not a very good model, it takes some tries to make good things
#🏞｜general-with-images
charred crystal
could get some great tabaxi models with that, but I don't see any tail
smoky oak
could get some great tabaxi models with that, but I don't see any tail
I am telling you, anything furry v4 can do great tabaxi and such, as its literally trained off of furries. But you have to really heavily weight the prompt towards SFW 😅
smoky oak
IS THAT SHREK!?
glossy herald
yeah, in cats the musical :p
smoky oak
thats so cursed lmao
charred crystal
himbo thanos with cats ears
shadow sage
Hi veryone. Glad to be here
I would like to ask something because I'm getting mad about and I don't understand what is going on
The image is processing, with detailed structure of buildings, but then... Every detail it gets destroyed... Why? This kind of thing happens to me so often and I don't know what I'm doing wrong.
smoky oak
The image is processing, with detailed structure of buildings, but then... Every...
your images are too blurry when upscaling?
is that whats happening?
@shadow sagenot sure if my reply pinged or not, sorry if it did
so like, good detail on the normal gen, blurry details on the upscale?
if thats the case, I can help
it blurs at 50%
shadow sage
if thats the case, I can help
I get said in tech-support that the problem is the latent upscaling
And denoising level
shadow sage
And the results are much better now
smoky oak
some can go down to 0.05, some need like .8
realistic vision v1.3 works between .3-.45
shadow sage
Yes, I just learnt that
smoky oak
anything furry likes .4+
smoky oak
Yes, I just learnt that
awesome! Happy generating
shadow sage
Thank you so much 🙂 very appreciate your help
shadow sage
wooof
smoky oak
that was my second day, now I am 15k+ images deep 😅
Just made this wallpaper for my phone :>
shadow sage
Just made this wallpaper for my phone :>
Beautiful!
smoky oak
Also, let me share some quick tips with you@shadow sage
you can press control + enter to generate/stop generating
and a new one I just learned
shadow sage
you can press control + enter to generate/stop generating
Best tip ever, I was looking for that
smoky oak
if you highlight text and use control + up or down key, it will increase/decrease the weight by .1
shadow sage
if you highlight text and use control + up or down key, it will increase/decreas...
That I knew it, very handy
smoky oak
so it will take it from
pretty
to (pretty:1.1)
shadow sage
Just made this wallpaper for my phone :>
I suppose it's 9:21 AR
smoky oak
happy generating
shadow sage
happy generating
TY!! Same for you!
smoky oak
I suppose it's 9:21 AR
20:9 haha
just a litttle off
my monitor is 21:9
2560x1080 vs 2316x1080
dense tapir
smoky oak
(its actually 3088x1440, but 1440p is a waste on a small screen lol)
shadow sage
Nice
charred crystal
it's terribly unrealistic though
smoky oak
it's terribly unrealistic though
this AI isn't exactly trained on photorealistic depictions of lavender sprigs haha
the cherry blossom one came out quite realistic, tho I am going for a very dreamy out of focus look
charred crystal
would like the smell of lavender as well
I think perfume makers are safe from the AI revolution for now
smoky oak
hexed bronze
smoky oak
I love non standard aspect ratios
hexed bronze
been experimenting a lot with the manhwa style stuff
entirely vertical web novels, and they have these long panels
smoky oak
oh man, the AI decided to get
✨ F a N c Y* ✨
random hexagonal bokeh
I... love it haha
wait- does that mean I can specify the shape?
raw quiver
charred crystal
I wish you could specify shape and texture
hexed bronze
smoky oak
damn, these look sick man
charred crystal
are you just usng prompts @dense tapir or are you using inpaint?
dense tapir
just prompts as I suck at inpainting
charred crystal
love the superman you posted yesterday
craggy rune
nice
mossy night
Took me a while to get a sort of eldritch horror that actually made sense instead of being a nonsensical blob 😅
mossy night
also, this is a great library for hands position
That's amazing. Thanks!
dense tapir
glossy herald
That's amazing. Thanks!
yep, I just found it on reddit and I share the love, it's just great
dense tapir
pseudo aspen
Take a look at some beautiful sceneries I made yesterday and today
pseudo aspen
also, this is a great library for hands position
A solution to bad hands. Nice!
tropic shell
Take a look at some beautiful sceneries I made yesterday and today
Love all the flowers and reflections~! Breathtaking!
glossy herald
Hey aZeo !
nice to see you around too, always happy to see old faces, I took a break at some point too
this chan has calmed down a little, but it's getting more alive by the day recently !
ancient trout
Smart baby,cartoon, profile pic
glossy herald
Smart baby,cartoon, profile pic
Hey, I answered you in another channel earlier
we don't have a bot to make your prompt into pictures
fresh hound
cheers buddy - nice to see you too!
crystal deltaBOT
fresh hound
I'm seeing some incredible results here
glossy herald
you tried controlnet yet ? ^^
fresh hound
what is that 😮
glossy herald
this is SD made
and controlnet showing off
it's a set of complementary models that help shape what you want
fresh hound
oh cool
jaunty hearth
Portada de comic
fresh hound
dang thats really cool!
fresh hound
Sweet! will deffo get that added
glossy herald
it's another level, for texts in particular
fresh hound
this looks like it would (hopefully) be useful for video too
fresh hound
aaaaaaaah yeah i saw that
glossy herald
I'm having so much fun for it
fiery carbon
@smoky oak Very nice work, i like your UltraWide wallpapers ❤️🔥
outer spear
Whats PCIE 40G? I couldnt find anything on y outube
dusty needle
the image shows the logo of an ice fennec with geometric textures
cyan snow
dense solstice
it's big brain time
cyan snow
tropic zinc
I am going to run a campaign for Nike.com, can you provide some content about history of nike and how they have one of the best community in the world and provide images that are some of their best shoes they have ever sold
wispy nest
El sol
opal tangle
so my upscaler app is going fast af now ( for size comparison, it's using ai upscaling not just resized )
simple GUI atm but its runnin ;'D
cyan snow
so my upscaler app is going fast af now ( for size comparison, it's using ai ups...
that really isn't how you should be upscaling your generations
opal tangle
that really isn't how you should be upscaling your generations
this is a tool I am making to host online for free
a general AI image upscaler
aka: it's not, this is a web app I've been developing in flask/html/css/js/python
cyan snow
there already is a button for it
opal tangle
in automatic1111 webui there is yeah, but that's not what I'm doing
I'm making a website ( and online app )
cyan snow
gritty bluff
lean oriole
how do i remove the white bands with negative prompt ?
glossy herald
Fight club, half is Edward norton, other half is Brad pitt.
source for controlnet
using couple latent
controlnet on clippy
smoky oak
<@296720689468145664> Very nice work, i like your UltraWide wallpapers ❤️🔥
Thanks!
smoky oak
controlnet on clippy
genuinely trippy haha
glossy herald
if tweeter was a real bird
this one fits the logo too, technically, but I didn't see it coming ^^
smoky oak
how do i remove the white bands with negative prompt ?
enter into negative prompt:
(boarders:1.4), (white boarders:1.4), (Crop boarders:1.4), (Blank space:1.4)
stuff like that
smoky oak
if tweeter was a real bird
I'm pretty sure making the twitter bird more realistic is against their TOS 😅
glossy herald
x)
let them ban me !
but I can't unsee the last one
I'll always think of that posture seing the logo now
smoky oak
Take a look at some beautiful sceneries I made yesterday and today
would you mind sharing your model and basic prompting for these results? I have been trying to make photo realistic wallpapers for my PC, and they haven't been capturing this level of fidelity
ok, I actually used the PNG inspect on it, and now I am more confused than ever before lol
you use the same model as me
pseudo aspen
would you mind sharing your model and basic prompting for these results? I have ...
I am using Realistic Vision 1.3. Basic prompts I used were"mountains, lake, clouds, sunny weather, (photorealism:1.4), (ultrarealism), (extremely detailed)". Also use Highres fix
smoky oak
very interesting how different our results are
I made this one with realistic vision 1.3
1080x2560
pseudo aspen
Hmm.. pretty nice result as well
smoky oak
All of my flower images have been realistic vision as well
no shit, your prompting does get wildly different results, damn
lean oriole
enter into negative prompt:
(boarders:1.4), (white boarders:1.4), (Crop boarder...
thanks
glossy herald
Whiterun
pastel bluff
glossy herald
edges detection of some kind ? it feels not just img2img, but going after the controlnet vibe
pastel bluff
edges detection of some kind ? it feels not just img2img, but going after the co...
Raw image-to-image with XL
glossy herald
that XL seems enticing. can't wait to ear more 🙂
lilac pilot
Is this something new?
Trying to fix hands later
Sometimes masking and trying higher resolution does not make the face better idk why. I see others do it
@pseudo aspen Felt like doing some as well. 
glossy herald
Elden cat
text went up in fire lol
btw controlnet source (thanks paint3D)
delicate wharf
what is this at the bottom of txt2img tab:
it changes the output a lot, and I don't know what it is.
I can select one of them or both.
does someone know?
glossy herald
cursive lance
smoky oak
lol
west kiln
cursive lance
is this generated?
It could be said that yes, except the letters above
smoky oak
peeng
smoky oak
It could be said that yes, except the letters above
haha, awesome lol
crystal bloom
testing studio ghibli Lora. Used ControlNet canny for the fist image.
mossy dagger
How can fix
gaunt heron
Hogwart's Lega-Z
dry crow
help me<:habby:1017789370532646913>
You can try to delete both xformers Folder an run then try again
edgy shard
okay
this looks like an alright place. i made this board, i am really leaning towards one of these (this is from img2img from MJ), and I am curious to see if you guys can pick what my type is, based on this simple character category
which one do i like best out of that group :D
also curious to hear which one you guys like best
dry crow
maybe yours one is mid row right,
my fav is bottom left 😄
edgy shard
those are both good ones, bottom right is indeed a cutes
my favorite out of this selection is actually the center left though
edgy shard
i heard there's a checkpoint that allows for character sheets, but i don't know where to find it
dry crow
maybe CharHelperV4 but idk
dry crow
thats an embedding
edgy shard
okay so do i download this?
dry crow
if you want such pose then yes
edgy shard
where do i put this?
dry crow
in embeddings folder
edgy shard
okay
dry crow
then you activate it with the trigger word
edgy shard
how can i turn on hires fix with im2img
stoic pivot
photo of realistic([cat]|[hamster]) sitting in a watermelon boat
edgy shard
ok i am going to try this turn script
dry crow
very nice 😄
edgy shard
man
this has me considering buying a second sabrent so i can load checkpoints
not sure if i need terabytes of checkpoints tho ._.
dry crow
haha i dont think so, but of my 1tb ssd 350gb are models
dry crow
why reload after restart
edgy shard
wanted to try something new
ivory nacelle
hello does anyone know if you can use latent couple in easy diffusion?
dry crow
yea just restart xD
stoic pivot
Donkey + hamster = Ugly rabbit?
edgy shard
haha
okay so @dry crow how do i use this to activate it? do i just type it and it does it or should it show me something to know it's active
dry crow
you need the trigger word in your prompt and then you describe your char and add mutliple views or something too
edgy shard
i don't understand if i'm doing it wrong or not
can i img2img because i don't want a prompt
dry crow
for Example:
1girl, 21charturnerv2, multiple views, standing in kitchen, black hair, brown eyes,
edgy shard
dry crow
can i img2img because i don't want a prompt
your resolution must be wide view
in output
edgy shard
is that like 768x512?
edgy shard
it's not finding her
dry crow
didnt test it in img2img
edgy shard
can i run 2.1?
silk gull
does anyone hear know how to colour correct an image one makes in 3D, and feed it into SD so it spits out the correct lighting and colour results (contrast, exposure etc) and add better details? What is the proper way to set this in action? the above image is a combination of a photo and a 3d environment. I need to improve it. How would I tell SD to match better, and give better more believable results?
dry crow
can i run 2.1?
sure 2.1 are just models, but they need a .yaml file too
edgy shard
i really would like to figure out how i can img2img a character sheet. stable diffusion gives way more control than MJ but MJ has it built-in
stoic pivot
You never asked for it, so here it is, a baby cat monkey.
tender glen
Where can I activate Dropdown VAE selector next to model selector on main tab?
Do I need a plugin?
dry crow
Where can I activate Dropdown VAE selector next to model selector on main tab?
D...
its answered right now in #📝｜prompting-help
green plover
I've been pondering. I may have come up with an idea that might get the traditional artists off our backs.
This is the first image in the Laion improved 6p dataset. Checked the terms and it's not really available for reuse, so I downloaded it, ran it through blip2 on a beam search returning 10 results and used those results plus the original caption to generate a description
Applying a photo style and a negative prompt and running it through an aesthetics checker I came up with this.
Higher aesthetic value (the requirement of my generator script was the same or higher) and they can't say we stole it if we created it
green plover
we have all the tools we need to do it
plain patio
I understood that artists / stock images site don't really have a leg to stand on in their lawsuits?
green plover
Some of these don't come from stock image sites.
wispy nest
If your using it for personal they don't care
oblique nimbus
coarse jetty
oblique nimbus
dense tapir
spice sorrel
Wtf is the background? Did the AI actually noticed the Paint Tool SAI interface?
lapis geyser
hollow knot
crea fondo de galaxia con muchos colores y estrellas
fiery carbon
He looks a lot like that famous belly dancer, you know?
vague oasis
crea fondo de galaxia con muchos colores y estrellas
el bot lo sacaron, si tienes un buen pc puedes usar stable diffusion localmente usando tu gpu
hallow sparrow
Is there a different place to ask questions? I haven't had one response so far? And i have another question, my dreambooth does not show any models'? Although all is loaded and working fine...
fiery carbon
It will show your model as soon as you have saved it.
glossy herald
Is there a different place to ask questions? I haven't had one response so far? ...
usually #🤝｜tech-support is a lot more monitored by people doing bug fixes than #🏞｜general-with-images , this is more of a chat, and image sharing board than that yeah. Not sure about that problem
(I know when I'm doing that I rarely have even time to read general)
From playdought
to life
dense tapir
pseudo aspen
Ohh nice ones 😊
lean oriole
hows this?
(((full body:1.4))), rogue warrior king, (Indian:1.2), (wearing multiple leather and stitched armor), wearing Heavy Armor and metal, with a sword in hand, long hair, medium beard, (no helmet), (insanely detailed), smirk, aggressive dynamic pose, beautiful detailed eyes:1.4, cinematic, intricate details, (masterpiece, best quality), (cinematic light), hyperrealistic, hyperdetailed, 8k, HDR, intimidating, dark fantasy, detailed armor, (war and destruction background:1.2), drawn by Greg Rutkowski, Yoji Shinkawa:0.6 <lora:epiNoiseoffset_v2:1> <lora:realisticVisionV13_v13:1>
Negative prompt: bad-hands-5, (face covering:1.2), (face mask:1.2), ((mask:1.2)), (cover face:1.2), (hood:1.2), (cowl:1.2), (flower:1.2), nude, (bad art, low detail, pencil drawing:1.4), (plain background, grainy, low
quality), (mutated hands and fingers:1.4), (watermark, thin lines:1.2), (deformed, signature:1.2), (blurry, ugly, bad anatomy, extra
limbs, undersaturated, low resolution), disfigured, deformations, out of frame, amputee, bad proportions, extra limb, missing
limbs, distortion, floating limbs, out of frame, poorly drawn face, poorly drawn hands, text, malformed, missing fingers, (((cropped)))
Steps: 50, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 4272978727, Face restoration: CodeFormer, Size: 512x512, Model hash: e04b020012, Model: rpg_V4
Just my daily thanks for not being part of midjourney anymore, but instead I am part of a community that actually welcomes change and diversity
static tusk
hows this?
they look really nice. You should inpaint the eyes though to complete the pics. Upscale using extras then inpaint.
wispy nest
if i want to try getting prompts from an image, where to ask
You can use png info if it was generated with sd
rain hedge
The link to the post i got the image from is below
https://www.reddit.com/r/NovelAi/comments/11cza4s/shes_pretty_good_for_about_my_10billionth_attempt/
dense tapir
wispy nest
The link to the post i got the image from is below
anime promts are pretty easy, unless you want something specific
sour swift
been working on generating textures for 3d assets using SD. This one was made using the prompt "a truck in the style of the grateful dead, psychedelic, 70's"
worn carbon
cowboy man just crossed a fence in a field and meets an angry bull coming towards him (animated mode)
cowboy man just crossed a fence in a field and meets an angry bull coming towards him (animated mode)
summer raptor
just lurked from chinese telegram, they got websites with a bunch of models from community
ruby flicker
How can I get started generating images?
green plover
how far along are you?
glossy frigate
Thought this turned out well
glossy frigate
yes pls, someone know how to start this ai
Follow the youtube guide called "Install Stable Diffusion Locally (In 3 minutes!!)" by Royal Skies. This ai can be run in many ways, but the most common is the Automatic1111 webui. Since this ai uses your gpu to generate the images, you can expect good results with 8gb or more of VRAM. There may be solutions for lower VRAM builds but I am not familiar with them yet
There are options to rent a space on HuggingFace that hosts your server with an off-board GPU with high VRAM, so that's a non-free option that can still be used with the Automatic1111 webui
raven beacon
thnks a lot gritings from peru
smoky oak
we have rights to sell our images from stable diffusion, right?
green plover
we have rights to sell our images from stable diffusion, right?
Yes
glossy frigate
we have rights to sell our images from stable diffusion, right?
https://huggingface.co/stabilityai/stable-diffusion-2/blob/main/LICENSE-MODEL Section III 6 mentions that the Licensor (copyright owner of stable diffusion 2) does not claim rights in your output using the stable diffusion model as long as you do not breach the other provisions in the license
That is the CreativeML Open RAIL++-M License that stable diffusion 2 uses.
smoky oak
https://huggingface.co/stabilityai/stable-diffusion-2/blob/main/LICENSE-MODEL Se...
Alright, sounds good. Just wanted to know if I could make money off of a wallpaper app like zedge for my generated wallpapers
glossy frigate
As long as you don't breach the Use Restrictions at the end of the license, you should be good
Not that the use restrictions are extremely specific, and your viewpoint on what is covered under those restrictions is subjective
smoky oak
I'm just generating pretty flower wallpapers, I am sure I will be fine 😅
Like such:
If this is against use restrictions, crucify me :p
@glossy frigateThank you by the way, I forgot to say that previously
glossy frigate
Looks great!
patent onyx
tryed somthing inspired from a reddit post...amazing what SD can do
spring vector
a beautiful girl
green plover
a beautiful girl
I've rendered quite a few of those. I can't post most of them here. See #start-here for instructions on how to get start. There's no renderbot on this server
dense tapir
Welp, seems until Colab updates pyton to 3.10 most SD related stuff is DOA since they are still using 3.8
green plover
I'm running automatic on 3.10
dense tapir
glad you are but you aren't on Colab, heh
Google will take forever to upgrade to 3.10
green plover
\stable-diffusion-webui>python -V
Python 3.10.0
oh. sorry I misread
I thought it said they had upgrade to 3.10
sometimes my things don't register fast enough. I'm sorry
patent onyx
hey, atm iam "working" on a bath scene and all the images i receive are kind of unsharp. Has someone an idea how to fix soft images and make them sharper?
wispy nest
how does high res fix work? i get blurry results, like if there was vaseline over the lens
gaunt heron
High res fix should give sharper results. it upscales with whichever Upscaler you choose.
wispy nest
which upscaler do you recommend
patent onyx
i´ll try the high res fix
gaunt heron
Latent should be fine.. I haven't used one other than UniversalUpscalerV2 in a long time. There are lots (https://upscale.wiki/wiki/Model_Database)
wispy nest
i tweaked a bit, but the results aren't too coherent
patent onyx
high res fix is crashing on my macbook 😦
gaunt heron
yikes haha
wispy nest
yeah, it's truggling with my 8 gb vram
patent onyx
ok i can´t do more then 768 i think 😄
wispy nest
wow, how did you make that
patent onyx
upscale with "Ultimate SD upscale" works pretty good
wispy nest
stoic pivot
What do you get when "photo [(ugly male)|(cute female)] portrait"?
wispy nest
eerie
gaunt heron
I want a big robot TV
dense tapir
🙂
oblique nimbus
wone ward prompt "stuff"
stoic pivot
Merry-Springmas, "muscular colorful [(female fairy:0.5)]|[(male Santa:1.2)] dressed in [dungaree]|[(tutu:1.3)] dancing on a farm road"
oblique nimbus
one ward prompt "simon stalenhag"
dense tapir
dense tapir
oblique nimbus
one ward prompt "bug"
oblique nimbus
crystal deltaBOT
dense tapir
stoic pivot
My Unicorn embedding failed so hard, I'm still learning.
dense tapir
robust valve
I can't create
patent onyx
why
cyan snow
fiery lichen
wispy nest
portals are kinda interesting...
hasty nova
portals are kinda interesting...
i've been wanting to make a portal embedding recently
wispy nest
i've been wanting to make a portal embedding recently
I think already existing models doing decent job with it , the only problem is - it's always continuation of same thing inside of portal, hard to make it look like it's leading to other dimension
hasty nova
I think already existing models doing decent job with it , the only problem is -...
maybe I could make an embedding to fix that... would be a bit more complicated I suppose than a normal embedding. I could probably make it work though. I could start working on it tonight.
wispy nest
maybe I could make an embedding to fix that... would be a bit more complicated I...
sounds fun, I'd check it out
wispy nest
maybe I could make an embedding to fix that... would be a bit more complicated I...
When I think about it - it's probably easier to inpaint insides of portal with another thing or do region control with another prompt inside portal (well, when region control will work better)
hasty nova
When I think about it - it's probably easier to inpaint insides of portal with a...
region control worked well for me, but inpainting is another solution
tame crest
Хуйня
glossy herald
for now yeah, it seems so... not sure I get why either
glad grove
hello, can someone help me get a skin change for this character?
i’m trying to make a “food” version, specifically ice cream
green plover
and I got Pandora to play nice
faint sage
Help why does it look like this https://discordapp.com/channels/1002292111942635562/1002602742667280404/1082114152232931358
sinful mirage
Help why does it look like this https://discordapp.com/channels/1002292111942635...
try installing it again maybe, just install a fresh version then drag and replace, if youve already tried just restarting it
stoic pivot
I do write odd prompts, this one was dirty and based on "Religious art" that included words like "perverted", "absurd porn" and "Trending on artstation", but then I changed to a Protogen model and it gave me this, and it was much better 😄
Edit: It was not as perverted as it sounds much involved mushrooms creatures in a magical forest".
cursive lance
echo apex
Hey, my first time playing around with the platform and I've noticed I'm getting much better results from v1.5 than v2.1 with my test prompt. v2.1 just gives me animated style images
What's the big model difference here?
(How) should I change my prompt around to adapt for v2?
Prompt:
In the midst of a thrilling space battle, Yoda sits securely in his seat on a Republic gunship. In front of him, the controls and monitors illuminate the cockpit with blue and green lights. Yoda's wise expression reflects his anticipation and readiness for the fight ahead. Outside the window, laser blasts from enemy ships trail past the gunship, illuminating their hulls with brilliant flashes of red and yellow light.
dense tapir
Up your negative prompt game for 2.x
regal sonnet
Can anyone help me, my hard drive was formatted and before that I was having VAE menu up here and clip skip but now I can't remember how I can make them appear again
gaunt heron
regal sonnet
thank you so much
waxen sigil
vagrant shell
EREEEEEEEEEH
wispy nest
plain seal
sponge bob cyberpunk
green plover
Just the results of a request from a public image requests group
pastel mango
catgirls say hi, in two different art stylees
smoky oak
anybody know how to image to image gen at higher resolutions than the 2048 limit per dimension?
smoky oak
change the max in there, use any text editor
oh! amazing, let me have a look see
pastel mango
"txt2img/Height/minimum": 64, "txt2img/Height/maximum": 2048,
for example
"txt2img/Width/minimum": 64, "txt2img/Width/maximum": 2048,
glossy herald
yep, best feature
pastel mango
yeah
it's great
just make sure you create a backup
in case you change something that shouldnt be changed
glossy herald
you can even have multiple different ui-config.json
And have an argument in your webui-user.bat to say witch one to use. (or duplicated your bat to have both options)
pastel mango
no problem
glossy herald
well, that crashes for me, even on 24GB ^^ hope you make it work :p
pastel mango
if you tile it with hi-res fix
smoky oak
I am not gonna go that high on both axes lol
pastel mango
you divide the work
glossy herald
yeah but still, after 15 minutes of making all tiles, it crashed on the last step, trying to stitch them together
at least in my last try
smoky oak
I can do resolutions totaling 4m pixels pretty confidently on my GPU
wait, the tiling function divides the workload?
pastel mango
wonder if its possible to set up and img2img loopback where each time it increases the resolution slightly
smoky oak
I thought it made the images seamless textures
smoky oak
wonder if its possible to set up and img2img loopback where each time it increas...
I was wondering about that as well, to keep more original detail
glossy herald
it starts by doing the picture small, then splits it in tiles and upscales them by generating more details, and then stitches together all the pieces for a last pass I think ? I'm not sure tbh, it seemed to be doing something like that
smoky oak
I had no idea
pastel mango
what you think of my catgirls? I think my model merge does a pretty good job
smoky oak
I am already confidently doing 4m pixel images on my GPU in single generations
smoky oak
xformers really is magic haha
pastel mango
gpu that does it :p
smoky oak
wanna buy me one too?
buy you one what?
glossy herald
GPU I think
smoky oak
what GPU do you have?
pastel mango
3070 ti, great gaming gpu, bad AI gpu
smoky oak
hahaha lmfao
pastel mango
too little vram
smoky oak
I am using a 3060ti lmao
pastel mango
ah lol
smoky oak
8GB of VRAM is plenty for well over 1080p generations
not quite 1440p from what I have seen (without tiling)
I had no idea thats how tiling worked
flabbergasted lol
what
:C
smoky oak
ok, so I am much more VRAM limited in img to img, interestingly enough
granted, this is AI upscaled, the actual SD output was 2560x1080
pastel mango
I tend to run hires fix though, but I should get the extension one that's improved and start latent coupling
smoky oak
I use high res fix as well
though I find at really high resolutions it kinda just starts making noise rather than detail sadly
like here
pastel mango
brb food
smoky oak
thats gonna crash 😅
jesus, is tiling like way slower?
this gen is taking forever
what in the hell did it do to my image
glossy herald
when you go over what your VRAM should allow, it splits the work, but that means lots of loading/unloading from VRAM
so alot slower
strange result though lol
smoky oak
but, that doesn't make any sense. I can do this resolution super comfortably
let me turn off tiling
static tusk
tiling make the image into a texture or similar
smoky oak
Thats what i thought, but they said otherwise-
glossy herald
lol on the sizes I'm trying for fun, it droped me at 52s/it currently xD
not sure why it is not even taking 13GB VRAM though
I have plenty left
smoky oak
so like wait, am I misunderstanding what you guys are saying about tiling?
Cause did you guys not just say it splits the image and then combines?
glossy herald
we may be talking about something else here
there is the "tiling" checkbox, that does what pure_fire just said, make your image into something you can tile, like a texture in a videogame
smoky oak
there is the "tiling" checkbox, that does what pure_fire just said, make your im...
so then what tiling were you referring to?
glossy herald
and there is high res fix, that does not do the same thing, but I said "tiling" there because there is a tiling process inside how high res functions, I was explaining what I understand of it
sorry for the confusion
glossy herald
but tiling is also a great option, not even remotely useful on pictures like yours though
smoky oak
I worked with the barium project, standard stable diffusion is massively outclassed for game textures
sadly I wasn't taken with the team ._.
I know they just made a massive sale to unity to incorporate their phenomenal AI generated fully passed 3D textures into unity itself
these were just the beta test materials I did the promotional renders for
that tool is reallyyy damn good
glossy herald
wow that slaps
smoky oak
generates displace, color, normal, ambient occlusion, emission, reflection, roughness, and metallic maps for 100% tilable AI generated images in like 20 seconds flat
static tusk
nice
smoky oak
and before it was sold, plans were to add far more ontop of even that
did these promotional render for the site as well
it used all stable diffusion, but mixed with his insanely good tweaks, and man, that shit was fire out of the box
(all of these results were achieved within less than 1 week of starting the project)
glossy herald
well, you were right, I OOM'd
smoky oak
of course you did lmao
glossy herald
^^
Time taken: 11m 26.88sTorch active/reserved: 19440/23474 MiB, Sys VRAM: 24564/24564 MiB (100.0%)
but I should be able to do 5120x1440
smoky oak
you reached for a 50 million pixel image
glossy herald
reach for the stars, get slaped in the face
smoky oak
I on 8 GB VRAM can pull just over 4M lol
smoky oak
so theoretically, you should be able to do like a max of 14M
still well over 4k
but I have one issue with high resolution images in stable diffusion, and that is that their detail just... falls off after a certain resolution, and I don't understand why
glossy herald
yep, it's because the model is really not made/trained for that. 1.X is trained on 512x512, and 2.1 has a 768x768 variant, but it starts repeating after that
I have a nice faq on that
crystal deltaBOT
glossy herald
it's duplication, and it's also "less details" that happen, but some more steps help on the quantity of details on those sizes sometimes too
smoky oak
for example, this image is 1080x2560
glossy herald
I'm doing another test in my walpaper size right now to see. it's still taking forever though
smoky oak
proper kelp
I thought I was getting some of this crap down.. I come in here and am absolutely lost. Lol :/
smoky oak
but the actual details are horrible
proper kelp
Like how did you produce that? What command, app, etc
glossy herald
I thought I was getting some of this crap down.. I come in here and am absolutel...
nah, don't worry, you are getting things down. It's just, it's an hydra. Each head you cut, there are 3 more notions to understand under it
smoky oak
this is one of my favorites
proper kelp
Wow
smoky oak
I am using A1111, realistic vision 1.3, high res fix, and good prompting
proper kelp
So how are you creating these? In this channel, in downloaded software?
glossy herald
5120x1440 worked !
smoky oak
5120x1440 worked !
bruh lmao
glossy herald
such a bad pic though
proper kelp
:/
glossy herald
I asked for "a cat"
Time taken: 3m 44.05sTorch active/reserved: 17382/23732 MiB, Sys VRAM: 24564/24564 MiB (100.0%)
proper kelp
Heh
glossy herald
lol wtf cat
glossy herald
duplications :p
static tusk
you got a litter of cats
glossy herald
smoky oak
I mean, the image details aren't bad! haha
proper kelp
Where can I attempt some of these?
glossy herald
I would need some controlnet on it
crystal deltaBOT
smoky oak
Where can I attempt some of these?
we are using local software we install and run on our comptuers
glossy herald
here are the main ways to get to the AI itself, in the faq I just linked
proper kelp
Ok ty 🙂 what software ?
smoky oak
I made this one as well
glossy herald
gtg, later guys
proper kelp
Ty!
smoky oak
gtg, later guys
peace
smoky oak
Ok ty 🙂 what software ?
it should be noted that us talking are like the top 1% of stable diffusion users 😅
just a sec
proper kelp
Nice 🙂
smoky oak
do you know what GPU you have?
(i said that to say don't get discouraged if you are not getting generations on our quality levels for a while)
proper kelp
I’m old as shit and have used graphic apps and 3D modelers etc for ever .. just new to all this, i guess most are
smoky oak
I am 26,792 images in lol
smoky oak
I’m old as shit and have used graphic apps and 3D modelers etc for ever .. just ...
are you able to check what GPU you have in task manager?
smoky oak
I have 1800 hours in blender
Oh man, I have many more than that 😅
proper kelp
Cool 🙂
smoky oak
4 years of daily simulations
proper kelp
I was a wanna be
smoky oak
sucked the soul out of me
proper kelp
Sweet
smoky oak
I hate blender now lol
smoky oak
got a taste of well made software and never looked back
proper kelp
Oh
smoky oak
I just need to know what GPU you have
static tusk
Ive used Maya for over 15 years
smoky oak
Free though I’m poor!
what we are using is free
proper kelp
I have gtx970
smoky oak
Ive used Maya for over 15 years
maya-pologies
proper kelp
4gb
proper kelp
Yeah
static tusk
heh
proper kelp
Old 😦
proper kelp
Trust me you do not want to hear my fucked up story. :/ I’m lucky to be able to still game so I’ll count my blessings
K
smoky oak
tho i would say it starts getting good at 8GB VRAM
smoky oak
photoleap, just a sec
smoky oak
oh right, another one of those AI image generator apps
chances are, its running off of a watered down version of stable diffusion
which is... us lol
proper kelp
Mhmm combined with video edit
Cool
I want in
You have no idea lol
Guess after I work up a1500 video card
smoky oak
sadly its a minimum spec requirement, and you fall below it
proper kelp
Or more? 🙂
smoky oak
doesn't have to be anything extreme
proper kelp
K
smoky oak
I have i9 and 32gb ram etc just old video
sadly stable diffusion only cares about GPU, I know it sucks
proper kelp
I get it man 🙂
I started building my PCs at about 12, first was a 286, I’m 47 now
Been hooked on pc and graphics since I was a kid
smoky oak
if you'd like, I am a free for now, i can try some generations for you
smoky oak
basically, yeah
there is a lot more to it than that, but I can use my experience to translate it 😅
proper kelp
Outer space, Blackholes cover his eyes, digital art, slightly dark, confusion, dynamic lighting, high resolution, 8k, great detail, octane render, blender render, slightly gothic
That ok?
Heh
smoky oak
oh yeah, you get it kinda
smoky oak
yeah, SD uses even more excruciating detail haha
sterile kiln
Guys, do you train your LoRA models with default 1.5, or with custom models ?
smoky oak
Guys, do you train your LoRA models with default 1.5, or with custom models ?
I use 1.5
alright, lets see if I can get anything cool with that prompt
proper kelp
I cringe at the site of all these young ones running to play ‘retro’ games.. I struggle through each year dying as graphics went from ascii to actually colors to sharp color images etc.. no way in hell am I going back
Ty
proper kelp
How about this? Half rabbit half woman gazing at stars, digital art, slightly dark, dynamic lighting, high resolution, 8k, great detail, octane render, blender render, slightly gothic, master piece, unseen technique
Wow
Awesome
sterile kiln
And, Aitrepreneur says that you should not exceed 1500 steps, that you have to do number of images divided by 1500. Is this true in your opinion?
I see on CivitAI, lora models that are several hundred MB, while mine do not exceed 30 MB, do you know why?
smoky oak
like any of them specifically? I can upscale them
smoky oak
And, Aitrepreneur says that you should not exceed 1500 steps, that you have to d...
truthfully, it depends so much on subject, base model, image res, and your end goals
smoky oak
That is so crisp
these are low res gens 😅
do you want me to full res any of the ones I sent?
sterile kiln
these are low res gens 😅
i'm training on 512x512 with 15 images
smoky oak
i'm training on 512x512 with 15 images
I would say 100 steps each, but do yourself a favor, in the settings, set the epochs to 2
proper kelp
And these won’t do any NSFW or whavter?
smoky oak
that means it will train one time, and then a second with what it learned from the first
smoky oak
And these won’t do any NSFW or whavter?
they will. there are hundreds of models dedicated to porn 😅
proper kelp
And it’s free? I just need the gpu?
smoky oak
yessir lol
sterile kiln
I would say 100 steps each, but do yourself a favor, in the settings, set the ep...
This is what I did the first time, sometimes I have good results, like with my face, but I have another model that is not very successful, I put 200 instead of 100 Is it going to be better?
smoky oak
I'd say its a reasonable system haha
proper kelp
Can I use the results to make animations and post them on TikTok?
That’s what I want in the end
smoky oak
Can I use the results to make animations and post them on TikTok?
yup, as long as you don't break the guidelines (basically misinformation or violence and stuff)
proper kelp
Ah awesome
smoky oak
I don't remember where the licensing is
but yeah, you can sell the images you make
but it is against the rules to impersonate people with it
so if you generate an image in a specific artists style, it is against the TOS to pose as said artist
sterile kiln
@smoky oak ?
smoky oak
This is what I did the first time, sometimes I have good results, like with my f...
its all a balancing act. It depends so much from case to case that i try something new with each one and tweak until it works for what I need
smoky oak
for example, I did a taylor lautner LoRA and it made it great in just 1500 steps, but my shrimp one took 12,000 😅
(aka, 8 epochs of 1500)
sterile kiln
for example, I did a taylor lautner LoRA and it made it great in just 1500 steps...
12 000 steps with same amount of images?
proper kelp
I think I’m going to try and hang out around you guys more often if that’s ok.. I got shit to learn
Heh
sterile kiln
it was 4 epochs
what change epoch?
smoky oak
I think I’m going to try and hang out around you guys more often if that’s ok.. ...
you have a lot, I have just shown you the smallest little bit, have a look at this
you can use several different AI powered methods to extract information, poses, depth, and all sorts of shit from images now
oops
marked as NSFW
proper kelp
Cool
smoky oak
proper kelp
I read some crap about chatgpt ai being able to write in python for you? Is that description based as well?
You know?
smoky oak
we also just got the power to draw where we want things, and also divide images, as well as run entirely different AI models on different sections
smoky oak
I read some crap about chatgpt ai being able to write in python for you? Is that...
yes, but its not very reliable
proper kelp
Wow and ah k
smoky oak
"code me a python script to do such and such, free of bugs, syntax errors, blah blah blah"
there are even models for architecture extraction
proper kelp
Still can you freakin imagined the world embracing all this AI? Every simpleton in the world able to dream up master pieces and have AI spit em out, eventually novels, applications..
The world would like morph over night once everyone was taking part
I’d think
proper kelp
And then.. Elon will sit back laughing as it becomes self aware and enslaves us
Lol jk
Oh it’s coming
Clearly
If we don’t rush WW3,first
Heh
smoky oak
true lol
proper kelp
I have my iPhone tell me ‘now charging master’ when I plug it in.. so I cna enjoy ai calling me master awhile before we have to call ai master
Hehe
smoky oak
oh also, you can train the AI on any subject, style, or concept in just liek 1 hour if you have a GPU with 8GB VRAM lol
proper kelp
Epic
proper kelp
I see a future of people creating insanely detailed personal profiles of themselves
Job info
Banking
Everything
And feeding it to ai
And ai running it from there
Every best price in town
Your fav bands coming in
Best gas price what’s in your fridge
Heh
What to wear
I’m sure
Women. Etc
Heh
smoky oak
I am only 19, and I am ecstatic for human and machine integration and the world that will bring. With enlightenment and real understanding of our world
hasty nova
it's gonna start turning into something like the Thunderhead from the Scythe series by Neal Shusterman (though we donn't have actual ai like in those books)
smoky oak
lol
proper kelp
For a long time and you meeet the right one.. but!’
Even then
Life with her two years
First
Still love her?
Great
Hate her?
smoky oak
how assumptive of you :p
smoky oak
here is the other prompt
smoky oak
there are also models for actual furry/anthro generation as well
smoky oak
she can be, but I can't send it here
proper kelp
DM?
smoky oak
Sure, tho I am not too fond of NSFW women generations. Call me... not partial lol
smoky oak
I am down to do it, just not much lol
smoky oak
I am not into the female body, heh
proper kelp
Ah that’s cool
My sister isn’t either
Heh
Truth
What I meant earlier
Don’t get married to ANYONE
Lol
People suck
smoky oak
haha, I get it lol
proper kelp
Lol
But yeah .. my situation is fucked
The vaccine..
I took my family
Heh
We all came home
Few hours later I lit up like the forth
Head to toe
Neuropathy
:/
smoky oak
Hey no way
proper kelp
Have been burning day and night since
smoky oak
I suffer from neuropathy
smoky oak
yeah, i know, I have it too 😅
smoky oak
cold for me
proper kelp
Not all have the same
smoky oak
being cold makes my neuropathy worse
proper kelp
Yeah mine is heat i can’t even lay in ed
agile eagle
Hi
smoky oak
My mother and I have fibromyalgia, genetic, non injury related, so we have an onslaught of issues from that. One being neuropathy and allodynia