#✨｜sdxl
stray warren
that's the best image i've seen this year 🙂
thank you malicor, you make me 
pallid path
wow nice
shy kelp
HDR photo of woman, dark-brunette hair, light-blue eyes, (f_stop 5.6), (focal_length 28.0), f/5.6, 28mm focal length, sitting in bed, contemplating, sunset, romantic, warm, revealing clothes, alluring, white sheets . High dynamic range, vivid, rich details, clear shadows and highlights, realistic, intense, enhanced contrast, highly detailed
minor pecan
spice skiff
I set out to create a 5K desktop wallpaper for my Studio Display.
I used SDXL base to generate the initial image in 5K resolution. The resulting render was lacking in coherence and exhibited all sorts of perspective issues. However, I liked the general feel and the overall composition so I decided to continue working on it.
I wanted to maintain the 5K image size without resorting to any latent or pixel upscaling. I decided to do a little inpainting to see where it would take me.
I used ComfyUI for image generations and I used Affinity Photo on a PC to create the individual PNG mask files and to handle the compositing of inpainted elements back into the master image.
Once ComfyUI had generated a suitable replacement element using inpainting (e.g. a person, banner, scooter, etc.) I would paste that element back into the master image file in Affinity Photo and then I would create a new masked section of the image for the next element.
Rinse and repeat quickly turned into an inpainting-palooza.
No fine-tuned models, loras, or control nets of any kind were used. Everything is base SDXL.
I created a short timelapse video to illustrate the process for those of you who might be interested. I've posted a higher-res version to the Stable Diffusion sub-reddit.
copper kraken
HDR photo of woman, dark-brunette hair, light-blue eyes, (f_stop 5.6), (focal_le...
Here is the image you requested.
gloomy lark
elder plume
is this using fooooocus?
opal swan
rugged bluff
A book cover titled "Principles of Marine Sediment Dynamics", designed with bright colors and a simple, abstract style.
shrewd timber
Please draw a picture. The ratio of the picture is 16:9, and the main color of the picture is dark blue. The theme of the picture is technology and intelligence. Thanks.
meager canopy
Here is your image...
gleaming chasm
Chinese city with a distant view of a rolling mountain range with snow on top, the image is of a boulevard lined with traditional Chinese buildings, skyscrapers in the distance, and an ancient well with a Chinese dragon hovering over it in the near distance, the image is colorful and Pixar-esque.
minor jungle
Chinese city with a distant view of a rolling mountain range with snow on top, t...
#1047610792226340935 bots offline. You can run Stable Diffusion locally or use online services.
void yacht
I don’t know if this is the right place to ask , but any good SDXL instalation tutorial? Thanks
minor jungle
I don’t know if this is the right place to ask , but any good SDXL instalation t...
https://github.com/AUTOMATIC1111/stable-diffusion-webui you can try this web ui first
south flume
I don’t know if this is the right place to ask , but any good SDXL instalation t...
Its as good a place as any, but first you gotta choose a platform/interface , SDXL is a model , to interact with it you may choose one platform or another considering ease of use, hardware, expectations, etc... you might wanna follow some YT guides to get started , apart dad jokes I like the https://www.youtube.com/@sebastiankamph , he is concise and slowly breaks things down , but there are plenty others
bright lance
north nebula
一个女孩被一只棕熊追逐
minor jungle
一个女孩被一只棕熊追逐
#1047610792226340935 Bots currently offline, please use Stable Diffusion locally or use online services
molten gull
whats the status of stable diffusion 3 by the way ? anyone got access yet ?
minor jungle
there are some people got access to sd3 but afaik they won't release the whole model yet
north nebula
一个女孩被一只棕熊追逐
Are you an administrator or？
minor jungle
no, just a normal user here.
north nebula
<#1047610792226340935> Bots currently offline, please use Stable Diffusion local...
So there’s no problem chatting here, right?
minor jungle
yes. but i assume you are asking for a image generation
north nebula
haha，It seems that there are only xl models here, not 1.5?
minor jungle
yes. this is a place for posting and discussing sdxl
north nebula
Which one is the 1.5 model? Or is 1.5 outdated?
north nebula
Which country are you from?I mean do you want to come to my place to play?Bring you barbecue and skewers😋
minor jungle
wait a minute you are chinese right
north nebula
ang，Chinese pure man
minor jungle
两个中国人用英语讲话讲了半天草
north nebula
那就是 中国纯爷们😆
minor jungle
哈哈
regal shell
hi
uncut steeple
gloomy lark
gloomy lark
gloomy lark
cyan crown
mellow tendon
rich olive
https://civitai.com/models/367854
Awesome
rich olive
Lol looks like fun, might check it out
copper kraken
it's even better
holy shit this is amazing
better than the cascade clipvision fun we had recently
copper kraken
glass forge
With so many cretaive crazy amazing things people come up with you'd think consistent characters and stuff woudl be like the most basic most boring thing they'd just do to get it over with.
fuse watercolor and cubism a portrait and add some rodent teeth and floating cucumbers and a dragon in a glass oven. got it. Get the image of the same caharcter from two angles... nope
copper kraken
that's not really that hard to do, all the tools exist
glass forge
yeah but not 100% its always kinda wonky
you get the pose kinda but not really i mena try getting squatting people or people laying in bed. and the angles... forget it anything that veers too far from flat frontal angles gets messy fast and then of course the textures, skin blemishes the minute details... forget it
copper kraken
yeah, if you want total consistency then you need to actually train a lora
glass forge
its like were halfway there, this is awesome for spitballing and coming up with amazing stuff but when you want to use that stuff for anything things fall apart fast. imagine a 3D program and you made a nice 3d object and everytime u move the camera the object changes a bit
when this will be sorted THEN this will change the world
until then not really
copper kraken
there's no way around that, no genius trick with diffusion models that can accomplish that without independently rendering a 3D model of a character first with more traditional methods
glass forge
i did loras
made some for 1.5 and some for sdxl
they dotn give total consistency
yes but if we go and just do the old way than how will this improve our speed and productivity? were back to square one almost lol
and even if we do use some 3D poses as references still we cant "smear consistent AI over it"
like clearly ai is having an understanding of 3D objects at this time... couldn't there be a way to promt something and select it with inpaint and "name it something like woman in red dress" and lock it down and really tell the ai to keep that "concept" as is and not change it anymore except the pose and lighting. and then whenever we promt that "woman in red dress" it will give that character. lol this could work for objects too maybe? and then you could generate up a bunch of props and characters and then mix and match them in photoshop or whatever and bring them back and just fix the lighting with AI and late paint in motion / lipsync, etc. feels like something that should be possible maybe... idk lol. wishful thinking. or is it
i never know with this, things that used to take 10 years take 2 weeks now
runic elm
||dog||
minor jungle
||dog||
#1047610792226340935 bots currently offline, you can run it locally or use online services like https://seaart.ai
opal swan
rich olive
opal swan
shy kelp
A boy
meager canopy
I'm pretty sure it's not a boy
peak dove
new ipadapter is very impressive
Some say it breaks ComfyUI? Has that been fixed at all? 🙂
meager canopy
Some say it breaks ComfyUI? Has that been fixed at all? 🙂
The screenshot is in ComfyUI, so clearly didn't break it there 😄
south flume
@copper kraken I'm experimenting with Ip-Adapter and Photomaker, If you dont mind,wich IP-Adapter model were you referring to?
opal swan
meager canopy
Still not sharing prompts?
opal swan
frigid willow
copy space, ultra detailed photo, forest silhouette in the shape of a wild animal wildlife and forest conservation concept in the winter night, northern lights dancing. Beautiful design for wildlife preservation, environmental awareness. ¯_(ツ)_/¯
copper kraken
Some say it breaks ComfyUI? Has that been fixed at all? 🙂
Breaks how so?
copper kraken
<@1208924372299939890> I'm experimenting with Ip-Adapter and Photomaker, If you ...
Sdxl plus vit H
warm helm
meager canopy
atomic crag
made some for 1.5 and some for sdxl
I trying to make a Lora now. And I am kinda confused on if I should make it on 1.5 , sdxl, 2.0 or 2.1. Or 2.1 blob vs 2.1 base
Also I am running dreambooth v23.0.15 and my layout is completely different than other videos I’m trying to follow.
I’m trying to make a hyper realistic model and I’ve gathered a bunch of photos 50+
peak dove
Breaks how so?
Apparently it overwrites the old IPAdapter, so stopping all the projects using the old node?!
gloomy lark
copper kraken
Apparently it overwrites the old IPAdapter, so stopping all the projects using t...
correct
tbh, it probably should've been released as ipadapter v2, so people could keep the old one too and eventually uninstall it once they were done
gloomy lark
pallid path
ok now THIS ended up being very epic
can't seem to get metal claws though
adding wolverine into the mix just makes him look like wolverine of course lol
lusty wolf
tawdry current
tbh, it probably should've been released as ipadapter v2, so people could keep t...
Nah, better to not split the user base up like that. It's mostly all done by a single dev, so he shouldn't have to deal with the headaches of separate repos. You should always consider everything in ComfyUI to be subject to change. If you like the old version, redownload it and put a gitignore in the directory to prevent it from being updated
copper kraken
true
im fine with it myself
its really not hard to tweak a workflow for the new nodes
takes 30 sec
pallid path
yeah the prompt coherency is really failing on me, even when doing this regional prompting
im trying to replicate something I have in my SD3 prompt list
and no matter how much I tweak the settings or the prompt (or even use a base image with high denoising to get good composition) it just won't work 90% of the time
copper kraken
whah is it
pallid path
look at my previous images here, two fighters, one with metal (wolverine-like) claws and the other with a katana
looking at eachother whilst doing a fighting pose
I bet festivalman could get this one right, but I just can't
burnt shadow
pallid path
epic
gloomy lark
look at my previous images here, two fighters, one with metal (wolverine-like) c...
lol, this is what happens when I accidently paste the giant instruction set with examples into the regional prompter.
but i post the stuff here that works. there's a fair amount of stuff that doesn't.
this regional prompter stuff can do amazing things, but it still hits the limitations on interactions that sdxl has unfortunately. regional prompter with sd3 will be epic.
pallid path
yeahh, well thank you anyway for trying
gloomy lark
it's definitely a weird one. i can get it to do all sorts of stuff but can't get it to do both of them. even if i make one of them a cat (so not a man at all), it only draws the cat 1 out of every 4 tries.
gloomy lark
yeahh, well thank you anyway for trying
look at that. sometimes claws guy, somestimes samurai, sometimes neither.
not sure what regional prompter extention is really thinking here.
pallid path
SD3 got this on first try (even if the result is a little distorted)
also idk why it's pixelart, I asked for digital artstyle
gloomy lark
OOOH. i have the answer. one sec.
pallid path
maybe "video game" confused it
gloomy lark
getting closer. 🙂 back in a few
pallid path
cool
mellow tendon
Still not sharing prompts?
There is always Interrogator: "a woman in a hooded jacket and hoodie walking through a muddy sea of water, splashing dynamic pose, long brown hair blowing wild in the wind elden ring cinematic lighting, monks!!!!!!!!! fire, tarot card goddess of death, wearing brown jedi robes, portraits of a woman enraged, dangerous cliffside, sansa, power pose, 2 0 0 0 ad magazine setting, by Sigurd Swane"
meager canopy
There is always Interrogator: "a woman in a hooded jacket and hoodie walking thr...
Interrogator never does a great job though.
pallid path
What about vision models
mellow tendon
It can be a good base to work on.
pallid path
What about vision models
okay no that would be OP with SD3 though
pallid path
Nice
gloomy lark
not happening with sdxl, even with regional prompter. sd3 is the answer
pallid path
yessir
gloomy lark
I've learned that you can still do a LOT of amazing stuff, but interactions? nope. possibly with inpainting, but I ain't got time for that
pallid path
can you show the ideogram result in like SPOILER or #🏞｜general-with-images
absolutely
gloomy lark
as you can see, ideogram results are pretty rough, but they give a great basis for img2img to clean it up
gloomy lark
well, sd3 should look a lot better too
pallid path
but unlike ideogram we can do highresfix, which will fix the faces!
or just adetailer, but I don't use those
gloomy lark
yeah exactly
pallid path
if I can do highresfix on the 8B with 12GB of vram I'd be fine for like.. almost ever?? idk how much the finetunes will bring as well
copper kraken
this plus this with juggernaut 9, prompt steampunk machinery in a city
gloomy lark
yeah with tiled upscale, i can even do it on my 3080 with 10 gig.
copper kraken
equals this
gloomy lark
one last one then back to work for me, one from a while ago, ideogram but with tons of img2img sdxl upscaling. i have high hopes for sd3.
pallid path
yeah with tiled upscale, i can even do it on my 3080 with 10 gig.
yeah im just afraid of tiles ruining coherency between tiles of course
unless there's gonna be some controlnet tile which is really good at stiching it all together
idk
your images look good though
gloomy lark
I've found that I need to latency upscale 2-3 iterations to maintain consistency, THEN I move to tiled ultimate sd upscaling. Then it works well, but yeah, those first couple generations need some vram.
pallid path
decided to sing up to ideogram to test stuff, here it is with my exact prompt
It's pretty good actually, I expect this quality from even the 2B model
copper kraken
same inputs sdiff prompt
pallid path
you know the style of the images and faces kind of remind me of 2.1 for some reason :|
for Ideogram
It's fun to use though
stone storm
are these the standard or base line sdxl models making these?
copper kraken
not entirely either exactly, but juggernaut9 which is sdxl
im on comfyui, load these images and youll get the workfflow
gloomy lark
I had no idea merging models and lora was so easy. I combined all my favorite things, dark arts images, proteus-rundiffusion, and andrea75c's cute3d cartoon lora into one checkpoint that does it all quickly. from left to right, dark arts, proteus in middle, final combined 3 way merge on right.
hah I should do that midjourney mimic lora onto this as well for gawdy over stylized rainbow brightness.
works with regional prompting well as well.
pallid path
this merge looks good
gloomy lark
this merge looks good
yeah it combines 2 models that have good prompt adherence, with the smoothing out factor of the cute 3d
pallid path
wow
pallid path
can you try the fighting one with this
gloomy lark
can you copy/paste the prompt that you liked from ideogram the best? the magic prompt one?
I'll see if that helps at all. My hopes aren't high for that one in particular. I think this giant at the gate works well because there's no subjects interacting with each other.
copper kraken
pallid path
I had this detailed prompt in my SD3 prompt list, I hope this helps:
A game screenshot of a fighting game in digital painting style. There are two yellow health bars. The characters are both black silhouettes against a colourful background. The background is a beautiful landscape of a lava mountain. The left silhouette character is a ninja holding metal claws and the one on the right is a japanese samurai holding a katana.
you could opt out the "yellow health bars" and "game screenshot", those might get in the way
gloomy lark
so this is just as is, that original prompt pasted in without changing anything or regional prompting.
pallid path
not bad
pallid path
SuperPrompt V1 gave these:
In the center of a vibrant canvas, a fierce battle scene takes place against a backdrop of lava-covered terrain. The left silhouette character is a majestic ninja, holding wolverine claws and the right behind it is a japanese samurai, holding a katana. The black background is reminiscent of a vivid landscape, with the lava mountain contrasting sharply against the stark white background. The painting exudes an aura of calmness and serenity, making this an unforgettable representation of martial arts in the most unexpected way.
In the center of the canvas, a fierce battle unfolds before your eyes. The black silhouette against the vibrant backdrop of a mountain range is depicted in striking contrast against a captivating backdrop. The ninja holding metal claws and the samurai holding a sleek japanese samurai holding a sleek katana are both prominent figures against this stunning visual masterpiece.
gloomy lark
I'm gonna try running it at a higher cfg and see if that helps with drawing the second person.
pallid path
these look stunning
gloomy lark
yeah other than the regional thing, the quality looks very aesthetic.
i'm running an x/y/z plot, from 4 cfg through 11.
I'll run the super prompt after this is done.
pallid path
thanks
gloomy lark
nope, all those cfg values, all at 1920x1080 native res (sometimes that helps). only 1 person ever. the extention just doesn't want to do it.
gloomy lark
thanks
this was with the super prompt thing you pasted.
rather good actually, that first one.
pallid path
oooh
if you want I could run a couple of prompts through it for you
but I can also show you how to run it offline, it's all on CPU and it's super fast
pallid path
it seems proteus is more coherent
gloomy lark
I think it gives the best combination of them all..
gloomy lark
it certainly is for that.
pallid path
im going to try merging it but in real time using comfyui
gloomy lark
the merge is amazing
here's the workflow if you want to make it for yourself. you just have to plop in the checkpoints and lora files.
gloomy lark
yep, amazed at how easy it was.
pallid path
honestly darkarts reminds me of illuminatidiffusion
I used to use that model all the time
back in the 2.1 days 👴
gloomy lark
yeah, that's the only part that makes me a little sad. I worry that some of the people who made/make amazing stuff for sdxl might not be around for sd3.
pallid path
Me when 2.1 had 1 step images without any special implementations
models almost done downloading 
pallid path
here's the workflow if you want to make it for yourself. you just have to plop i...
oh I have to have the one WITHOUT clip?
gloomy lark
wow
pallid path
is it cause of the ponyXL stuff?
gloomy lark
yeah he went nuts with ponyxl and everyone revolted.
so he quickly dropped it. version 2 might come out if he decided to release it. I think he wants to try and make money off an image bot for a while first.
gloomy lark
it's funny, midjourney mimic lora makes it look even better, but the prompt adherence goes right out the window.
actually this isn't bad with MM at 0.3 strength
pallid path
gloomy lark
I'm finding that dpm++ 2m karras 20 steps, followed by a latent upscale of area/1.5x and then another dpm++ 2m karras with 20 steps at 0.5 denoise cleans up any messed up swords etc.
pallid path
lol
gloomy lark
hah that's really good, although i have that love for the slightly pixarish look
pallid path
gloomy lark
@copper kraken
pallid path
kind of unrelated but still looks cool
gloomy lark
that also with the new model?
pallid path
yes
gloomy lark
inpainting?
because I could never get that even with regional prompting.
this however, is amazing. the previous generations from either of those models could never pull this off.
pallid path
because I could never get that even with regional prompting.
get what
copper kraken
This shit is espec good with the res momentumized custom sampler and a custom scheduler
gloomy lark
that with the merged model or the ip adapater stuff?
copper kraken
Oh nice, what'd you do to merge
Any specific weights etc or just straight up model merge simple?
gloomy lark
proteus-rundiffusion-without clip, dark arts, and 0.5 of cute 3d render
there's the merge json
copper kraken
Ohhh that's a comfy workflow, flew over my head ha
Def will check this out
Try the params in that nuke image I posted btw
That sequence is giving me great results with the res sampler
Includes a non tile upscale which you might like
gloomy lark
ok looks like i need another update
gloomy lark
do i need to download any new models compared to yesterday?
copper kraken
Nope
copper kraken
Oh oops you don't need that
Forgot I had junk off to the right
Go ahead and delete that entire island lol whoops
That's the new SDXL tile controlnet
gloomy lark
that is a wonderfully random image compared to the sources. 🙂
copper kraken
Awesome
Yeah this blows cascade clipvision outta the water
Now if only we actually had non shit sdxl CV...
It'd be really interesting to hit from both the unet and conditioning side
gloomy lark
that is some weird wild stuff
pallid path
nice
opal swan
gloomy lark
opal swan
gloomy lark
gloomy lark
copper kraken
Still need to try attn masks with this
opal swan
gloomy lark
Still need to try attn masks with this
hold on, I'm having a apocalyptic picture-off with trustory here.
opal swan
terse pewter
Need some advice, I want to turn this ugly looking image into prettier photo, the text can go away, but i want to prettify it, make it seem more 3D instead of being like a 2D drawing.
Is it possible to make it happen?
copper kraken
gloomy lark
is that my house when all 3 nodes are running?
copper kraken
is that my house when all 3 nodes are running?
No kidding
gloomy lark
reordered some stuff, so now it's just 30-40 seconds for 9 images.
copper kraken
That's awesome
Btw did you try forge?
And also. That exploding house.... Ipadapter is god
gloomy lark
I didn't. i should. 🙂 but I have to not ignore the fam so I should stop for today.
another day, another no sd3.
copper kraken
Exact same params with ipadapter weight = 0
This was the image I fed into it
Compare to the violence of this image
Incredible improvement
gloomy lark
hah it is definitely more splodey
copper kraken
Vastly
The first one is what I hate that it loves to make
An intact structure surrounded by fireballs
Destruction/flame/whatever tending to be adjacent to the image with no real connection
No splintering structure, charred paint adjacent to the flames etc
Then feeding one crappy low res images in via ipadapter and boom, literally
Prompt: house exploding
gloomy lark
it's not gonna make this kind of nonense though. 🙂
I think the regional prompter command is good, I just need to get forge going on the nodes so it's actually fast though.
copper kraken
Forge was a lot easier to get set up than I recall with a1111
Came with more shit ready
opal swan
copper kraken
tawdry current
oh ffs, i figured they would have fixed the VAE for the lightning models... in case someone doesn't know, the sdxl1.0+1.0vae is bugged and makes these horrible artifacts like on the right side of this image. they even reverted it back on the HF a while back so that it's sdxl1.0+0.9vae. i guess the lightning models are using the bad vae. the left side of the image is decoded with the 0.9vae and the right is with the baked in vae from the 8step lightning model
this is super zoomed in btw
though i guess to some, the "chromatic aberrations" are a feature and not a bug lol
tawdry current
yeah youre tellin me. the only reason why i noticed is because it will absolutely destroy your image after even a few rounds of inpainting. it will get progressively more and more blown out like a bad feedback loop
other than that, i actually really love the 8step lightning model(i dont do a ton of stuff with people people though, so i dont know how it stacks up against the base model at that)
terse pewter
Need some advice, I want to turn this ugly looking image into prettier photo, th...
great ! my message ignored
gloomy lark
great ! my message ignored
terse pewter
thank you, exactly what i needed ❤️ can you please do this for these 3 ones as well?
minor pecan
gloomy lark
another update to deep blue is out. https://civitai.com/models/128397/deep-blue-xl?modelVersionId=400487
copper kraken
thanks!
gloomy lark
i'm going through my merge stuff, i thought the cute3dcartoon lost prompt adherence and it just turns out it's because i was doing square aspect ratio. it's fine at 0.5
so now i'm testing merging with deep blue instead of proteus, but I think deep blue is too animated and pulls away from the style of dark arts too much, whereas proteus is complementary.
copper kraken
gotcha
yeah there is def some of that
there's some more advanced merge nodes too where you can pick out the different weights within the unets themselves or whatever it is
tbh, just baaarely have the weakest understanding of that stuff right now
but i've poked at it like a caveman a bit and noticed a pretty big difference depending on which parts you use
that was with loras but i believe it translates over
gloomy lark
It's really neat to play even with just the ratio numbers and strengths on the loras. to keep hitting render and see how things change and see which is better or not. some unexpected results, where I thought a mix would be awesome but it totally wasn't.
yeah, and using the clip from any of these loras just outright destroys the prompt adherence of a good checkpoint. I had no idea until i started playing with this.
copper kraken
but it might be possible to avoid pulling over the animated style and get more of how it structures the image... idk
gloomy lark
the midjourney mimic one basically just makes everything into a portrait with clip on. gets rid of anything in your prompt other than the single centered subject.
copper kraken
if that's the case, carosello is a top candidate imo to throw in
gloomy lark
with it off, things come back to life
gloomy lark
you can go pretty high with the model, with the clip off on the lora and still be in good shape.
copper kraken
this is what i'm referring to
something i think that would be really beneficial for us to wrap our heads around eventually
exactly how to manage this that is
but yeah i saw some really interesting effects with loras by dropping various blocks
i'm ready to get an ipadapter tattoo on my forehead
gloomy lark
haha yeah
concerning that block merging, I think that goes beyond what I'm willing to sink time into. I'll leave that to the people who make the major models. they're clearly very good at it.
copper kraken
well if you look at that first scheme
it appears dropping the first few blocks from deepblue might prevent the animated effect from carrying over too strongly
gloomy lark
I think datavoid with the new proteus model he has on his paid bot has really struck gold with what you're talking about here. luckily I got the version just before that at least. hopefully he'll release this new one at some point. it's a pretty big leap even over what we have publicly from him.
copper kraken
ahh i haven't seen it, is it a closed model now? no weights?
gloomy lark
so let me say this before I forget. I just ran my big giant at the gate on my channel. one before it is with 350, now with 40. hard to know what's what because of different seeds, but it looks like it's little more detail oriented. high detail anime is awesome.
copper kraken
i think i heard you or someone else talking about it the other night when i was in a daze from overwork + SD induced sleep deprivation
gloomy lark
yeah proteus-rundiffusionv2 and now v2.4 is closed... he says he needs to make money to be able to keep doing this.
gloomy lark
with sd3 right around the corner, doesn't seem practical.
copper kraken
yeah
gloomy lark
I'd pay to be able to download the model, but I'm not going to use his bot.
I think a good number of othr people have said the same thing.
copper kraken
yup i'm on that list
it's local or it's something i'll mess with for a few hours at best
gloomy lark
yeah, and i want my llms and all that other stuff. otherwise i'll just use midjourney
i want discord bot stuff AND comfy, not only one.
copper kraken
yeah exactly
all the tools are what makes it so addictive for me
https://civitai.com/models/143043/starlight-xl-animated don't know if you've seen this one, but this is another very good model
doesn't as hard toward animated style as deepblue
can do a lot of photography
gloomy lark
that was my go to forever. then i found proteus, and now i've found all these other new ones.
copper kraken
gotcha, cool
gloomy lark
i did a lot of A:B testing with that one, and proteus, even the early ones, were way more prompt adhereing than starlight, even though starlight looked the best. it makes me said that starlight hasn't been updated in a long time
copper kraken
agreed
lots of stuff that's just sitting around that sucks to know it's not getting updated...
gloomy lark
hey what was that new aetherverse or some such that came out recenlty? have you tried that?
copper kraken
downloading, thanks
my SSD doesn't say thank you..........
haha, i really need to just buy a 4TB
honestly, i'm procrastinating because i don't feel like having to take the 4090 out to stick that in
need to just do it
gloomy lark
well, do you really need all those? 🙂
copper kraken
yes >_>
lol
honestly part of the reason i hoard is because my connection isn't that great
the main issue is a lot of these models i know are good at doing one single thing
for example, thinkdiffusion i fire up for one thing only, nukes
gloomy lark
it's funny. i'm sitting here slowly increasing the cute3d render. and it goes from stone giant off on the side, which is especially prompt following, and it just gets more and more centered and portrait-ey
that first one is aetherverse, the new one, which is rather prompt following for what i put.
the higher deepblue i add, the more centered and portraity it gets.
but more unrealistic.
copper kraken
they def have their tendencies
that's where the block merging looks so interesting
def going to spend a bit more time looking into that once my schedule clears up a bit
the experts are def going to run circles around us on that probably forever, but it'd be really nice to be able to get custom models
stuff like... noticing you really like a certain type of composition a model is giving you, but not liking the style
so extracting the part that's leading to that composition and swapping in blocks that will guide it toward a diff style
gloomy lark
hah yep. I can't wait until sd3, where I feel like I can really describe a scene where something's actually happening, so what you just posted means something as part of something bigger
so you could have reactions from other characters to that, or where that thing is climbing in the window...
or crawling out of the television
gloomy lark
hah we need a mecha made out of exploding housing.
copper kraken
that's one place where dalle3 doesn't
gloomy lark
anthropomorphic robot made of exploding houses is crawling out of a television.
sigh.
gloomy lark
my channel
anthropomorphic robot made of exploding houses is crawling out of a television.
gloomy lark
i have playground, my proteusrundiffusioncute3d merge, and deepblue40 on slash image now on that bot, so it's one of those. 🙂
copper kraken
awesome
copper kraken
looks really good
gloomy lark
wow, this is from my merge.
that's actually got it all.
robot/ houses/ exploding/ tv.
this is that same combo but deepblue40 instead of dark art images
it does a really good job too
copper kraken
yeah that's really impressive
gloomy lark
my merge , merged with carosello is just nightmare fuel
copper kraken
🤪
gloomy lark
i like that one. a nice place to live
copper kraken
so many possibilities with the new idadapter
one i'm finding i like is subtracting the positive embeds, adding the negative embeds, using ease in-out for the first pass
then a img2latent2img2img 1.5x upscale
switching to style transfer for that one, and changing some of the params for the res momentumized sampler
zero clue what i'm doing but it's awesome
gloomy lark
zero clue what i'm doing but it's awesome
image of your endeavors
copper kraken
res momentumized is something absolutely no one is talking about afaik
you've used it, aimingfall has used it a bunch, but i don't know if anyone else has really dug into it?
it's a really special sampler
just throwiing my old images into this
those were the inputs
my cats compete to be the AI Art Cat of the hour
my PC is set up so all the heat blows behind my UW screen
best cat warming station ever
gloomy lark
it's really good, just takes time. I think it gives SDE like results but faster, but its all still slower than 2 passes of dpmpp_2m. alone, it messes things up. second pass of 0.5 denoise, and that crappy fast sampler tuns amazing.
hah
copper kraken
i think they're a lot more interesting results
the compositions are different from any other sampler
gloomy lark
those are neat
gloomy lark
haha
50/50 my merge with deepblue40, another 0.5 of cute3d, and 0.8 of midjourney mimic. it's style overload
gloomy lark
merging aetherverse and deepblue with 0.6 towards aether is actually giving some off angle shots for once. and is actually a stone giant, not just a giant, and actually looking down on little people.
this might be a good one for the bot instead of plain deepblue.
copper kraken
wow masking is really nice with ipadapter
you can use masks to mask out stuff you don't want to incorporate now
as well as still using attention masks to specify where things should be generated in the image
copper kraken
oh nice
gloomy lark
this was a good prompt. might work for your stuff
astronaut in a faded space suit covered in samurai armor that is adorned with noh era trinkets and meija era pieces. the helmet is covered in sleek shiney black accents with various antennas and cameras.
lots of detail that your stuff could riff off of
copper kraken
lots of detail that your stuff could riff off of
used modelmergeblocknumber with juggernaut and carosello which obv is pretty much only illustration outputs
if you knock out input block 1 and output block 1 you don't get that style anymore, you get photographic
so swap in block 1 from juggernaut and you get photographic style
prompt: a shark car
juggernaut
carosello
carosello with blocks 1 from juggernaut
carosello
juggernaut9
juggernaut blocks 1-3 input & output, rest carosello
carosello, then carosello with blocks 1&2 from juggernaut
indigo wagon
这里可以生成图像吗
glacial flame
I'm excited to try out any textloras/models with SD3, cause SDXL is pretty decent already with certain methodology
drifting hinge
这里可以生成图像吗
drifting hinge
5
honest edge
1
neat sand
one dog icon
noble shoal
one dog icon
Here is the image you requested.
gloomy lark
signal shale
opal swan
opal swan
inner kelp
gloomy lark
Wow very neat
burnt shadow
lots of detail that your stuff could riff off of
pure base model with a slight influence of 3 input images
not as detailed/clean
gloomy lark
pure base model with a slight influence of 3 input images
Very cool. I used to work for an art gallery that had a Japanese noh mask exhibit. There's probably more cool stuff with that keyword
burnt shadow
Very cool. I used to work for an art gallery that had a Japanese noh mask exhibi...
yes, can imagine
burnt shadow
nice one👍
rich olive
burnt shadow
cloud canyon
Humanoid robot with clear structure
gloomy lark
Humanoid robot with clear structure
compact spade
mossy rampart
Create a photorealistic image showcasing the 'Fire Guardian' smart fire safety system, based on the AIOT platform. Include a variety of fire sensing devices within a small venue setting like a retail store or office. The devices should be depicted in both close-up, to emphasize their design and technological features, and in use within the setting, illustrating their integration into a comprehensive monitoring and alert system. The composition should also hint at a user interface or data visualization that represents real-time monitoring and analysis. The lighting should be clear and bright, emphasizing the system's reliability and the safety it provides. Highlight the system's ability to safeguard life and property through advanced technology.
late abyss
pure base model with a slight influence of 3 input images
upbeat radish
cats
indigo wagon
生成一张树叶
glass forge
https://tenor.com/eSqTFJo3pni.gif
Is SD3 out yet?
opal swan
opal swan
ashen otter
Is SD3 out yet?
that is a gif of a dog
pallid path
^
neon junco
create an image of a cat playing a recorder
little whale
Beautiful pixel art of a Wizard with hovering text 'Achievement unlocked: Diffusion models can spell now
steep turtle
little whale
cat
minor pecan
rich olive
noble shoal
rich olive
shy kelp
Two huge monsters fighting each other brutally
steep turtle
stray warren
rich olive
gloomy lark
Two huge monsters fighting each other brutally
gloomy lark
Two huge monsters fighting each other brutally
gloomy lark
Two huge monsters fighting each other brutally
ashen otter
Two huge monsters fighting each other brutally
No
noble shoal
burnt shadow
gloomy lark
burnt shadow
pallid path
lol
glass forge
that is a gif of a dog
Yeah but it looks like a dog looking for soemthing. So yeah... funny
noble shoal
pseudo swallow
Image
burnt shadow
safe stump
Lots of colourful balloons in the sky
gloomy lark
copper kraken
Lots of colourful balloons in the sky
Here is the image you requested.
upbeat summit
burnt shadow
gloomy lark
glass forge
Politics in a nutshell.
faint trellis
美女
eternal fog
copper kraken
rose smelt
gloomy lark
opal swan
ashen otter
mercevolvopel
glass forge
Whay can;t I use Playground 2.5 with Fooocus? Isn;t it a sdxl model? It's sposed to work yes?
tawdry current
@glass forge It's an sdxl architecture, yes, but it uses a different noise scheduling method. EDM I think. Comfy has since it came out. Foooocus hasn't implemented it yet
cloud lily
cat
prime bison
upbeat summit
who's gonna clean up that mess?
gloomy lark
Here's the image you requested
Wow
opal swan
copper kraken
finally sitting down to do some methodical tests with the new ipadapter since we have no documentation still
gloomy lark
Prompt was : Exploded cat (by nychos-(by Van Gogh:1.5))
Wow that's a really good prompt
copper kraken
holy shit
crazy
results of round one testing: all "Combine Embeds" methods are commutative except subtract.
rich olive
Wow that's a really good prompt
Wow really good
rich olive
Wow that's a really good prompt
Try this one quickly : Exploded cat (by nychos-(by Van Gogh:2))
copper kraken
clipvision doesn't seem to have any effect on the output so i'm not sure why that input exists... hm
rich olive
(Shark by Zdzislaw beksinski:0.5), (shark by nychos:0.5)
copper kraken
not sure what the neg embeddings do...
with concat, the answer seems to be nothing
rich olive
character (by Zdzislaw Beksinski-by Esao Andrews)
humble cape
Made with sdxl turbo
pallid path
I think I've seen your post on reddit
gloomy lark
Made with sdxl turbo
humble cape
I think I've seen your post on reddit
Yess thats me
humble cape
You applied image to image and image to video ?
pallid path
nice images
rich olive
Reminds me of liminal spaces, i love em
trim orbit
i love the concept of liminility and feel it's been hijacked by backrooms memes. The idea of thresh holds and gates entering new spaces. You've probably felt it when you enter a room and something switches in your mind. not only a new physical space but a new head space at all. even goes back to ancient times. Towns would have gates. Chinese would have lion sculptures guarding their gates. very cool stuff. Gateways.
trim orbit
TIL that theres a specific LCM for animatediff. very neat.
humble cape
Reminds me of liminal spaces, i love em
Yes, I generated this image with sdxl in order to make a liminal space
trim orbit
heyyyy i got it working in automatic1111 animatediff extension. 4 step animations neato!
humble cape
My best is this one
rich olive
Wow those are amazing
trim orbit
theres a lot of panic about the coming eclipse that'll hit eastern north america. i think it'll line up with the devil's comet being in the sky too. oOoOooOoooo spooky
solar eclipse instead of a lunar but 'm still reminded of it
humble cape
Yess thats terrifyng ambiance
My technique is to start my prompt with “a photo taken in” and use the key words “urban horizon”
humble cape
Amazing it's surreal
copper kraken
These are my kind of liminal spaces
reminds me of control
humble cape
@rich olive Try the prompt: "A photo with a lot of image noise and visual artifacts taken from the top of a building, night sky, clouds, moon, urban horizon, dark nothingness"
With sdxl base model only
trim orbit
having some difficulty with the i2v model in a1111 . looks like that's not supported as well
humble cape
stray warren
Base SDXL is powerful
humble cape
The render is amazing
rich olive
ashen otter
The render is amazing
Saving
humble cape
With better hands
tawdry current
In the future, rather than trying to wrangle seeds and RNG, just Marigold a depth map out of your own hand, takes like 60 seconds
opal swan
gloomy lark
prompt: a cat watching tv. superprompt: a cat watching tv, 8k, masterpiece, in the style of greg rutkowski
gloomy lark
honestly it's all about length. giant prompts don't matter all that much if they can't also restrict themselves to stay inside the 75 token limit. i know the really big models have that issue.
hopefully they put in logic in the node to force the issue.
pallid path
yeah
pallid path
prompt: a cat watching tv. superprompt: a cat watching tv, 8k, masterpiece, i...
It's quite dumb but it's fun
a fluffy orange cat sits on a windowsill, its paw resting on the television. Its t-shirt is folded neatly around its body as it watches the latest scene of a bustling city skyline. The cat's eyes are fixed intently on a tv set in front of it, and its tail wags excitedly as it watches the tv with curiosity.
gloomy lark
It's quite dumb but it's fun
a fluffy orange cat sits on a windowsill, its paw r...
certainly gets the job done.
yeah i'd say that works fine.
and that prompt was 74 out of 75 tokens. 🙂
pallid path
Yeah I always trimmed my superprompt prompts to 75
But with SD3 we don't have to trim anything 😌
copper kraken
gloomy lark
I hope so. I heard lykon say sd3 is 512 tokens, and then someone else said it's still actually 75 and deals with them in the same way as sdxl where they all just get averaged. I have t read the paper on it
pallid path
Well if T5 applies conditioning to the image separately then yeah it's 512 Tokens, even if there might be a fade off towards the end of that number when it comes to efficiency
copper kraken
pallid path
I mean even if it only doubles or triples the context length we'd be happy
copper kraken
I mean even if it only doubles or triples the context length we'd be happy
especially if it actually listens
there's already not much point in hitting 75 now
you hit 15 and it's often already ignoring stuff
pallid path
Most of my SD3 test prompts are hovering about ~50 to ~120 tokens
idk how to count prompts though so I'm just doing gpt2 tokenization
copper kraken
well, this is an interesting new interpretation of "supermassive black hole in a city park"
pallid path
This apparently(?) just snuggly fits into the prompt limit, no idea though.
Old analog film photo of a giant room with a big monitor in the center. The monitor has a giant head in it with an ominous stare whilst speaking. There are massive rows of chairs with people in them, mindlessly watching the giant head. There is a caption for what the head is speaking on the bottom: "No Fun Allowed"
And the rest are just longer and more complicatd
(It's a 1984 meme described so that any intelligent image generator like Ideogram can do it)
Ideogram did it with flying colours, so I suppose SD3 will do just fine with this much
Festivalman could easily do this with a simple regional prompt
(Big giant monitor with head in it on TOP grid and rows of people watching on the BOTTOM grid)
interesting, SDXL almost did it actually
with no regional prompting
with regional prompting
gloomy lark
That's very good
copper kraken
ipadapter shines even more with diff diff
really improves consistency if that's the goal
gloomy lark
Festivalman could easily do this with a simple regional prompt
I think this is one of those times where regional doesn't do much for you. Base works well enough, other than getting that text subtitle on the screen which sd3 would probably do really well with. I tried throwing it at my regional script, but I think it understands big monitor as just one that's close to you.
copper kraken
maybe something like projector screen instead of monitor?
pallid path
these are great
gloomy lark
Hah, zoom out. Zoom out more.
pallid path
oh no nipple!!!! 🤯
gloomy lark
these are great
An old, grainy analog film photo of a colossal screen on a front wall, in the style of a dystopian movie still ADDCOL The ominous, enormous head of a man projected on the screen, in the style of a dystopian movie still ADDCOL The giant head staring straight ahead, its mouth moving as it speaks, in the style of a dystopian movie still ADDROW Countless rows of chairs filled with people facing the screen, in the style of a dystopian movie still ADDCOL The people sitting motionless, their eyes fixed on the giant head, in the style of a dystopian movie still ADDCOL A caption at the bottom of the screen reading 'No Fun Allowed', in the style of a dystopian movie still
I changed up a bunch of words like "in the front of the room"
pallid path
that is SO epic
copper kraken
oh no nipple!!!! 🤯
got timed out for 12 hours a few weeks ago cuz i posted one on my phone and didn't notice you could see nips poking through the black dress (some bat freak girl i made in response to someone who was spamming prompts)
pallid path
💀
copper kraken
just the shape alone was enough, apparently
pallid path
makes sense, you are a massive criminal
copper kraken
lol
pallid path
copper kraken
i don't even know which i like better
this is part of the fun with this diff dif workflow
you get five shots to have an interesting output with each run
darkarts works really, really well with diff diff
frigid holly
opal swan
opal swan
opal swan
humble cape
agile sage
deer
humble cape
he want a huge
wet nacelle
wet nacelle
wet nacelle
pallid path
bro im chair 💀
sleek sun
Emily a beautiful woman 30 years old, long brown hair, light eyes, frontal portrait
wet nacelle
minor pecan
have you heard the good news
rose smelt
placid dust
stray warren
I asked for a vampire in red... 😆
upbeat summit
I asked for a vampire in red... 😆
these head wings keep haunting my images
gloomy lark
stone fossil
gloomy lark
glass forge
giger
copper kraken
_Emily a beautiful woman 30 years old, long brown hair, light eyes, frontal port...
Here is the image you requested.
rustic stone
A polaroid photograph of modern day jesus christ hunting for easter eggs while holding a basket radiates golden light
gloomy lark
A polaroid photograph of modern day jesus christ hunting for easter eggs while h...
Here is the image you requested @copper kraken
copper kraken
A polaroid photograph of modern day jesus christ hunting for easter eggs while h...
Here is the image you requested.
gloomy lark
hahaha so good
copper kraken
that regional prompt feature is great
gloomy lark
yeah, even with the regional stuff doesn't quit hit, the regular often does with the same prompt.
@copper kraken For the ones that didn't, I tried running it manually on sd through lots of seeds and about 1 in 5 was perfect, and the rest were just a mess. Not really sure if that's fixable or not. we obviously don't want to wait for lots of images to generate. (or at least I don't)
copper kraken
yeah def don't want to wait forever
i usually do small batches (if any) cuz it's often better to change parameters than to get extra dice rolls
but, that is also without the benefits of llm enhanced prompts
i'm starting to think i need to post this workflow on civitai
i'm enjoying how it's producing such wildly chaotic images while becoming increasingly chaotic itself
gloomy lark
definitely.
gloomy lark
hah I want to see the comment section on that.
copper kraken
i need to somehow ensure that no nodes are just sitting there
must ensure all nodes are essential
the heart of the action
upbeat summit
the heart of the action
you don't need to generate images, that's already art 
upbeat summit
very cool!
copper kraken
dare you to try the embedded workflow 😄
with a diff prompt, but everything else the same
upbeat summit
with a diff prompt, but everything else the same
interesting world, lots of details to explore
copper kraken
tawdry current
Rookie numbers compared to some of the blueprints I've seen in UE4 and 5 lol...
Where people don't know the sacred art of creating functions or macros
opal swan
glass forge
This is scary and impressive but does it really yield better results than Foooocus?
copper kraken
This is scary and impressive but does it really yield better results than Fooooc...
Oh my god yes
Wipes the floor with it
wet nacelle
opal swan
opal swan
wet nacelle
lusty wolf
rich olive
rich olive
median matrix
minor pecan
thats me in the chair
rose smelt
unkempt stone
APPLE
opal swan
pallid path
@uncut steeple @smoky patrol Scams ^ (leads to an incorrect link)
smoky patrol
<@170273907889143809> <@304180422357549056> Scams ^ (leads to an incorrect link)
Thank you!
gloomy lark
So what tool are you using for your images? I assume it's not just txt2img, but some kind of regional painting or some such? Clearly it's something I need to start using. 🙂
rotund chasm
Pls where are u guys generating these images from
I find it difficult
glad grove
Pls where are u guys generating these images from
I find it difficult
from our GPUs
rotund chasm
from our GPUs
Okay
But it's not in this server
glad grove
no
humble cape
rich olive
wet nacelle
not gore. Just a bit much
burnt shadow
So what tool are you using for your images? I assume it's not just txt2img, but ...
It´s simply txt2img along SDXL´s base model and in this case 1 step Refiner (which I usually refrain from using) on top 🙂
gloomy lark
It´s simply txt2img along SDXL´s base model and in this case 1 step Refiner (wh...
Wow ok neat
burnt shadow
Wow ok neat
playing with img2img as well, mostly for variations of an image or simply as a creative tool, in this case it's prompting only though. Haven't yet touched IPAdapter and regional yet. Next on list 🙂
gloomy lark
I'm impressed at what you're able to get on screen straight. I've been working on regional prompt stuff a lot.
burnt shadow
I'm impressed at what you're able to get on screen straight. I've been working o...
Thank you 🙂 Regional prompting I heard about not that long ago, then I just recently learned it´s about regions of the image while before I thought of it potentially being naming actual regions of the world like countries for region-specific outcomes 😄
gloomy lark
Thank you 🙂 Regional prompting I heard about not that long ago, then I just rec...
lol that would be pretty funny. Yeah check out the rpg diffusion extension for a1111 that works with the regional prompter extension. Does amazing things when it gets it right.
humble cape
empty plaza
It's scary, it's great
burnt shadow
lol that would be pretty funny. Yeah check out the rpg diffusion extension for a...
Having just started using Comfy more intensely and Easy Diffusion otherwise, not too much of a fan of A1111 in terms of interface, I feel like I´m gonna experiment with the approach of how I think it works for starters, thank you anyway 🙂
gloomy lark
Having just started using Comfy more intensely and Easy Diffusion otherwise, not...
Yeah I only use a1111 because nothing beats its regional prompter extension for text to image ability. Clownshark does it via images in comfy, but the text method allows for llms to do cool stuff for you
burnt shadow
Yeah I only use a1111 because nothing beats its regional prompter extension for ...
yes, can imagine it being easier, then I like the manual approach experiment for its nature anyway 🙂
rustic garnet
I mean, it's super easy implementing that in comfyui same way. I just think using images is more precise most of the time
copper kraken
Yeah I only use a1111 because nothing beats its regional prompter extension for ...
it's definitely more convenient than the one in comfyui
can't say i've explored either in a great deal of depth - afaik, the comfyui one is more powerful in the end, but you obv either need a workflow thrown your way or to spend a bunch of time making one, and comfyui's api is a lot less friendly from what i'm hearing
certainly more work to tweak than ADDCOL etc
comfyui could really use a node that can efficiently generate mask grids etc without it turning into a big chunk of the workflow
and a regional prompter node that can actually take those as inputs or something like that
burnt shadow
burnt shadow
native knot
Holy shit...the focal distance on this one...this looks like it's legitimately floating in front of me.
Oddly enough, the thing I find most disturbing in this image is the fact that he has a button-up collar on a shirt with no buttons.
copper kraken
it's always the subtle things that make a creepy scene so unsettling
native knot
Yeah...exactly. Not phased by the makeup or anything else...it was the shirt.
gloomy lark
there's probably a subreddit for that, fashion horror
glass forge
Darth Haus?
native knot
Darth Haus?
He's got this crazy knack for figuring out exactly what totally rare force choke you need.
peak dove
I updated my InstantID, and produced this "Splendour" SDXL Turbo/Pearl Mountain Collage Maker
rustic garnet
comfyui could really use a node that can efficiently generate mask grids etc wit...
dunno, I don't think that these things turn out as big workflows. It's just that the workflows people publish are unnecessary huge, because they add too many features to them.
I would say ComfyUI needs more modularity, a way to easily share and save parts of a workflow
native knot
copper kraken
dunno, I don't think that these things turn out as big workflows. It's just that...
That would be really handy too
Save and share functions/group nodes
noble shoal
You could just build group node components and share them 🤔
copper kraken
I guess I mean something through the manager but yeah
noble shoal
Components you can just re-use in every workflow
copper kraken
Yeah
Not a big deal to me anywaym
Im not afraid to go wild making workflows that imitate my art lol
marsh zodiac
@river stratus Goku in his super sayen form
gloomy lark
<@1047950355822022736> Goku in his super sayen form
minor pecan
Goku became tentacled
meager canopy
Save and share functions/group nodes
Doesn't this already exist?
copper kraken
kinda, guess i had some vaguely diff in mind at the time
noble shoal
Doesn't this already exist?
With comfyui manager you can pack your group nodes. I guess you could share those packs if you want.
meager canopy
With comfyui manager you can pack your group nodes. I guess you could share thos...
Yes, there's a Component Builder that allows you to do this.
copper kraken
yeah, i guess i meant more share them as separate subworkflows that can be neatly opened and closed in a separate window without ungrouping the entire thing into a mess all over the current workflow
high spear
amber remnant
The on-site background wall design of the headphone exhibition, with the headphone products as the main focus, and the overall yellow color tone.
copper kraken
The on-site background wall design of the headphone exhibition, with the headpho...
Here is the image you requested.
noble shoal
The on-site background wall design of the headphone exhibition, with the headpho...
Here is the image you requested
high spear
gloomy lark
native knot
gloomy lark
composition idapter
copper kraken
composition idapter
kiss a week of your life goodbye now that you've also discovered how insane that update is lol
gloomy lark
kiss a week of your life goodbye now that you've also discovered how insane that...
i have no idea what you're talking about. i can stop whenever i want.
gloomy lark
will it do SD3 or is it still stubborn about a number following a letter?
copper kraken
will it do SD3 or is it still stubborn about a number following a letter?
still stubborn... lost most of my men but i'm winning the war
gloomy lark
oooh that's cool
copper kraken
can get it to be a bit more subtle by skipping the first few steps with the controlnet, then hammering it hard after
the qrcodemonster with sdxl is not anywhere near as good as with sd15... with sd15, i wouldn't need ipadapter to help
this does come with the advantage though of some control over what goes where
native knot
Still waiting on the next version of SDXL:
SEXL
gloomy lark
@copper kraken
i don't know if it's ipadapter or just this art model, but it does text really well (click the second one)
copper kraken
that was the source image on the left?
gloomy lark
yeah moofi posted that one
that was just "movie poster stability ai" obviously it's using what's after the movie poster as part of the image basis
copper kraken
think you might've gotten lucky
like i did with my MEAT BAG image a few days ago
it's generally not spitting out any text with "CLOWNSHARK"
did get this one though hm
native knot
wet nacelle
Am I also to assume that your pfp is also stable?
Can you share the details on how you gend that?
gloomy lark
ok, best shark feet ever
wet nacelle
wet nacelle
ok, best shark feet ever
true
native knot
