream master to merge his changes in
if you git pull --rebase it will first rewind your changes, apply his, and then apply your style changes on top. a merge is a bit nastier
Yeah, I know how git hub works but if he can't work with my symlink then I don't need to use him anylonger and go back to auto1111.

you can delete them as there are pruned versions that are 700mb each
thank you
autopnoe is merging with Vlad anyway so I suspect his will go bye bye.
here are the new ones:
https://civitai.com/models/38784/controlnet-11-models
CS1o i'm sure you know but if you wrap the url in <>s it doesn't show those <>s but also doesn't show the embed
@dry crowI sure wish we had the tiles and the styles models for 2.1
I think I need to reinstall all my controlnet stuff. It's ancient.
Thanks for the link.
Looks like they're finally safetensors, too.
i think we will have them in a few weeks
I have a pickled .yaml file for CN.
I finally got the author of it and ticketed on his, as I was instructed, but no answer.
i love having it create equipment
man.. .this delorean lora looks funky at the start of its life
so the bug i fixed, was putting a blank value for the positive prompt into CTU
Compel does something with nothing to create prompt embeds but it's nooot a good result
now that it passes that actual prompt value through, directly from the given image, it's a lot better
it would be super nice to have 2.1 CN models because i wouldn't need to generate 1.5 and 2.1 embeds
and likely would save memory by being able to reuse components from the text2img pipeline, in the img2img piece, maybe..
the 12GB Colab instance really struggles with my current 2.1+1.5 combination workflow 
guys what does the lycoris extension do?
i wish he would converge
๐ง 200k? what is 200k?
Why channel icon changed?
in discord or what?
yes
you can search for server in compass icon and see number of members
200 742 ๐
shark
I will never be ballin'
not bad
not perfect either, but better then I'd expect for image containing both feet and hands
man... i think doing that generation screwed my browser up
my browser completely shut down after starting that gen
i dont think im going to need to restart the console
thats seems to be fine
ill do a test gen just to check
would love have such but getting somehow weird
yep everything is working now
i think everything crashed because it was my first use of a lycoris model
those seems weird
what seems weird?
how to say picture above o.k. this for AI i think bit messy
rollsroyce! Dont want to drive it ๐
You don't drive that you are driven in that.
i know just check front wind shield and lenght ๐
If driver see something closer than 20 meters, i dont know

If in trouble you mean free loaders can't free load from MJ so come here instead? I guess.
Only lasting for 1 month.
In general if this gets as good as MJ peope might migrate since its free in the first place
People
Maybe since both MJ, and SDXL, are just for button pushers though SDXL requires a beefy card to run locally, and the bots are here for 1 month.
True
Im hoping for Adobe to release inpainting very soon
For Firefly
fuck firefly that stuff is shit
Wouldnt say shit, especially more artistic stuff rather than photorealistic
But anyway its pretty much just a beta playground
Naw, @smoky oak Was on here with comparisons it is shit
The actual beast will be unleashed later
We shall see.
No, in beta AI from Adobe in PS
Ooh you mean the delete one?
nope, way more
The one that deletes what you mark and replaces it with whats in the background
Look, I am saying that is just the tip of the iceberg as in beta the other ones are being worked on.
I think there is a miscommunication ๐
people that say firefly is shit don't know firefly xdd
Show us what it can do good then 
For once I think the open source models will win in the end, top people at google and other places have said as much. They can't keep up with the changes being by the open source communities. Unless they manage to squash them by lobbying to have laws passed to force everyone to pay I think 1 or 2 years it'll all be open source.
Yeah, put up or shut up
My first somewhat decent generated image, yay
Now down side to that is the already ridiculous price of GPUs will sky rocket
Isnโt this good?
How do you get these good images
seems alright for firefly
- as few keywords as possible, while informative;
- keywords placed lile a "ladder";
- good negative prompt.
Also: Deliberate for the win!!
RealisiticVision2 imo is a beast of a model that and edgeofRealism
My card can't do it that big
Do you find Deliberate2 to be worse than 1, I experimented with it and wasn't impressed
Tiled VAE my friend, takes a while but can go pretty huge
Idk, Deliberate 2 is kind of the only model I've used.
Several of my Discord buddies and one youtuber think that Deliberate 2 is best
Can't use it as I am 2.1
Model is coming
Ah ok never found a good 2.1 model I liked so just haven't bothered with it much
The new model, XL or whatever its called is pretty amazing but not feasible locally yet (not even available I think?) but hopefully getting a new rig later this year
I have no want for SDXL as a trainer unless SAI suddenly surprises me and gives us the tools designed specifically to train it. Otherwise no need of sdXL
oh yeah, firefly as absolute ass at the moment lol
Hate to lower the tone but the problem all the commercial models have is they censor everything. So they hamstring themselves from the get go. Some of the best models are trained on nsfw content even if you don't want to create nsfw images.
Agreed
That looks like some movie decorations, amazing!
deliberate for those, does some great photoreal stuff
This looks a bit like Ciri btw
also firefly's true power won't be text2image more other stuff such as skecth2image 3D2Image creating a 3D scene and transferring to image and all these intergrated in the tools adobe already owns
leads to artifacting
@dense tapir how do you make these? What model, what prompt?
Haven't noticed it myself going up to around 1.5k to 2k but perhaps above that it does ?
I don't have a fast card so can't do them 100%
What GPU do you have?
1060
Something my 6gb just will not do.
Well, I have 8 gb
even at 1920x1080
it's especially noticeable if you put 2.1 on a tiled VAE
Palit Super Jetstream
Yeah as I say don't use any 2 models
Did this earlier needs another upscale pass and more inpainting but that uses tiled vae
can you see it?
your image is more synthetic and so, it will be less noticeable
try to make a natural scene and stuff feels "off". another thing i noticed is that tiling the VAE leads to more deranged faces when they occur on the transition lines.
that's only 784x1024
Wtf, Collie bear
also, what VAE are you using? MSE 840k?
I could easily take it bigger, that isn't artifacting your seeing imp its exposure
well if you're happy with your images, that's awesome, man. but i wouldn't recommend VAE tiling, in general, personally. it's good for people to know they're possibly enabling something that's going to reduce the quality of their outputs.
Well as I say I wouldn't say the above is a 'low quality output' (even allowing for subjective taste) but to each their own
it's generally better results to generate a smaller native image and then use eg. BICUBIC upsampling and then ControlNet or some other 'fixer'
you upsample it to the blurry and disgusting resolution you want ControlNet to fix it at
๐
There are numerous ways to upsample, loopback is very good as well
yeah multidiffusion as well
Yeah that works extremely well but tbh I found that artifacts. I did these 3 tests on different upsample methods, same base image
bottom right is multidiffusion
I kind of like it least despite the added detail
Damn, all 3 images are so different
@smoky oak https://www.youtube.com/watch?v=CBvdo3JnLX0 Google teams up with Adobe's Firefly.
BIG AI News. Midjourney starts to research 3D from Text. Google partners with Adobe Firefly. The Google PaLM 2 and Gemini Models are very powerful. Google indetifies AI images and Watermarks them as Synthetic Content. Shap-E can create 3D Models from Text. And many other news
Support my Channel:
https://www.youtube.com/channel/UCCKx8mAHiFus-XYQ...
oh yeah, two corrupt AI companies working together
just throw in open AI while were at it and we can kiss everything goodbye to their "good and ethical" dataset collecting
BINGO, we think alike on this one.
It's funny that Midjourney is usually thought about as supreme art AI
I never liked it as I like my art creator to not inject its own style. If you have a style, that MJ style, then you are doing something wrong imo.
Anyone figured out how to run SHAP-E locally? 
IMO Mj has gone downhill, I mean quality wise cool it can create photos but its so opinionated now, it just throws to much of its own bias in
I liked V2 myself when you could create blurry nonsense and upscale them into all sorts of outputs
Yeah.
Also MJ is really limited. I think you can't even really train it on some character like you can with SD?
SD has so much freedom it's crazy
What makes them more corrupt than Stability AI,
The fact that they are corporations?
what lol
They are both massively censored, with huge goals and political gains from controlling the use of AI in the media, while also using crowd source testing to make better tools in house that only they will benefit from.
Also, Adobe is one of the most problematic companies out there. They shake down artists and creatives for severe amounts of money yearly just to fund their never ending program bloat. They are just trying to monopolize yet another thing in the industry, of which they will then hike their prices up once again (like they always do) and force people who want to stay relevant in the professional subsects of design to pay it
Yep pretty much this, its why I only use the open source stuff now, won't touch a paid service
Also, RIP, Adobe Flash Player๐ญ
most problematic to whom? And how is someone better than this who uses Stable Diffusion which is also very controversial?
or is it just because its open source that it is justified?
There is the OpenSource tool called Ruffle, that supports Flash games ๐
I wish png info worked with Controlnet
and then Open AI themselves are breaking the law, and I am just waiting for them to get nuked out of existence
yet you use Stable Diffusion
Sure, it may be controversial to some, but being controversial and literally monopolizing a several billion dollar industry where professionals have no choice but to spend 100's yearly just to stay relevant are not the same
SD isn't breaking the law lol
Stability AI got a lawsuit for a reason
well OpenAI too
Yeah, and they won it, for a reason lol
Shutterstock lost, cause the use of their images is not illegal
who cares, people use those for a reason and stay for a reason not just "because they are forced"
same with OpenAI
nobody is going to nuke them out of existence
They literally are if they are employed in a professional scene where Adobe is the pipeline, which is 98% of design professions
OpenAI is breaking the law in other separate ways
Like, you know, the "open" part
which they dont pay for anyway
i mean this
Open AI is supposed to be Ai for the people. That is their business model and mission. They started as Ai for the people, where they would give their research to the people. It is an open source company at its inception, and thats why even Elon musk doesn;t support them anymore
They are an open source company that is not open sourcing their work anymore because of "competition"
and those who arent working professionally for such companies are still subscribing to Adobe CC and they arent forced to
them buying some smaller companies, some bigger ones or some startups is not corruption
and for us who profit from the products its not the biggest deal anyway
yeah it is, maybe no to you tho
The deal is that you have to or it doesn't work.
obviously people arent not upset enough with it to leave
Screw software as a service and all companies who embrace that crap.
with exceptions
Get you hands out of my pockets.
There are no other options. Thats what a monopoly is. People are forced to stay, or go without
I myself was in that position, and many of my professional friends still are
they arent forced, they choose to
Because no alternatives???
its not like only professionals that work in a studio are using Adobe CC
I think we have a troll, or a shill working the room.
there are, but they arent good enough or convincing enough or the pipeline is established
and then they sell themselves out to them eventually lmao
Luckily open source development will kill the big players, the real money is likely going to be in server farms
wishful thinking that is screamed out for decades
Adobe, Google, now OpenAI, Autodesk and more are said by Open Source cultists to be obliterated
People at those companies have said as much, open source development is just much faster and more dynamic
and this for a longer time
and then those big ones are even deploying open-source stuff themselves
they dont have to fear open source
Wait, what is wrong with Google?
Its the reason why stable diffusion craps on Dalle 2, and Firefly.
They do not have the time, resources, or man power to keep up with the sheer amount of crap we are able to do and learn
people have claimed that Google is about to die soon, especially after OpenAI came with ChatGPT
As I say those models are hamstrung from the start they can't compete with models trained in the wild and the mass of extensions and bleed over SD gets from China and its efforts
this conversation just went over a cliff
google is never going anywhere
SD has its own niche of users, Firefly will have theirs
dont worry about that one
anyways, apprently this guy thinks that google and Adobe are not controversial companies, and that it is totally not problematic that they are working together on Firefly in order to try and monopolize AI
Probably thinks that John Deere isn't problematic, and that autodesk is a saint as well
Either this guy is a troll, or I am genuinely concerned for his safety
anyways, typically when a whole group of people well educated on a topic are in opposition to your point of view, that tends to mean you are the one in the wrong lol
"cultists"
bro, try harder lmao
so now it's insults, that's not how conversation works
we use an AI image gen software, praise be thy anime tiddies I guess
and who would have thought, on a SD server that i would have multiple people against me lol
Its ok to be wrong, I just suggest not doing it confidently, as thats when people have "problems"
nobody said that SD isn't gonna stay around, nobody said that firefly is gonna be bad, all we said was an objective truth, that goodle and Adobe have a history of monopolizing and preying on people who cannot leave their ecosystems
goodle oops lol
this is pretty much a matter of interests, not ethics and morals between us here
i wasnt kidding when i said im playing the devils advocate
So the constant syphoning from passionate creatives who have no other choice isn't a matter of "ethics?"
but i can also say what i really think
both of them have kinda a monopoly status if that is what you wanted to know from me what i think
do i sh*t on them for that? No
im using both of them
it would be a liability for me to do that
I wish there were than two phone options.. to go off on a tangent
Firefly page on Adobe website makes it look pretty impressive, ngl.
Doing AI generation + photoshop in potentially one app also seems great
exactly, you are dependant on them
We all use google and adobe crap all the time, I am sure of it. The difference here is we are not complacent about it. We know we are being screwed over
im not dependent on neither of them
same with AUtodesk btw
Autodesk is just 3D adobe
kinda
but still
im actually not dependent on them
i use their products for reasons
Maxon buying up zbrush made me feel ill
not because i want to be cool or because im dependent
they probably would do the same with Substance if Adobe didnt do it before
nobody said this was about being "cool"
of course they would, as all of these companies want everything
and then they buy these startups, and change them from the glory they once were
I should know, a project I worked on was bought by Unity, and now its not even recognizable
do you know what my actual issue pretty much is?
Really if you remotely care about any of this there is no reason to use paid services, GIMP, Blender and SD local (and many more) are excellent. Anyone who pays for PS unless your a professional artist is crazy imo
At this point, I don't think we really care. You have made it clear your goal here is to oppose our ideas
look if someone who is really heavily into FOSS and doesnt want the proprietary stuff and actually is consequent about it and also is consequent with the reasons he doesnt use those and sticks to FOSS im really fine with it, everyone according to his needs and wishes
none of us here (at least myself) have a problem with closed soruce crap. It can be massively beneficial. We have problems with how these companies force it on people with no alternatives and then take them for all they are worth
the problem starts with those who say they are this above but actually are subscribers to other products and are also smokers and alcohol drinkers and subscribe to Netflix and what not. If they would at least say that its just because of their interests and priorities
wdym force people with no alternatives?
To say that there is no mallicious intent behind google joining Adobe on firefly is just flat out wrong. Firefly is gonna get so good that creatives can't not use it, to which they will begin to raise its price. Its adobes whole business model. Make something decent enough to get adopted, then hike the price
photoshop on it's own isn't a massive expense, but it's pay as you go. I get why people hate that
there are alternatives to Google, Adobe, Autodesk, OpenAI
the question is how viable they are
Is it an alternative if its not as capable?
Wait, you complain that company makes something great and then sells it?
I just hate adobe for taking all SD and community work, including extensions , integrating them into theiw own ecosystem...
No credit given to SD or people who made extensions.
All they are doing- stealing other people's free work and making money out of it.
and its okay, but why do also people hate on that that subscribe to other products and buy stupid stuff on a monthly basis like alcohol and cigarettes?
If you wanna work in an adobe based studio, there is no alternative to adobe
they dont
slippery slope I guess, price can always go up
no need to get political about addictions. Thats a messy topic, and it is monumentally insensitive to lump genuine addiction in with cooperate shenanigans
addiction?
come on lol
is that the excuse?
Do you think that smoking and drinking are not addictions?
alright, I am done
isn't this a good time to stop all this.. getting silly now ๐
this is a waste of time
yeah, agreed
you are excusing their double standards with "addiction"
agree to disagree or whatever the saying is
true
look at the fluffy cat slug, he hates arguments
they will integrate them into CC apps
โค๏ธ
You can say that when they'll make their tools based on open source community work free.
Until then don't even pretend they aren't selling something they didn't make, without even giving credit to community or SD.
why would they give credit to SD
their models dataset isnt based on SD
it would be a stupid idea anyway
@smoky oak somewhere in this discussion you started to complain that companies find their niche, make great product, then sell it for relatively high price and there're no alternatives.
But that's natural, isn't it? Usually whatever company gets to do something first, sets their own rules, that's how market works.
Tools they are using - community tools based on SD.
the amount of assumptions this guy makes is insane
no
did you even look up on their current and future features?
Unfortunately it is, however there are different ways about going about it. There are companies that continue to innovate, and really thrive as a community, where they keep prices reasonable and do their own thing
But then there are massive conglomerates that see that a company is trying to just be peaceful, and do their own thing ethically, and they offer them so much money that they cannot deny, buy them, and then hike their prices
None of them will get close to SD currently its months behind and hamstrung by a censored limp dataset. And MJ will always beat them to the more casual crowd. The only use PS will have for it is stuff like smart content fill and such for retouching.
its in early beta
this state doesnt matter much right now
Can be in beta for the next 10 years dataset suddenly won't become good
for the more casual crowd part, it may be
although, not sure about MJ
Stable Diffusion is beating MJ if you ask me
the fact that a mullti billion dollar conglomerate decided to release such an unpolished product in beta shows that they do not have much confidence in their current tech, and just needed to get their name in other peoples mouth's to not go irrelevant TBH
and if they keep this SD XL bot its a even bigger threat for MJ
I agree SD is beating MJ but for casual users who don't want to muck about with python and models, MJ Is still the easy best choice
"unpolished product" is that what you say about SD too? :X
true
and they use it on mobile too
but then again SD might do the same now
with the bot
no? Considering its the best AI image generator available...
and you think this will remain so
it has been, up until now, sooo
also did you say the same months ago?
It also has that early 90s Playstation factor where it just became the defacto cool one to use, for reasons so most people not super into AI art auto think of midjourney
or when SD was in "beta"
sure, things can change, but as it stands, firefly is extremely behind the curve
uhhh... Yeah, actually
i notice more influencers doing marketing for Leonardo AI
i mean more and more
not more in overall
so Adobe Firefly is unpolished....because just so
because you dont like them
Like the only people who will use Firefly are the captive crowd who already use Adobe products, nobody would choose it for image generation over local diffusion or other free services
Because it is extremely bad at certain things
artists
Yes Leonardo is excellent , I use it to generate ideas then run them through SD to make them better
if firefly was like really good, I could definitely say its cool, but its not right now
one cool thing i like there is that you train right away on your images if you want
Even then not for generation its garbage, a proper artist can train a model on their own work in Auto1111 and get 100xs beter output than anything firefly ever could offer
not yet, its not even available for commercialising yet
you have that watermark when you generate as of now
its literally just a playground for a beta product
that is 2 months old
they might eventually completelly ditch Firefly because they will integrate all the features into CC apps directly
dunno if they wanna keep Firefly as it is
I will say that firefly is not bad in all aspects, but its image generation in general is unusably bad
its behind in photorealism mostly
also ofc its mentionable that you cant use famous artists and their styles and copyrighted characters for example
that one for a reason tho
that's worse in every way
Firefly will get this function too
I have no real experience with the other tools besides the image generation in firefly (which is the most relevant cause of where we are) and the image generation aspect is unusably bad, genuinely
I didn't ask you nor do I care what you think as it is all subjective.
That is why it will always be garbage, no artist reference, no likenesses, no NSFW content (even a hint of bare navel) its just a hamstrung model, same as MJ but at least they allow some wiggle room and it produces high quality images
this breaks down to groups of consumers alltogether as some of you already mentioned more or less
agree to disagree
but we will see in the coming months and even years
at the end of the day people have their reasons to stick or go for either of the products available for the masses
who knows if some new contender will join too
(and who it could be )
That's the point though while they are playing catch up the competition isn't standing there with their thumbs up their butts. I have no doubt it will improve but it will never beat the current front runners. Leonardo has a chance but there development plans are odd, they seem intent on making it focused on 3d texturing and characters
i would understand this if we talked about some small group of devs, but we talk about giants like Adobe and Nvidia
Leonardo, along with all cloud based nonsense allows no uniqueness. I want to add my own models, loras/locons/lycoris/embeddings, etc... my special sauce.
cant you?
A group of lead devs at google said they can't keep up with open source development. Partly cause its not limited by an overarching structure trying to monetise it, devs can get it into peoples hands fast then ideate on it live, in development terms thats priceless. What make take a team of 30 or 40 hobby coders a few weeks on SD will take the big players twice as long, running it past all the relevant departments and testing phases
No, I can't. I can't train my own model, and hypernetwork, and lycoris, and embeddings, and select how much I wish each to beused (weighting) with them.
Leonardo (the free service) is decent for running ideas, thats it though imo
just buy a W7900
you mean Luke Sernau
its not that easy to say tho what you said above
it is possible tho
there are factors to consider there
btw he also said that the one clear winner in all of this is actually....Meta
with the leak of Llama
No he said what I related above, that open source collaborative development is just fast, more efficient and better at making advances. Their multi million dollar 13b paramanter models were replicated within days by random people with laptops
he actually said what i meant above
with Meta
and someone has to read what he said carefully
yes, he said open source is developing very fast
he also says Google should work with open-source community
No your taking the one sentence that supports your position and ignoring the entire rest of the memo. He says it helps meta cause the iterations of the LLM are being done on their model, so they can reabsorb that work. But his overall point is neither them nor Meta can keep up with open source, a group of students in Korea (I think it was) worked out a way to train a LLM on 50k compared to the multi millions it cost Meta and Google
id still want to see what he has to say about that leaked document
And they did it in 2 weeks rather than 2 years
if he didnt respond already
dont know about the case with koreans
but the idea to work with open source community isnt bad per se btw
i mean some of the big bois for example fund some open source projects and organisations and softwares already and deploy their own too
yet it still applies: Certain factors are to be considered before we talk about who is about to win
what is this argument, even
I don't think there will be a 'winner' just different things for different people. Adobe and its ilk will be integrated into existing platforms and serve a role, MJ and Leo will cater to people on mobiles and without good GPUs, SD will remain a slightly clunky hard to penetrate thing that is more powerful than all the others when used correctly.
Jesus, its still going?
oh nevermind, it stopped
I could't bother 
Vlad's page says I need to put --upgrade to update, but what folder do I open the CMD in to do that?
FuuuuuuuuuuuuuuuuuuuuuuuuuuUUUUUUUUUUUUUUCK
no idea, I don't touch vlad anymore cause of all of the issues I have had with it
there has to be a word stronger than cumming
Going
Ded
My girl beats yours in a fight ๐
pass tence
Zovya do you want to use my model
I like this vibe :>
I want you to blind test it vs. yours
also if u guys arent using the added details lora ur not diffusing
you can also inverse the minimalist Lora to actually add detail too
are negative loras worth it?
I feel like, how much more detail could you possibly add
these are already godlike
No idea I never actually tried it lol
Should I reveal my KFC special recipe o:
I'm convinced nothing can beat this model
fuck it I'ma give away the percentages cuz I'm gonna do it in civitai anyways
this is only like what, the 5th time you have said that?
๐
Maybe he keeps breaking his own record
I know it's so depressing
no wait
Zegova's model is probably at the same level
but it cant be dreamboothed
this one can
just release it already so I can determine if its actually that good, or if its just like senblend 1.7 or whatever it was
increases are marginal, we are indeed reaching the limits
but still an increase o:
It takes about 6 hours to upload that joint
but I'ma work on it as soon as chores are done
also, after using this new anthro model, I realized just how far realism models are behind in terms of control compared to anthro models. Its kinda insane
whats that
Like I am not messing with realism cause its so much less impressive than this ne mode
I want more tools for my arsenal o:
its a furry model with a level of quality, consistency, control, and adaptability that makes realism models look like a joke in terms of capability
it can do humans as well of course
but this models capabilities are several levels above what even Zovya can offer in terms of instruction and minute control
I cant thumb down posts for some reason
I have you blocked, thats why
r0fl
just waiting for you to get banned again
it really is sad that you try so hard to be in a place that keeps wanting you to leave
this is goofy, and I like it
Love that new toolkit extension for A1111. Was able to cut the garbage out of my bloated 5GB model and get it down to 2GB
Ignore the hideous ugly hand ๐คฃ
I like this cute little style
prompt is: conceptual abstract illustration of happy happy joy joy sunshine lollipops rainbows
Its an image done on Leo originally then run through SD using the JIM TERUMODE model from civit AI
I used to spend ages touching up inpainting etc. Now I'd rather generate 100-200 images and pick my fav from them
please, god, do
Frankensteining is my thing tbh, I think you get more interesting results running through multiple models.
prompt: conceptual abstract illustration of lighten the mood
oops, mucked up a prompt before hitting start, but getting some pretty cool gens anyway
oh i passed by that while perusing civit earlier

05/16/2023 00:09:33 - INFO - __main__ - Num examples = 224
05/16/2023 00:09:33 - INFO - __main__ - Num batches each epoch = 224
05/16/2023 00:09:33 - INFO - __main__ - Num Epochs = 8
05/16/2023 00:09:33 - INFO - __main__ - Instantaneous batch size per device = 1
05/16/2023 00:09:33 - INFO - __main__ - Total train batch size (w. parallel, distributed & accumulation) = 2
05/16/2023 00:09:33 - INFO - __main__ - Gradient Accumulation steps = 2
05/16/2023 00:09:33 - INFO - __main__ - Total optimization steps = 800```
the default settings are likely going to fucking nuke whatever i'm working on anyway


