#🏞|general-with-images
1 messages · Page 64 of 1
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 #⭐|pow-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 #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝|tech-support for any problems you encounter while installing it locally!
captain shrek
shrek lora?
this new GPU is so nice for training bro
no, art style LoRA
my first style LoRA
here is hoping it works
Yo
that's the illustration of this stage
@obsidian sky you're an incredible person and a great friend
You are an incredible person and great friend as well
So far that's a high level to start at
Don't fuck it up jk 🤣

The other models I have messed with are Edge of Realism, and Cyber Realistic, both of which can match Z in some regards
elijah wood, football player at the movies, the warm glow of cinema encompassing him and basking him in its soft emanating light as he smiles and laughs at the movie on the screen
fp16 are smaller with negligible quality loss
^
well "non fp16" implies fp32
They also use less GPU resources to generate, and FP32 also tends to use more VRAM
it'll possibly work better for CPU inference
we aren't. you knock out it of in the. park! every of the time
awww, thank you haha
I have never done a style LoRA before, and this one worked pretty damn well I'd say
Tested it, lost to cyber real and edge of realism
it has a tendency to make things look like anime
made something like this earlier with analog madness
its good
yeah, its not bad, its just not as good as other options out there
is it 2.x?
You of all people commenting on plastic is peak comedy to me lmao
♋
tried to recreate that lady
that's a lot of hair, god dayum
knowing you it's probably awful 
I have made 6 LoRA's since you challenged me, hurry the hell up bro
analog madness is ok 🤷♂️ just gotta use good negative prompts
Again, it is a fine model, but you can get even better results out of some other models
i am listening 😄
my model collection is growing every day
Cyber realistic and Edge of realism beat analog madness in my testing
super easy to LoRA
Civit. I would recommend Edge of realism, or Zovya (the V1 model, not the ultra)
IDK how many times people have to disprove that for people to stop spewing that nonsense lmao
ok mr rich man
sure bud
but dreambooth also costs money no ?
omg, not that video lmao
That video is almost 4 months old, and has no information on new things like Lorycoris, or any of the new technologies lmao
4 months old is already considered lost in time, ancient, last century, lost knowledge
also whats Lorycoris
The only limitation of LoRA's is they have a token cap, so you can only fit so much in. It has 0 to do with "lower quality"
A new subset of LoRA
There are several new ones
Its not a fair comparison since you have no idea how to train a LoRA lol
Alright, lets see it happen
watch an Aitrepreneur video on it, I'm sure you'll be making trash in no time lol
you can train one in 30 minutes if you have a non-shit-tier gpu from 2017
?
my 1080 TI is kind of out of breath nowadays
i have epilepsy and live off disability so my options are limited, gotta live with what i got
maybe new hardware gen soon
I just got a 3080 for $320, keep an eye out for local sellers
just stay away from 4000 series lol
gotta find my local data center, observe the high ranking employees
find out what kind of bars they go to
befriend em
- ????
- cheap A100 gpus
my friend has an A100, but he won't let me play with it lol
hes always too busy doing NDA stuff on it
you can do 8k on a 6GB GPU
yeah but you also grow a beard and hemheroids while doing that
I did 16k on a 8GB GPU, just gotta wait for it
You need a 80Gb GPU for 4k smh
16k on my 8GB GPU about 22 minutes
i did 2500x4000 today and it took 40 minutes
now that I know what I am doing, I could do it in much less, and I also have a much faster GPU now
oh, cause you are on a 1080ti, that right
I forgot that
big rip
Hey, it could always be worse
Sell you liver
my friend runs SD on a 4 core 3 year old windows laptop
I offered to just let him use my hardware, but instead he chose to sit through 14 minutes per 20 sample 512x512 DDIM image
He trying to heat up his room with laptop
@bold cave do you do prior preservation
is that what you mean when you say "regular images"
Laptops put out way less heat that desktops, unless I didn't understand what you said
oh, fruit is typing
thats a rare sight
i am eagerly awaiting the new hardware gen, praying its gonna be photonic chips with lots of tensor ops per sec
fruit 
👁️
👁️👄👁️
furby face
👁️ 👁️
👃
👄
cursed
Who is fruit
admin
😄
i want to do it
but what should i finetune DF on
With git I force push when it's my bosses lunch break and then add a bunch of commits
fruit 
Fruit
fruit
WOOOOOOOOOO
can you upscale it
like in SD upscale
tiled controlnet thing
to what fuckin resolution bro it's already at the max
I always forget that you don't
when i look at an image, i want it to suck me into its reality so i can get lost
that's called mescaline bro
@sterile temple ha, tytana
Create a pointillism-inspired artwork that showcases the technique's hallmark use of small, distinct dots of color to create a visually striking composition that plays with light, color theory, and perspective, and captures the essence of the subject matter in a unique and captivating way```
i stole this prompt from someone and edited (added like 2 words, now its mine)
What is this resolution. This isn't the 90s anymore smh

i'll take it as a compliment that you all want such high-res versions
high res this
sets VGA monitor on desk
that's Floppa's prompt
it fucking HURTS my eyes
thanks bro
i have never known such pain
oh
Looks like art that takes 5mins to make but sells for 5 million cuz hipsters
that img
is that the prompt i posted
ye
I wish I was a cat
just glue some fur and whiskers on your skin and demand people call you a cat
dots, lots of fucking dots, dipping dots, dots inside dots, dots morphing into dots, osmosis, the mitochondria is the powerhouse of the cell
pointillism in its essence
your prompt
it's being the powerhouse of the cell
try the cheesedaddy models, they're my fav, i like v35 more than 41, but 41 has more photorealism in it. my mix is mainly those models with a bunch of other random crap thrown in
where sauce
I like cheese
yeah but
thank you
here's the same pic with cheese daddy v35
you can see the similarities, but it's more artsy
nice. sasqwatchrek
upscaling anything that was made with fractal in it just gives insane results
just compare
still have to figure out the tile borders tho
here's v41 version. a bit more realism in there, but loses some saturation
first you look and see, you think its a glass with marbles
but its actually only marbles made of marbles
and there is no glass
padding to 128
mr paddington bear
which padding
dang
have you tried the multidiffusion upscaler
never heard of it
only did SD upscale with controlnet so far
whats the difference
ty
not exactly sure how it works. lots of overlapping tiles? some optimizations to make it use less vram. too dumb to understand
ah it is regional prompts, i thought of that earlier
because SD upscale uses one prompt for everything
@bold cave which model of mine were you trying to train btw?
I don't think it uses regional prompts, but you can use the controlnet tile with it. Here are some upscales I did using it
insane
To second the conversation, I have had no issues with training LoRA's for it, just for that bit of info
probably no way to train the ultra tbh. you're not supposed to train pruned versions as all the training stuff is removed. just try the standard V1
if the V1 doesn't train, then I'm at a loss. it's just a mix. nothing special
I know, I am just sayiong its not exactly an additional network in general issue

🤷♀️
like I said, it's just a mix of models, nothing weird going on like the ultra
can you train a lora and mix in the lora?
you can, but Sen thinks LoRA's are trash lol
it's fine for a face. I honestly prefer a TI for a face. no need for hundreds of megabytes
nice, not seeing any seems
Yeah, I can agree with that. I do LoRA's cause I have a lot of space and they train outrageously fast
my new style LoRA took 3 minutes 39 seconds lol
My Raihan (from pokemon) LoRA I did for a customer took about 11 minutes to train, and the results are still flooring me haha
the show
vs my LoRA
This was a digital art take on him
damn thats some quality
Didn't corridor crew do that or something
I saw the froot typing again
this was my first time doing an anime style kind of guy, and I am still shocked how well it worked haha
its so hard to keep up with everything
at this point there is new things every week
or every few days
hell, some things are every day, or even more frequent
i've been tracking https://github.com/trending
and its insane
some repos get 10k+ stars in a single day
Too bad only a max of 340mm vid card.
oh god that is a specimen
at this point wouldn't it be better to just buy computing hours on lambda or other hosts
Hahaha
no, thats WAY more expensive
no, the money factor is what makes it expensive lmaop
I even had 1990s to 2010 flashback with Inwin cases. Damn.
compared to buying a good gpu nowadays idk
I guess there is a market
Does depend massively on where you live,t o be fair
like here where I am, good GPU's are not hard to get at all
my new GPU is a testament of that lol
What I find amazing with so many cases is they 100% straight up block airflow as they put some glass in front of it. Luckily Steve from GN calls those that do that out.
a normal 3080 here is like 600-700€
Thats what they are supposed to cost tho
They are $700 MSRP
Oh, retailers are open boxing cards and some 4080s are falling below 7900XTX prices now.
Ah
I got my used one a few days ago for $320
but that was a lucky find
I know this I am buying a Crucial MX500 1TB just for SD. Only SATA but it took 3-5s to load a model on mine yet on my HDD can take up to 3m just depends (on what I have no idea).
hell, the 4090 was just tested not long ago on PCIE 1.0, and it only had a 22% perf loss IIRC
wtf
just need a crane so it doesnt sag
yep, which is all my b450 offers
850W is fine
i got a 1000W with like 4 years of warranty left
yeah, here it is
24 GB
PCIE gen 4/4 = gen 1 at 16x
ooooooof
you realize that the new connector for Gen 5 is a max of 600w, right? That means the card is not going to go above that.
i'm still kind of reluctant on buying anything new because silicon electronics is kind of dying
You can go above it
but it is reallyyyy not smart
Linus did it on his 4090 IIRC
he hit like 900 Watts I think
BUT
the question is, can i pay off the gpu by making images
they had to use their own power solution including janky ass clips, and severe overvolting lmao
potentially yes
I am making decent money selling LoRA's right niw
Sure, I guess you could make it but the wiring is not made for it so a lot have clamped it down in bios
yeah, they had to use a special nightmare bios
A 17% perf hit for gen 3 at 4k is amazing
which removed 100% of the current and temperature limits lmao
so if they go too hit, the card just died lmfao
That's Linus
yeah hahaa
damn, Linus wasn't himself that day then.
you read that graph wrong lmao
rthats gen 3 and 4x, not a full 16 x slot
???? My gen 3 is 16x
full gen 3 can still run a 4090 no sweat
IIRC Gen 3 is like 32GB/s
other slot is gen 2 x8 I think
buying hardware is pure agony when you're poor
I am waiting for the case manufacturers to catch up. They are trying
this is the equivalent of gen 3 16x
98%, which is within margin of error
what are the chances that we will see silicon photonics next year ?
each PCIE improvement is 2x the speed of the last
its probably already happening, but for the consumer, probably no chance
if NA doesnt do it, china will
I see, X16 is 100. Cool
its kind of also the reason why they fight over taiwan
yessir, 3.0 is still plenty fast for GPU's
the new PCIE gens only get leveraged by super SSD's
X8 has a very slight perf hit I see so in essence I could even slot it in my second slot. Amazing
which sabrent just released a maxed out nvme 5.0 drive IIRC
smartphones in 3 years may have multiple orders of magnitudes more tensor ops per second than our current hardware
its like 14,000/14,000MBP/s
Yeah, Gen 5 for NVME M.2 is insane
yup, and thats in gaming, which has WAY more throughput than AI
gaming doesn't even use AI for gameplay elements other than graphics at this point in time
its boring af
Gen 2 x8 is the same as gen 3 x4
you're not having to send millions of pixel values dozens of times per second when doing AI, unlike gaming
welp, 6% loss in a game. Can you even feel that?
only if you are at 4k, and not CPU bound lol
LOL, even a 13900 that stupid site said was a bottleneck for it
at 1080p, it is
but it depends a lot on the games
but if you have a 4090, you sure as hell should not be at 1080p lmao
unless
1080p and over 9000 fps
your eyes can see more than 24 frames
unless you have a 500hz top of the line 1080p monitor
I think its 500hz now
let me check
idk i think the new gpus are cringe, for whatever reason i feel like they are not what i want
Yeah, 500hz now
4000 series is a huge flop when not looking at the 4090/4080
even then, its still yikes
I am still sitting back, get my computer parts piece by piece, having a brew, eating some popcorn and waiting to see what a 7950XTX is like and what happens with Nvidia prices.
https://www.nature.com/articles/s41586-022-04714-0
i want this but for SD
Nature - Using a three-layer opto-electronic neural network, direct, clock-less sub-nanosecond image classification on a silicon photonics chip is demonstrated, achieving a classification time...
I am pumped for the 7950XTX
In each neuron, linear computation is performed optically and the non-linear activation function is realized opto-electronically, allowing a classification time of under 570 ps, which is comparable with a single clock cycle of state-of-the-art digital platforms
I am hesitant now as all of these things have been dismal failures
that is fast as light
I just hope AMD goes for the kill shot, cause if they can get decent perf at a good cost, they are gonna own the consumer AI market
at the end of the month at the show
these photonic devices can run off usb power
and the second AMD is viable for this work, you better believe people will jump ship and start doing their implementation work for them
After all, a LOT of the AI enhancements we have now are are not from multi billion dollar corps
they are from dev teams and individuals
combine that hardware with this https://www.nature.com/articles/s41467-021-25427-4
Nature Communications - Development of deep neural networks benefits from new approaches and perspectives. Stelzer et al. propose to fold a deep neural network of arbitrary size into a single...
and we have infinite power
manufacturers holding back hardware advances
You know I can't even pay 500 USD for a case that has all I want so I just sort of said fuck it
it feels like i did something wrong idk
the case manufacturers really got screwed and caught off guard with the new lengths of the cards
4080/4090 are big bastards
FE is decent and no big deal for most but Nvidia stopped making it
Oh, Oh, I actually just ran across a Strix 7900XTX
Yeah, the 4090FE is actually a lot smaller than I expected
Maybe it was 7900XT but it was 3 fans, and only 8GB
my new GPU is longer than a 4090 FE
Yep, 12inches for the FE
Going SSD there I have 1mm of clearance but can't buy a FE short of scalper pricing
my GPU is 12.7 inches
fucked
also, I found its only 323 mm, not the 338 I saw somewhere else
I OFFICIALLY could not put that in there
https://arxiv.org/abs/2303.01287
this sounds like some shit from a sci-fi movie
The optical neural network (ONN) has been considered as a promising candidate
for next-generation neurocomputing due to its high parallelism, low latency,
and low energy consumption, with significant potential to release unprecedented
computational capability. Large-scale ONNs could process more neural
information and improve the prediction perf...
Just think the FE is over engineered for its heatsink and fan and could be reduced by 16-25% and still be good. Now why are AIB all shitting 3 fan nonsense and 5 million feet long?
I know for Ampere, only NVIDIA had access to their shorter and better designed boards
Well, I think a lot of it has to do with their heatsinks used. No idea but if I were a board partner I would have sat back and asked Nvidia why in the fuck are we working with your ass when we get next to nothing out of it, get no access to the stuff we need, and you publicly shame us?
Let Nvidia go the way of the Apple model which is what they are trying to do.
no wonder evga left the gpu business
Right
not a bad choice either because photonics is right around the corner
so the market will be completely scrambled
since it does not require expensive fabs and can be done with 100nm processes
the training of these things will be so fast
I think it was evga who first mentioned that is the goal of Nvidia so they were getting out while the getting was good.
You want a Nvidia card then it will only be made by Nvidia for whatever price they will to charge.
2023-05-13 03:31:52,594 [INFO] Upscaling tile 3 of 9
2023-05-13 03:32:04,317 [INFO] Upscaling tile 4 of 9
2023-05-13 03:32:16,023 [INFO] Upscaling tile 5 of 9
2023-05-13 03:32:27,768 [INFO] Upscaling tile 6 of 9

that's weird as shit
he was just mentioning he never gets banned on this discord so I guess he did a naughty.
you have to change the MAC address on your computer unless it's a PC, that's a MAC thing
enable privacy extensions on the nic
get GPT4 to solve it for you in LangChain, duh
LMAO
but look at my Ultimate Upscaler thing
points to the server audit log held by Clyde
Oh hey, welcome back sen
this is better than UU
it only looks bad because it's so highly detailed.
kno wha im sayin? 
boy, that's less intensely awful but still not great
reduced the noising strength to .7 but it just doesn't fix enough of the image that way lmao
Send me the base image and I will give it a try
Right when my hood's Owl shows up as I can hear it outside
We have a little owl where we live now, and it likes to perch near my room occasionally
This one is over 8 years old at least, has a mate, and is a HUGE white owl. I only saw it one time when the hood had a complete blackout.
and when driving one time near my house
metallic owls
well i've got the upscaler code refactored so it's "production quality" now, eg. i can make changes to it without too much pain. but yeah them seams still there for now
need to add padding to the tiles and try different blend strategies
can you create mask automatically that covers just the seams and then do an inpainting over?
i assume so but i'd need to prompt the inpainting correctly, no?
also it'd need to hold the whole image in memory
the seams are there because it's each tile being upscaled separately to 1024x1024
the idea i've got at first is a transparent edge blend on a chessboard pattern
it's tricky. the tile split needs to give them padding, and that padding needs to be made transparent, presumably after adding an alpha channel to the tile
mmm yeah that's true about having to do the whole image using too much memory
the effect of denoising = 1.0 is really weird
when doing this tiling upscale i mean
it's fine with the entire image
how to use the bots?
none here
where ?
try Blue Willow, perhaps
Probably dying due to all the gens
here is not working? Blue Willow is in bad quality
im assuming you busted something in there when screwing around with it lol
Watch what happens
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 #⭐|pow-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 #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝|tech-support for any problems you encounter while installing it locally!

I did not :<
When it tripped on you the first time that may have damaged something in it. Rather it was going and the strain has now weakened for it to just shut off on you at random times until one time it no longer works. Hope it doesn't pop and take the system with it
It never tripped
Yes, it did and you know it
turns out my PSU didn't trip, my GPU crashed from the OC I had
Did you try taking the cartridge out and blowing on it?
Oh, you overclocked it right out of the gate. I am done here
Those times when my PC freaked out were from GPU crashes, not PSU issues
it was before I realized that this GPU pulls WAY more power than an FE
know any AI that is at least a little better than BLUE Willow?
since then I have been running power limited with an OC that is stable, and it draws much less power
down from 360 watts to about 260 watts
which is not much higher than my 3060ti
it was like 220/230 with my OC
LOL OVERCLOCKING
you can overclock and under power
you can overpromise and under deliver
It was stable, but I am not sure what just happened
transient power spikes that surpass the card's VRM frequency
I think it was a power shutter, cause I wasn't even doing anything other than watching a video
NVIDIA is notorious for these
and we have had a couple electrical issues at this new house
you said your house's electrical was fine
I was saying that as in I knew that the time it was not that
when I had the black out, that was my GPU crashing
I overclocked too far
man... wish I knew ahead of time that SD was coming and bought a laptop with an 8GB card. The 3060 6GB is okay, but sick of the CUDA out of memory errors
I have done 1000's of gens and several LoRA's since I found the stable settings, so I am just gonna stick to those again and if it happens again, I will lower power even more
ah lol
ok, some windows error logging shows it was actually unstable OC, not PSU
interesting

Alright, I just confirmed with actual PSU technicians, that the PSU I have should be able to handle a 3080 just fine, even without power limit. They said that this PSU is rated for 650 watts reliable, and can take up to 1000 watt peak transients. Its a cooler master MVE V2, and is rated B+ tier in teh global PSU lineup
They also said it should be especially fine since I am not running my CPU at the same time
and they are suggesting as well that it was just an unstable OC, as it crashed when not doing anything
I like the first in the 9x9 grid
lmao PSU technicians
I am talking about people that actively run and help categorize the international PSU scale
#7
regardless, it explains why my 3080 running at 370 watts was causing no issues
and why now at 260 watts it didn't crash from PSU, but from unstable OC while being so power limited
that's kinda sick but it really fucked up
oop lol
I'm think of making taking the face of #7 and combining it with something else
Sytan shut up you can barely notice the seams
because everything else is jumping out at you
lol it's so damn weird
this stuff is hard to debug
true, true
the shitty contrast is way more distracting
the seams almost look intentional when it's this broken
started messing with pixel art style generations
neat, i like the light it cast onto the surface underneath
crystal skull generates some pretty cool images
I am using some more keywords than that, but yeah
pretty happy with this one.
nice! :>
generated pixel art isn't as good as a realism gen with a pixel art filter
kinda, yeah
there is pixel art script on github i believe.
Its a lot more than bicubic
Bicubic is a down sampling process, this is using a different process
bicubic is pixel averaging
this is not
can you train a lora on eboy's style?
Like on a specific E boy?
just all the work you can get together
they've been around forever and done tonnes of stuff
is that an artist, orrrr
I used to have a massive framed poster in my lounge before it fell down and smashed 😦
yeah, Eboys are like... online hookers lol
i love this sort of pictures @sterile temple
that is very nice :>
tho IDK how you would train a LoRA on that as you would need to tag a LOT
They are packed and always see new stuff when you look at it
Let me see if I can find something you can train
i know there were such puzzles
I have 0 confidence that I could train that
They used to post just single character designs
but I haven't looked them up in years
Dracula is sleeping in coffin 😄
omai
Also, I have finally found a collection of prompts that are genuinely giving me Midjoruney digital art model esque results, and god are they pretty
this kind of stuff
very flowy, painterly, abstract
thoses are the prompts ?
bioluminescent acrylic pour
good is when you interogate image with clip, it create txt files in integorate and it can give very good idea when clueles. Many of film emulation and so.
can it pick up film emulation?
let me show what i mean. Mmnt
kodak portra 4 0 0, kodakchrome, kodak ektachrome
just few kodak, you can find there movie material and i think it is working well
bioluminescent acrylic pour
fancy boy
deep blacks, vibrant colors
oh yeah I've heard of kodakrome before for 90s look
there are for 920s as well
you would need a noise offset mode to get actual black backgrounds
going to try panavision panaflex x
true but I don't mind the almost black in this case since they are supposed to be glowing
I'll give it a go anyway
darker blacks, but definitely adding it's own style to the images too
have you tried fluorescent? I believe much more light, tried it on mushroom
yes new hairdresser 😄
swirling vortex style, very fashion
just film grain i like it
with florescent added to the prompt
interesting on the rare occasion you get an image that is perfectly split into two different images
i like that tiltshift look
whats the go to colab notebook for training loras these days?
i dont know. I never do colab. And i hear something about colab and SD somehow not working. But i can be absolutely wrong
not best but quite good result for stadium
😦
hi all 🙂
bro i wanna remake ricardo milos
2nd is pretty good design, shame SD can't keep track of where the moon is when it's divided up
what model are you using?
there is extension to save each step to not miss such images
if you like cfg4 and set it up to 7 it will save all cfg 1-7. Otherwise you lost it
what is the extension called?
1.5 vs 2.1
to compare i think you have to have same models, and it is imposible
i have so far got cfg 1-5/1-10
OK, so getting "good results" is a bit harder with 2.1, requiring more finesse and work than with 1.5?
Yes
This is one TEENY example but I've been ... yeah ... trying 😅
In 2.1 negative prompting is more important, most times, than the positive prompt.
I guess I first need to unlearn everything I had to go through, in order to get good stuff from 1.3/1.4/1.5
for 2.1 models you need to set the resolution to 768x768
not always but yeah
yea but like 90% xD
id say more like 95
Cool. Thanks a lot, folks, I'll keep playing 🤓
nah, I had good results at 512x512 but it depends on the model.
I would LOVE to see good results with the vanilla/base/unmodified/etc 2.1 (released 5 months ago)
I hunted for websites with examples/prompts so I could test on my machine but could never reproduce, not even close
Been so long since I did that
hehe, I bet, 5 months is a lifetime
day one of 2.0 release, and 2.1 I switched
CFG Scale 1-6
Verra nass!
woah
I can has with parameters? 😛
Just trying to grab PNG info 😉 guessing it's not the vanilla model
Nice work 👍
This is actually using one of my old embeddings I published
My results 😆
what did you put the cfg scale on?
I just grabbed PNG info and hit Generate 😇
It's 6 in this case
"Model: aZovyaPhotoreal_v1" aahhh
Cool! I'll just keep playing in my corner 😄 come back later ... have a lovely day, folks 😎
Depends

spiderman?
spider man in acid rain
i prefer spider woman 😄
🤨
nice fingers hehe
my first ultrawide, and yes not perfect but glad i can do it
it is just outfit, glowes
could be nice cave with fluorescent slalagmites
yeah. just trying to get some with a mist or smoke around them
cool
i want it to cow or pow but must invent something different 🙂
cow or pow?
Is there a way to add line-art to images that are relatively flat (like basic vector art)?
This is a render I made in blender but it has no line-art on the buildings for example
I ran it through Stable Diffusion and it looks 2d and not 3d anymore but still no line-art
Yes
shparkly
tried a bunch of things, this is the best I could do
Thanks for trying, prompt? And model? There is a line-art generator for blender (Pencil+ 4) but it's $500. Professional anime studios use it. I found a "free" one but it's for Maya. Also, there is a add detail Lora and thicker lines Lora. I used both and it added a bit of line, at least...
3d blender person car city buildings tree blue sky line art cell shading thick black outlines lora:xrs2.0:.41
just not getting what I imagined, which is like a floor of fog rather than it in the background
i see. will try it as well, probably with mushroom
although really need to go to bed, need to get up in 6 hours to get mother's day breakfast ready
good night
goofy ahhh
ok nite
junk pile
mmm shiny leather seats
starwars are attacking 😄
and what those lasers?
lasers-
those lasers are lasers
two tone lion
Rally cars!
:O
why u do dis?
idk
I made a character and ran him through img2img 2 pass, but he keeps getting lighter and lighter in each pass, because almost all models are biased towards being more light. How can I retain the original color?



dots




