#🏞|general-with-images
1 messages · Page 132 of 1
Hey I’m looking to use AI to make future chapters of my comic book. I’m looking for an AI artist to collaborate with on this project as a hobby. Please DM if interested
discord user
I had trouble making this realistic.
"Humans are pretty stupid. But they got milk!"
cat
Here is your picture
cat being abducted by aliens
@late sorrel initial test (just wrappers for now)
LaVi looks best to me ...
A peaceful doodle of a meditating monk, in simple robes, atop a mountain at sunrise, doodle style, sunrise color palette, expressive, free-handed, line art, basic shapes --ar 3:2 --v 6.0
This is not the image you requested.
@drifting elk the mad lad haha, you already implemented something, nice, i will check it out later, busy now, thanks a lot for the work ❤️
how does it compare to ELLA so far?
a bird
no good news in that regard, overall it seems much worse :/
Some tests with ELLA
ELLA and LaVi
Thanks!
where my free rtx 5090 what i requested
🙁
Here is the Free RTX 5090 you requested.
No cooler needed! Cool! 😄
nice without cooler and extra power)
those hessians never knew what was coming for them
ultramarine
He was our most capable president.
Ella + Lavi. Why not? 😬
kind of disapointing that none of them even go as far as to do the helmet
SVD is my friend today 🙂
generating some babes I see
See you space cowboy...

I want to see this it uses this fan, is there anyone who can help me?
/imgea
I made something awesome I want to share.
Wait, wrong image, dang it
There, that's the right one.
Hi! Tell me, please, can such backgrounds for photos be created in stable diffusion? If so, how do I create similar images?
Thanks!!!!
but is not ideal
now make them easter eggs
from below,incredibly_absurdres,ray tracing,{best quality}, {{masterpiece}}, {highres}, original, extremely detailed 8K wallpaper, 1girl, {an extremely delicate and beautiful},,ray tracing,{best quality}, {{masterpiece}}, {highres}, original, extremely detailed 8K wallpaper, {an extremely delicate and beautiful},,incredibly_absurdres,colorful,intricate detail,artbook,,{best quality}, {{masterpiece}}, {highres}, original, extremely detailed 8K wallpaper, {an extremely delicate and beautiful},,full body,wide shot,electricity,aurora,starry sky,clusters of stars,starry sky,glinting stars,,starry sky,clusters of stars,starry sky,glinting stars,,raindrop,shooting star,colorful startrails,highres,photorealistic,realistic,goddess,floating,armor,unbuttoned clothes,
story from book?)
yes
Is it possible to create lora to make such stickers in SD? (Same style and mc)
Im so glad newer models are getting rid of all those bloated prompts, people will still use them because they dont know better, but it wont be necessary for a good output 
Senegalese black man, 40 years old, profile picture, real photo
Need a coffee? Not really creative I know ...
offering
Had the pleassure of working alongside @maxcoopermax on his latest project, called “Seme”.
He asked me to experiment around the concept of renaissance era, giving life to a set of AI intervened custom-recordings [ft. @chinalabaig], experimental oscilloscopes [tbr, hopefully soon], and @touchdesigner systems, for his shows last week in collabor...
186
Hello, somebody can help me with this prompt:
A photograph of a creature that has a black body like that of a penguin, has a wide yellow beak like that of a duck. It has six eyes in total, 4 spider-like and 2 snail-like. It has a robust carapace like that of a turtle and wet and slippery legs like those of a frog, it is friendly, its habitat is wet and swampy places, such as hollow logs near rivers, light enters through the leaves of the giant trees that are around it.
I know its a little weird.
The more specific your prompt is, the less likely you'll get what you want. That prompt is very specific with lots of details, so don't set your expectations high.
should i change it?
Your expectations? Yes.
@languid pebble I don't see civithelper, and I grabbing the browser instead? I got forge installed
Congrats! .... let me check ...
regular or browser+? o-o I don't know which is better, browser+ has more stars
Huh, it's not popping up, do I need to go grab it from git instead of the availaible menu?
Wait
You can try to install from git, yes
This prompt is better?
Can you draw a friendly creature that's black like a penguin, with a wide yellow beak like a duck, six eyes (4 like a spider's, 2 like a snail's), a strong shell like a turtle, and wet, slippery legs like a frog? It lives in wet places near rivers, like hollow logs, with sunlight coming through trees.
Found it while I was looking to add pictures to the LORAs to see wich ones I could use ...
Another question, can you get the 3D model tab like in automatic1111? So I can just on the fly make a wireframe that the image generation will follow?
I saw someone else have it working, but that was with automatic1111, and it refused to work for me, maybe it will on forge
Never tried that one ...
Fair. What's the best way to get a certain pose you want? Do you just find a similar image?
XD That is horrifying
ControlNet ... they say openpose is not so 100% put you could even try to paint it ...
Depth map ... you can combine different dontrolnets ...
Lots of thanks XD
XDD
There has been an extension to create openposes pretty simple ...
nice hybrid
I just copied your prompt so you can see what you would get ... 🙂
Oho? What is the extension? Sorry to bother you so much but I am very interested in this topic O^O
ChvitaiHelper is an extension
No no, I meant the extension to create openposes
I got civithelper like you recommended
I don't remember the name ...
it similar as i imagine so im thankful
The last step I need to do is transfer all my prompts from SD to forge
It's gonna suck
Are your prompts kept in a file I can just copy paste?
e-e I don't wanna do it manually
You can copy them into a text file with a blank line in between ...
I should do that anyway, so if I want to switch back and forth I can do it easier, ugh, I will do that now, brb
and use Prompt from Text ...
What? I don't even know how to use script, and a bunch of stuff appeared that I have never seen on forge XD
I am lost again
Bottom of the TXT2IMG Page ... just use dropdown and choose prompt from textfile or box
My new Avatar 😄
thank you man, its strangely perfect
Went from this, to this
A friend wanted it brighter
I love the stomach, but some of the features aren't as cool
You will become better ...
😃
Good time to post something on X 😄
X?)
Twitter ...
Found the Script thingy?
No, I had to go eat lunch, and have been gone since.
I am just now back to start fooling around with stuff.
Can't be easier than I told ya ... bottom of txt2img and select from the menu 🙂
:O anyway
I wanna know how to make this prompt better. A friend wanted me to make this image for her, but I think I didn't do the prompt very well.
Play around with a fix seed ....
Oh, my prompt is apparently not allowed, maybe because of a certain part.
oof
((best quality)), ((masterpiece)), (detailed), goth girl, short black hair, red highlights, glaring, tattoos, choker, mouth open, standing in front of mirror, holding phone, selfie, thick brush, HD anime wallpaper, 8k resolution?
Okay, so it was that part.
1.5 or XL prompt?
It's the model Version. 1.5 or XL ...
I think it's a 1.5 model ... prompts sounds like for a 1.5 model. So it's OK ... I turned to work with XL ... but others still prefer 1.5
I don't know if I can run XL, I haven't tried on forge
It worked okayish on Automatic1111
As long as you train I would stay with 1.5 OR XL ... cause they have different workflows and prompting ...
With XL I stopped using the start of your prompt and added LORAs
If you use the civitAI extension to download Model information you can get information about checkpoints/models and loras here.
XL is trained to work on 1024*1024 ... if you generate to small results will be bad ...
Oho
Than you should update your information in the civitai Tab
XL you can do 1200800 as well but always around 10241024
1200*800
It takes forever to generate 1024x1024, so I guess I will be sticking to 1.5
You can upscale the good ones later
XL was working fine in ComfyUI though
So maybe I will use it in that and just do most of my work in forge
I don't think you need to print your pics in 90*60 cm ... so it should work for you
Haha LeoB! Thnak you 😄
A.I. calculating video ... can't tell you ...
sing "Baby don't bite me ... don't bite me ... no more"
These hands are so good omg
I gotta throw it through ultimate sd upscale but it's so good
I have no idea why this is happening. I'm using sd_xl_base_1.0 (I can't seem to find 1.5 which I used to have). I asked for a woman walking along a lake shore, and it's giving me anime for everything.
I used it yesterday, and everything worked fine.
Really digging that shell man
@shell sleet How did you get such amazing hands? No matter the prompt I write I get ugly hands.
Which model have you selected?
Ok, I seem to have fixed the anime problem by changing to a Dreamshaper model. STILL have no idea what happened to my base SDXL models.
They should be still in the models/stable-diffusion folder if you didn't deleted them
I didn't delete them..at least not knowingly. They are not in the directory. I've been looking for where I can download them, but I'm not having much luck.
You can download models on civitai.com
Yeah, I would have thought so, but I'm not finding them. I DID however find links on the Stable Diffusion website, so I'm reacquiring them now.
I used openpose with the hands included
I want to do that, but I can't figure out how to get open pose tab, like I've seen in videos.
I want to be able to make a pose from scratch, and then use that to generate an image.
I'm sure it's asking for too much, but I wish I had like a 3D model I could use, to get the poses just right. Lmao
https://openposeai.com can help you make an open pose from scratch!
I just downloaded an extension that adds that into forge :O
It's being frustrating to use, but I can make exactly what pose I want with some work. XD
Oooh I was gonna point you at the one for a1111, but I didn't know what environment you used
Bruh. How do you expect it to provide guidance for your image if you start AND stop the guidance on the same step?
Save the preprocessor image, load it into the first slot, set it to open mose and clear the preprocessor option
Pose** not mose jfc
When the prompt puts dark lightning instead of dark lighting
Maybe you can help me with what I wanna do
I wanna just upscale the image, but not change much on it, like just add more detail. Is there a way to do that?
hypertiling with low denoise 🙂
You are speaking a foreign language to me good sir, but I shall try
That?
I just installed forge, so a lot of these options are foreign to me
im awful at teaching, but theres a lot of good videos on youtube. I, and im sure anyone else here could recommend several. But, ill give it a shot!
assuming you don't want to do any outpainting and just want to upscale and add some details to an image you already have? or did you mean on the initial text to image generation?
Oh yall using forge, what's forge about?
forge runs way better than automatic1111
I am getting way faster results, and better results, and I am on 4gbvram
forge is a1111 in the way that:
a car that has had the seats, ac, power windows, stereo, and any other extra weight removed for perfomance is the same as a normal car
you can get even better results if you tweak some things in settings. higher performance but less quality. but you can always make up for that with the magic of tiled upscaling
Mmmhhhhhhhh where's the link for forge
I need that performance since a1111 has been an ass to me
I can produce decent quality(for me at least) with the current settings
be warned: controlnet is integrated in forge and so it may be a slight pain in the ass to get some things to work correctly that are taken for granted in a1111
I been able to use controlnet, but only tried Openpose, the only thing I have tried
Yeah openpose is the easiest thing for me lmao
just familiarize yourself with where to put preprocessors and models in your install directory (and where to get those from) and you will be fine for 90% of stuff
I am now using posemyart to find poses I want, and then just adjusting the littler details.
Oh yeah, forge doesn't have the "Put this thing here" files in each place, but if you know where they go, it's no problem
The weirdest thing was finding the right folder for the controlnet models, but I found it
ilke what?
I clicked on a video for hypertile integrated and he was speaking french
in my experience, primarily extensions with controlnet as a dependency. not difficult to fix but requires some familiarity
lmao yeah, sometimes they throw a different language at you
Dive into the world of high-resolution digital art as we embark on a five-step journey to transform the ordinary into extraordinary 4K and 8K visual masterpieces. From text to image prompt engineering to advanced inpainting techniques, and finally to a groundbreaking upscale process, this tutorial is packed with invaluable tips, practical insigh...
Thank you
I probably advertise this guys videos more than he does... and that one doesnt use the integrated hypertile extension, but he explains what is used and the vid desc has links. should be easy to follow!
as a fun side note, if you mess up hard enough and barely listen to the instructions, you can abuse tiling to make some cool photo mosaics
XD
yup same. i made a person made out of people. when i realized what was possible i made a gun made out of guns lmao
I always get pruned because storage is a valuable resource for me
pruned is probably fine to be honest.
it should still get you results
I don't know how to use styles, so I am gonna hope I can find where to put it, never used it except for the ones baked into forge
When I saw the Legend of Zelda style, I about lost it. I fricking love Legend of Zelda, and being able to easily make that style is amazing.
I applied it to the same prompt I showed above, and suddenly she had super cute elf ears
aztec watch?
you need more tiles!!! I have to screenshot that to even upload it lol
oh dear god lmao
im planning to further upscale my people tree for my horror image posts on instagram. uh, too bad i cant find anywhere to post pictures that are 12kx12k
yeah i uh. :) I can't do that
my laptop will blow up
100%
we rockin a 1660ti in this laptop
it will explode
I need halp
The internet is confusing me
Do I put the style I just downloaded in the same place I put loras?
Or?
styles, might refer specifically to styles extension or to just adding a style into prompt (which is what that extension, or embeddings do anyway for the most part) so if you want say something in the style of an oil painting, or in the style of "whatever" video game, you can usually just prompt it in. shouldnt have to use loras often or embeddings, just makes it easier. Hell, you can even use controlnet instead of those if prompting doesnt work
what style?
Are you allowed to post links?
i assume as long as they aint nsfw
That
that's a lora
:O Confusing me
I see that now
sure you can. enable never OOM on forge, upscale 1.5x at a time with shittons of tiles, wash rinse repeat until you have an ungodly 200mb image that you cant do anything with
im not rubbing it in but yeah style is just a tag
Im not! I'd get confused too
I was really confused when i first started using civitai lmao
if you STILL cant do it without melting your graphics card. Split the image into 4, upscale each individually and then stitch it together. uh. There isnt any point to doing it other than that its funny though.
alright time to use forge real quick since i know how to use git
forge feels much smoother, and all the options you would have to dig around for are in the front
You don't have to go all the way into the settings to change VAE
I love it
^yep and if you want more options you can add in the comfy extension if you want to dabble with workflows
Oh?
Does it change the UI to the comfy UI?
That stuff seems complicated, but I like how it looks
naw, hold on ill re enable it on mine and screenshot
Heck yeah
without comfy:
Oh, yours looks different than mine
just different theme, extensions, and tabs instead of dropdowns
I want mine to look as sleek as yours
_>
I don't like the bright orange
CivitAI browser+, does that make it so you have civit baked into the application instead of needing to go to the website?
you can change it in settings, though some of it may be due to my browser & windows settings regarding colors
Oho
what
Idk, I had no issues
I just pasted the link from github and followed the same process as when I installed Automatic1111
o-o

this is gonna be some bs
there we go i fixed it
i just had to delete the stuff i put in the webui-user.bat
...nvm there's more errors
Ahh, I didn't touch the bat when I first installed it, because it automatically set me to low ram
XD
or cut paste it out of the folder, its probably just whatever cmd args you had in
why do you even have conflicts?? you shouldnt be installing this into your a1111 directory
i mean, i guess??? just install it into a new folder, and then move your models over
...okay that's fair enough
yeah dont over complicate it lol. get the 1 click installer or git pull into the filder
new folder* also, if you have a lot of extensions, be prepared for some bug hunting
controlnet, color cutoff, and ultimate sd upscale
ignore all the red nodes, i havent moved over custom nodes folder yet
BRO IT HAS COMFY INTEGRATION? 
its an extension
havent messed with it much , so not sure how well it works
my experience as well. used comfy backend on swarm UI. but, damn i love the look and feel of a1111 too much to give it up
That image....
That mess of cables
its spaghet
I would suffer if I had to look at that every time
its not too bad honestly. that workflow wasn't exactly the simplest one ever in the first place
The only thing stopping me from using ComfyUI is the fact that you have to detach things when you want to do other things. I can't just tab over to another tab to do imagetoimage, I'd have to redo wires
Like a damn electriction
XD
it was the jboogex machine learning workflow off civit, if you waqnt to see how it looks without all the red boxes 😄
OR just load your workflow where thats already done
What theme are you using btw? I found the theme part in the settings, but none of them have previews
@-@
gradio theme default ui theme dark mode
if youre going to mess with your ui i also highly recommend tabs extension. i have way too many extensions and it would be a mile of scrolling if i didnt have it.
Uhhhh, sheesh
even with just default forge integrated extensions, it removes a big eyesore from ui
OHHH i found what it is. its the dark reader extension in waterfox
thats with dark reader off
What controlnet do I need to download to get the one that outlines everything? I found this awesome figure with wings that would look so cool
imagine being able to use operagx when genning
Wym
im sure opera has some similar extension as darkreader if not the same extension
It's great
other side
bro id rather a page be broken than it fSDFSDF blind me with pure white background
Broken page > flashbanged
yeah but i cant use opera for genning or it gets really really mad at me
and it crashes the page
i honestly hate huggingface for that, i still havent figured out how to force dark on there
gives me aztec vibes
You never answred my question O^O
oh and whoever was interested in comfy on a1111/forge, just be aware that its going to dip into your vram a bit. looks like it was using a couple GB for me
?
Is it one of these? or do I gotta go to git
The extension for dropdown you were talking about
ohh one sec
eek, yeah nah nvm
it'll break my 4gb lmao
A fellow 4gb gamer?
;w; I feel your pain
4gb? ouch
i have 8gb of ram, 4 of vram
damn i have as much vram as you two have ram combined
4090 OC with 64gb ram and a 5950x
30gb ram 20 vram... i want 30 vram...
i AM getting a computer built with a 3060ti tho so :D
soon i shall join the betters with ...46gb of ram
;-; I am being left behind
This laptop is 3 years old almost, it's about time to retire it lmao
I have the Gigabyte G5 MD
you could really blow someones mind and run a browser accessed vm off of a raspberry pi and use that to load up your webui and make an image... just dont tell them its not the pi hardware being used 🤣
Do you at least have 144hz refresh rate monitor?
hey for you low vram folks, if you are able to do it you should run tensor rt as well. that will help with your generation times
What is that?
tensor rt?
basically, makes a version of a checkpoint that runs faster. takes a while to generate it, but if you use it a lot (like doing an animation) it is very worthwhile imo
Can I make versions of all my checkpoints? I don't have a bunch
huh. I'll see what forge does for me and i'll check that out!
_> I will let it run overnight to make them all, if it means it will run faster after
let me restart with comfy disabled and ill make a fresh engine on tensor rt, show gen time with and without it. may not be significant for me but should give an idea
LOL
I only had --medvram as a command
I am sure there are more I could be running, but I am noob
I had --medvram --xformers --no-half-vae
--no-half-vae because 1660ti is.
...very easy to frustrate
if gpu supprts it try adding the pin shared memory and cuda stream commands as well for performance
my gpu supports xformers since I could use that
o3o
I can use xformers yeah
I was using it in automatic1111
I don't know if I need to reput that line in forge
keep an eye on your terminal during start up, it will tell you something like
Hint: your gpu supports pin & stream etc
and will list the args
ohh that
it will be near top of terminal soon after starting
okay so i need to finish this up lmao
oh, note for tensor rt. the engines it makes, you need to also convert lora's to it if you want to use them with that version of the checkpoint
I apparently support using those things, but that error popped up after installing tensor
ignore it and press ok like 5 times
Okay
AH. hi i dont have the right python to path
Correct?
🤷 looks good , if it doesnt start just remove everything after xformers
I am installing xformers again
So every time I start it I will have to spam ok 5 times?
XD This is weird, but if it works, it works
iirc, you might have to install the nvidia computing toolkit. the extension SHOULD do all that for you but sometimes doesnt
; - ; You have made this a lot more complicated than I thought it'd be
well, it might not work at all. im guessing you got those errors because computing toolkit isnt installed or extension did it wrong
hey man, i never said it would be easy. i said it would be faster. if it works 🤣
Do I just install the toolkit from nvidia?
well try running it first
It
see what happens and keep an eye on terminal
it's up and running
i mean, export an engine from tensor rt tab
yep, default first
im glad im sitting here and watching all this first
just to see what the process is like
i recommend reading throught that info on the right while you wait
🤓 error =/= crash
It's still up, but nothing is happening in the terminal
hm
mhm
trying to remember, i feel like there was something i had to do to make this work in forge properly..
In the video I am watching on this, it said the minimum vram you needed was 8gb, but if it's working with 4, I will keep trying
forge do be big. it taking a while to install for me
minimum needed for, tensor rt or forge?
tensorrt
im honestly shocked you can anything with 4
if you know what you're doing, you can do with 4
oh i have a gtx. that aint gonna work for me
ahh
i think my forge install is stuck again. hm.
well if it makes you feel any etter, looks like its broken for me as well now. dont really use it often anyway though since i have uh probably around 30 checkpoints or so, and too lazy to convert all of those + the ton of loras i never use
Ahh, so it's just busted
it will look stuck in a few places
oof, so I just uninstall?
it's stuck on "installing requirements"
It'll hang, it will eventually keep going
Trust me
not stuck, just doesnt have progress bar. takes time
yeah, it's taking a bit longer than a1- oh there it goes
Mhm
yep lol
So tensor is a bust
yeah sorry, was worth a shot though
That makes me sad, guy in the video was saying it would double my speed
less difference with more vram
iirc it was a difference of a few seconds for me on xl model 1024x1024. not much, but if your'e doing a large batch or a ridiculous upscale (5x+) it adds up
alright time for me to test this thing out
i heard krita has got some integration with a1111, anyone messed with that?
............... why did it give me a woman holding an apple lmao
i asked for only an apple
i cant even share it because yeah
quit using them pony models you degen
im using the regular models!!! also my laptop whited out for a second
i feel like this is heavier for some reason
what do you mean heavier? slower generations?
it whited out my laptop when a1111 just makes it lag a little
check that your args are same as what you had, half vae half precision low vram etc
do the args even work???
All previous CMD flags like medvram, lowvram, medvram-sdxl, precision full, no half, no half vae, attention_xxx, upcast unet, ... are all REMOVED. Adding these flags will not cause error but they will not do anything now.
i guess i just use --always-offload-from-vram
i guess i can see what happens if i use cuda stream and stuff
Will it make it harder to make good images?
I am stuck at 9 seconds to make an image of that quality
;w;
Even with the changes added to cuda, and not being able to use tensor
If I can learn how to use Hypertile it won't hurt as much
Right?
RIGHT?
sobbing
that doesnt really have a yes or no answer to be honest. I mean, I'd say youre going to struggle with making anything good anyway because you probably use low steps and resolution to get reasonable generation time. And, even with everything configured for quality over performance i STILL dont think the stuff i get is good and i end up editing in 3rd party software. so, like i said, not a yes or no answer lol
heeyyy i got it
oh hypertile is gonna make it take longer. but it will in theory allow you higher res images when running on a potato
I primarily generate in 512x800, then upscale
i dont know what's wrong with yall's computers. most people agree forge is faster 😄
Oh i fixed my forge, i added the settings and whatnot
and it made it significantly faster
and it didnt crash
👏
I just have to be careful with it still, i dont want to test my limits too far
i mostly generate in 1024x1024 with no hiresfix, then outpaint to the ratio i want before upscale
half the fun is breaking it a dozen times to learn how it works
I'll take it
but yeah 512x768, 1.5 hires... and then throw it into ultimate sd upscale (an extension yu download which you find in img2img, down in the script section) and you can make it bigger and nicer
here's the usdu stuff
but YES it is such a good spot for full body as well, it can mess up the eyes though so you have to be a little careful
here's an image i made earlier at that res
..wait that might be the usdu version
yeah that's it
This is all that pops up when I searched it up
what model do you use for it? i like 4x ultrasharp
same same
the industry standard 😄
4x ultrasharp is nice
What, 4x ultrasharp?
Yeh
30s 😬
i dont know if it's on civitai, i got mine from a youtube video
yeah thats it
goes in the esrgann folder i think. let me checkm
this is pretty good, but def not the ultimate upscale method
good enough for most things though
ahh dang
:C
you can do ultimate upscale
lol
youre in comfy right? whats that image restoration thing people are using on there? i forgot the name
it's like...super light
supir?
Since my stuff is butter than urs
yep supir thats it
i usually just do tiled upscales
but yeah that's really good
for 4x upscale models, i really like lsdir
supir was IMPRESSIVE from what i saw, saw a pretty neat workflow using it with animations
you go to img2img, and click scripts, find Ultimate SD Upscale, and then set the settings, throw your image into it, and press generate
Let me try
i think the others do way too much smoothing (4x models)
I have 6gb of vram but i might as well have 4 since this laptop is low tier
LSDIR is the best
whoever was asking where 4x sharp went , webui\models\ESRGAN
you'll have to refresh your ui to get the 4x-ultrasharp to show up i think
if you put it in the right spot
you'll have to make the folder
So I have to make a ESRGAN folder
tell me more about this LSDIR.. i found smoothing was kind of dependent on settings
I restarted UI, and we shall see what it does now
most 4x upscaler models i've tried like the ultrasharp shit look kinda airbrushed
cheers! will look into that, will test it and compare with ultrasharp once im ready to upscale the samurai
Gonna see what it does to this hottie
did you run that generation with freeu? it gives nice results for non realistic images imo, and shpouldnt add anything to generation time
youre using forge, right? scroll down. its an integrated extension
he looks like he gives good hugs 
I see it, but haven't used it
should be right there in your text to image tab
The best hugs
yeah try running that generation with same settings and seed with freeu
its not huge, but should add a bit of detail
Uhhh, how do I do that with an image I already have?
default settings are fine
...oh my image broke. thanks vae
Idek how to do this
theres a few ways. i usua;lly just got to "png info" tab at top, open up the picture. that displays all info. then click "send to text to image"
jerkwad vae messing up my images
LOL
that will pull everything used to make it back to text to image (i think by default though wont change checkpoint, that can be changed in settings)
also hires fix really slows me down for some reaosn
wonder if i can do anything about that
yeah kinda figured
You'd have to use the original
:c it won't show the things, because I think I modified it in another app after
Not the edited
LMAO THA KS MY IMAGE FIXED ITSELF
God drevis looks so good... I just need the deer antlers to work
OH MY GOD I UPSCALED BY 2 NO WONDER IT WAS SLOW
It made a different image ;-; is it possible to make it make the same image as before?
THOSE GODDAMN EYES
if text to image, gens should be exact same as long as you use same seed
if you pull from png info, it should do all that for you
What model is that, I kinda want it
I did, but it is different, maybe I was using a different model?
no freeu left, freeu on right. ymmv
Seth I must know the model
added some nice small details here and there, and f'ed up some other things. but, the prompt was also complex with high steps and cfg so that probably didnt help
W I D E
not same at all. the changes are small but they are there.
pretty much the same as using an add detail lora with mid-low weight
I wonder what happens if I get rid of highres fix...the images look cursed in the process
tbh im not a fan of highres fix
you're prolly not using the right settings
are you using a 1.5 model? if so, you might get better results for sharpness & detail with inpaint controlnet img2img. and, that will leave your image lower res so you have an easier time doing the real upscale once you've edited to where you want it
Oh, bad news
I made that image in automatic1111, not forge
I can't find the original
I don't wanna figure out what happens if I try sdxl on forge.
I tried with a1111 and it crashed out
My vram just barely meets the req
what do you mean? you can upscale any image you want
Why was I being told I need to find the seed to get that image again ;3; maybe I got things confused
With usdu I'm p sure you can
and if it crashers, do it again but turn on the never oom extension on forge 🙃
thats for if you want to create same image in text to image, has nothing to do with upscale. its useful if you want to make changes in original paramaters like adding freeu, changing prompt weights, etc etc
theoretically, keeps you from going "out of memory"
havent needed to test it so not sure how it works
Oh she nuts
she got that clowndog nose
I also saw you had this on
go to text 2 image tab, scroll down
Changes are funny
shouldn't necessarily copy the settings i had, was just showing you where to find them. but yeah that one should help you a bit depending on your prompting style. shouldnt hurt anyway
🤷 hell turn everything on
enable every single extension, turn every slider to max, and see what happens
if we dont hear from you, ill assume you created a singularity inside your gpu and perished
tile width = image width, tile height = 0, tarhet size : scale to (6x), 4x ultrasharp, leave the rest the same. also scroll up and make sure denoise is set to 0.9 😈
Oh I thi k it crashed when loading the model in
no log?
it crashed when hitting generated? i was (kind of) joking about scaling to 6x lol
Oh dear God yeah it ran out of memory when loading the sdxl model
also that denoise at 0.9 with those other settings was going to create some horror. i mightve been messing with you a bit lol
Oh
XD
Caelus is cursed
Caelus being the laptop
Yeah I named the laptop
I turned on never oom
I don't think I have it set up right
I said fudge it, set the output to max
Says it'll take 4 mins
That is not bad actually
O_O
Okay, okay okay lemme turn off the speedup stuff
All it did was remove his nipples
What the fudge
What is the lowest level of denoise I should use?
I had it set to .45
_> Apparently .45 is the nipple remover
.3?
are you running never oom? its available on both text to image and img2img tabs
Yeah I have never oomph on
That would be even worse wouldn't it? I want it to make it better, not change less stuff
Oh I read that wrong
What am I doing wrong? sob I want to figure it out and sleeb
doing wrong for what?
I want to upscale the image and make it sharper. and I have it in image2image
So! I can't use sdxl, and that's fine
Don't worry, I gave up on SDXL too
I can run it in ComfyUI, but it's too complex for me
rn at least
and it keeps crashing when trying to upscale?
you can prolly run it in comfy no problem, since you have 2gb vram more than me
No no
it just doesn't change anything, it just stole his nipples
:O
Usdu why are you failing me
hold on one sec..
it's not bad, just copy other ppls workflows
ok yeah go into diffusers2 channel on this discord server and do screen share
Hello saw
Trying control net to see what happens
It really likes- ...oh it crashed
Don't crash this time? Please?
"TypeError"...kay
Oh gdi it broke
So it doesn't look like I can use controlnet..again
ok, almost readsy. prepping an image i was going to upscale anyway
TypeError: 'NoneType' object is not iterable
thats going to be a settings issue right there
Oh cmon
whats your preprocessor and model in cnet
..
It's not even on.
Wtf
Hello whiteout
It's not even genning regular images
Wtf did I break
Lemme turn off the speed stuff...
... :/ I'm assuming my prompt corrupted somehow, it works now
🤨
By corrupted, I mean the option to pull it from a previous window was busted
So lemme turn the speed stuff back on...
Here we go
Now it seems to be working
Tried it out with just 'apple' and it seems to work
There we goooo. It's working right
I am in awe with forge, it seems to be running cooler on my laptop that a1111
It is running SO much cooler
ok i may just put in screenshots if this upscale method works. because the generation will take 5 mins or so
Ho! Ho'! Ho! Good morning coffee!
getting close. .6 denoise, needs a few more tweaks and should be less abomination-like
logs?
One sec...I crashed out and I closed it
I'm hoping me using t2i adapters aren't the issue...
Those are so much faster and help me out greatly
cant imagine it would cause a crash.. what are you using em for?
Openpose



