#💬｜general-chat

1 messages · Page 79 of 1

warm junco
#

Every model thats larger than 1.98gb goes into that folder

sturdy mist
warm junco
# sturdy mist where would the smaller ones go?

The smaller files aren't models. These are additional files that can be used together with models.
On civitai.com you will find a lot of models and small files aswell.
Check the description (info card). If its for example a Lora, it goes into models/lora

sturdy mist
civic swallow
#

i tried updating my auto1111 and it broke 😭

warm junco
sturdy mist
#

what is the negative prompt for?

plucky shoal
#

for things you dont want to be in picture basicaly

sturdy mist
#

ok i assumed so just checking

fervent thunder
#

Is inpaint sketch broken? Having an issue where when I send an image to it via text2img, it just... disappears. dramatically. Like it flies off the canvas.

sturdy mist
#

also what is the point in using parenthesis in prompt instead of commas is that just specific to some models?

fervent thunder
#

parenthesis are used to group tokens to add emphasis. So say you have a prompt that contains "A banana split with chocolate fudge" - and it's not generating the fudge? You can try "A banana split with (chocolate fudge:1.2)" to further emphasize that SD needs to add some chocolate fudge to the generations.

sturdy mist
fervent thunder
#

Mhm. Just keep in mind that if the model doesn't know what chocolate fudge is, has no reference for it, you'll be outta luck regardless

sturdy mist
#

makes sense yea

flint jewel
#

is there a community guide to prompting for SD? i need like a noob to expert guide, if there is such.

karmic brook
flint jewel
#

thank you!

#

there is ALOT of content in there, is there a specific something i should look for? i couldn't find it using search

charred thistle
#

had to learn docker to get comfyui working with an RX590 to test it out

#

i don't think I'll be going back to automatic1111

#

i'm not sure if any of you remember the RAM issue I've been having where automatic would take up even an entire 32 gigs?... well, comfy hasn't gone above 3gb

#

should I risk it and try an SDXL checkpoint on an RX590?

compact nimbus
#

wow

#

the mj server sucks ass

charred thistle
#

are some samplers more prone to making mutants than others?

charred thistle
#

44 seconds doing 25 steps of 512x512 with an SDXL checkpoint on an RX590

#

1:53 at 768x768

#

and successfully completed 1024x1024 in 3:25

#

i guess that confirms.. 1024x1024 SDXL on an 8gb RX590, with 16GB Ram, is actually doable

primal current
#

Hello dear friends, do you know of any open calls, contests or events to exhibit AI/Digital/NFTs art or about the metaverse? I recently found a contest for Vogue with AI, but the deadline was closed :/

charred thistle
#

with SDXL, do the quality if images, meaning accuracy, detail, etc get better at larger sizes?

fervent thunder
charred thistle
#

so, i'm pretty sure that was a memory leak I was dealing with earlier with the version of automatic1111 I was using

#

a dozen pictures before having to restart.. with comfyui, I've done like 100 and Python isn't using more than 3gb RAM

narrow sluice
#

Can someone on local SD help me with one prompt as part of a project?

#

I just need it as an example.

west prism
#

some samplers are probabilistic, compared to deterministic ones

noble star
elder palm
#

hey guys

#

i was spitballing some ideas for myself for stable diffusion art projects

#

and i was thinking of making a wordless picture book

#

would my rx 7800 xt kill itself doing picture book sized pictures

#

640x640 is already kind of alot for it

safe isle
#

Hello!

------

| |

| |

------

forest trout
#

:.|:;

warm junco
storm crag
#

hi everyone, I need to do an in-paiting on some frames of a video of my company (remove a logo). I dont care about temporal consistency, nothing fancy.
These frames are scattered through the video.
Instead of manually extracting frames, manually in-paint them and replace them in a copy of the original video, is there some extension of automatic webui or some comfyui workflow that allows me to do it on the fly, in-place with some playback control? Or do you know any faster way?

hasty lily
#

How to generate images plz help I am new

wraith basalt
#

Is it possible to inpaint an outfit from an onlineshop onto a a generation?

#

Or what would generally be the easiest solution to do that?

charred thistle
#

so, both SD2.1 and SDXL are fairly new?.. what exactly is the purpose of each one?

tardy geyser
#

idk how ppl make these and figured id throw this out here in case anyone just knew

warm junco
tardy geyser
#

THANK YOU!

charred thistle
#

hmm, all my SDXL stuff comes out looking like 1950s cartoons, despite specifically asking for realistic

#

well that was easy to solve, I was using a resolution that wasn't immediately understood by the model

midnight crag
#

Can someone help. I installled Automatic 1111 and it works fine, but it loads not showing me all the screen UI. Its like zoomed in, i cant see the all of UI, i always need to be scrolling to see other parts of the UI. Is there some setting option to make it normal? Thanks.

charred thistle
#

sounds like your browser might be zoomed in

bleak matrix
#

Good morning, everyone! How are we all this beautiful day?

midnight crag
warm junco
midnight crag
#

Chrome

warm junco
charred thistle
#

I'd try another browser too

orchid spade
#

did you got an answer?

lofty shell
#

Hi Dear Friends. I got some SDXL Loras, it says Lora. I packed them to models/lora but i cant see them in stable diffusion. Can anyone help me with that? is it in the right folder? or do i need some other extension? As Example: https://civitai.com/models/124016/sdxl-chalk-dust-style it says lora. is it not a lora?

quartz helm
charred thistle
#

i may have to stay away from the SDXL stuff for a while, my GPU will do it.. but it doesn't like it

#

by that I mean.. it does take 3 1/2 minutes to do a 1024x1024

#

I'm curious, but how many people here support "open source" concepts?

forest trout
#

I mean if you prefer stable diffusion over the alternatives I'd say there's a 80% chance you also support gimp, blender, etc.

charred thistle
#

I'd imagine some people just run it because it's free though, or can be ran from their PC without paying for tokens

charred thistle
#

something else I'm curious about.. is there a reason we have a starboard channel, and not a port channel?

charred thistle
#

I'm testing out more SDXL models, see if I can find some more that I like.. the image I got earlier was using the standard model

fiery dock
#

Discord down for like 10 mins for anyone else?

warm junco
vocal vector
#

Yeah, the Discord status page shows an unresolved API error, but says a fix is being implemented.

charred thistle
#

yes, port and starboard

warm junco
charred thistle
#

port is left, starboard is right

warm junco
#

okay so what do you vote in port? im sry but i didnt have seen a port channel anywhere else just starboard for top images

forest trout
#

It's a boat joke.

wise rapids
#

Heey i need help

#

I launched ST but i try to input a image of a girl (ai) to remove her clothes. I get runtime error all the times

sharp terrace
#

Hi. I have a question - is there a good model for compositing a foreground and a background image together.

so if I have a fg image of a bottle, and I want to place it on a big image that provide (not one that is generated from scratch)

it would composite them and add shadows and other pieces to make them fit better together

charred thistle
#

what happens if you just copy and paste the the foreground onto the background and use it to generate something new based off of that?

patent solstice
#

Is there any possible way to run stable diffusion without a NVIDIA gpu?

hasty hornet
raw dagger
patent solstice
#

hmm, do I have to have a model installed for it to work?

charred thistle
#

are you able to open the webui-user.bat and run commands individually from a command prompt to see where it actually fails?

patent solstice
#

no time to do anything

charred thistle
#

open the command prompt alone, if you're on windows you should be able to hit the windows key and type in cmd

patent solstice
#

so just type in the name of the .bat file?

#

or am I just being dumb

jade mesa
#

where can i ask for help in warpfusion

jade mesa
#

dope thanks

quick pecan
#

Maybe a dumb question that can be found elsewhere… but what does putting a number less than one do when in parenthesis? Like (small dog:0.5), does that actually lower the “weight” of the word?

#

My mistake Mr. The Chad. Won’t happen again

charred thistle
#

@quick pecan 1 is 100%, 0.1 is 10%, so how much you want something to have an affect is determined by that number

fervent thunder
#

GUYS PALESTINIANS ARE GETTING MASSACRED, PLEASE DON'T LET WESTERN MEDIA FOOL YOU THEY'VE BEEN DYING AND SUFFERING FOR DECADES PLEASE,I STAND WITH PALESTINE

quick pecan
charred thistle
#

one direction gives what ever you're applying it to more weight, the other gives it less

#

if I'm not mistaken, a low number like 0.1 will have a more subtle result

prisma siren
#

Heyo

#

I have question

#

What’s the seed variation about? What does it exactly do?

forest trout
#

Basically the diffusion process takes the noise and tries to find an image out of it.

prisma siren
#

How do both regular seed and variation seed relate to each other? What does the variation seed affect when I have already input a regular seed?

forest trout
#

Variation seed is usually for fine tuning.

prisma siren
#

Seed: 803755656
Seed var: 3402361402

forest trout
#

I rarely use it but that's its main purpose.

vast ingot
prisma siren
forest trout
prisma siren
#

Then another question, if I change the initial seed but keep seed variation the same, what should be the expected behaviour?

vast ingot
#

seed variation is nice for small changes in a given image, yeah

prisma siren
#

Oh so it’s the other way around, I see

vast ingot
#

if you change the initial seed, then the variation will be done from the new seed, meaning a completly new image, close to the one from that new seed
but since you didn"t process that seed yet, it's useless

prisma siren
#

I should keep the same initial seed and change the variation seed then?

forest trout
#

If you want the same image yes.

vast ingot
#

yeah, variation is useless when you change the seed. better put variation back to 0 if you change seed

forest trout
#

Variation seed is useless if you change the main seed.

quick pecan
prisma siren
#

Is there a specific process to go about variant seeds? Or is any random number gonna work?

vast ingot
#

so once you are ok with the quantity of changes, modifying the variant seed gives you more "small variations" around your original picture

#

it's very usefull in combination with features like "X/Y/Z plot"

prisma siren
#

Like, can I type a completely random number and expect it to work or should the variation seed number be within some specific subset of the initial’s seed character?

forest trout
#

It doesn't matter.

vast ingot
#

you can type a completly random number as secondary seed yes

#

it's another random seed to go on top of your first seed, basically

forest trout
#

It's a seed so the randomness is equivalent for every combination.

vast ingot
#

what matters is the amount of variation

prisma siren
forest trout
#

He's talking about variation strength

#

Not the seed number.

prisma siren
#

Oh

vast ingot
#

it's a little like doing some "img2img" steps at the end of your image generation, this feature.
First, it process your base seed, as is
then it does some changes by adding some noise, using your second seed. the noise is random, as is the seed
but the changes are more or less important based on the variation strenght

forest trout
#

So if you want the variation seed to have a stronger influence over the image you adjust its strength.

prisma siren
#

I’m trying to figure out the variation seed functionality in general first

#

Before proceeding to variation strength

vast ingot
#

variation seed is exactly like the seed in img2img

#

it's a random number

#

type 1 or 10000000, it will have the same strengh of impact

#

this impact strengh is in the variation strengh

#

but there is not a lot more to learn from the seed

forest trout
#

Actually I wonder if the noise seed patterns are the same for variation and regular seeds. If you had the same seed for both variation and main would that even change anything? 🤔

vast ingot
forest trout
#

Makes sense I suppose.

prisma siren
#

As I understand currently, the seed variation is just a subset member of the initial seed family. Something like a A subseed of the initial

#

But if I make the case of having both the initial seed and variation seed be the same number then what will happen?

forest trout
#

Guiz answered that up earlier.

prisma siren
#

Yeah I read it, thanks for the explanation!

vast ingot
#

well, when a picture is made, the initial seed decides what "noise" is used as base
then the "steps" happen. it works with your tokens to make sense of the pictures, drawing what you asked using the weights in the model
then, it adds noise, using your variation seed. More or less noise is added, using the variation strengh
then it does some more steps

forest trout
#

So it just deepens the intensity of the noise in those regions if the seeds are the same.

prisma siren
#

Nice, I understand now

vast ingot
forest trout
#

I'm talking about the initial noise pattern.

vast ingot
#

yeah but I mean, there is nothing left, almost, from that base noise pattern, no ?

forest trout
#

When? After the diffusion process?

vast ingot
#

yeah, once the steps have first processed and we start caring about that variation

forest trout
#

Yeah I mean it won't matter at that point like you said.

vast ingot
#

there are others way to go about that feature, that other tools tried. like adding a small noise to the base noise, before doing any steps

#

in comfyUI, adding the noise of different prompts in the middle of the diffusion process, can help compose quite nice pics

forest trout
#

I mean that's basically the euler a scheduler...

#

Any ancestral schedulers for that matter.

charred thistle
#

learning about controlnet now 😄

charred thistle
#

at the mapping part of it

fervent thunder
#

Are images generated in DreamStudio saved to a cloud? Or is is download only, no internal save?

charred thistle
#

depth maps are cooooool

rare grove
stray adder
#

New Here not sure if this is the right spot, I have 2x A100 and a H100 on the way, Is NVLINK Supported on Stable Diffution, and on these cards will I see a improvement?

tender cove
#

Afaik 4090 is more efficient than a100, unless you've got free access somehow

#

Generally don't need all that VRAM for stable diffusion. As far as distributed inference across multiple GPU on the same machine, I've seen exactly one implementation and the author recommends against it. Dunno if there's anything newer.

charred thistle
#

I've actually done a lot of SDXL today with 8GB of VRAM on a 5 year old GPU

sharp terrace
green void
warm junco
#

Its the core base without user interface or additional features

gentle hill
#

With image inpainting we can remove photobombers, can we do same for video?
Any suggested method?

analog vector
#

Help, can I use my own civitai model in stability ai?

verbal finch
#

With sd2 they make a big deal about it not being trained in living artists. They dropped that with sdxl and started training on living artists again I'm assuming?

stray adder
#

To be clear I bought the A100 I am not renting a server

#

@tender cove OR did you mean in terms of power draw?

tribal needle
#

which model does @radiant meadow use to generate the images it is creating??

#

SD 1.5?

#

SDXL 1.0?

warm junco
tribal needle
#

and this is a dc server..?

green void
#

Is comfiai better for gtx 1070ti

warm junco
tender cove
# stray adder <@215227108916592650> OR did you mean in terms of power draw?

Price, not power draw. My understanding is that a100 is about capable as 3x 4090 when it comes to SD, but costs ~5x as much. As for power draw/tdp, a100 is ~1/2 a 4090, but some people in stable horde are underclocking/undervolting their GPUs pretty heavily for only ~10% performance reduction, though I'm not sure which GPUs or how well that transfers to the 4090.

#

Keep in mind that I don't actually have either GPU, so it's really just something to look into.

tribal needle
#

is ComfyUI capable of doing img2img?

#

it seems confusing to set it up

charred thistle
#

pretty sure it can, yeah

charred thistle
#

what determines VRAM usage?.. do higher resolution images require more?

#

seems simltaneous renders in batches will do it

#

at 512x512 SD1.5, it's like 900mb per batch size

unkempt heath
#

Where can we have hidden text in images? Please ping on reply 😄

charred thistle
#

secret watermarks?

unkempt heath
charred thistle
#

aah

unkempt heath
charred thistle
#

I just placed an order for a Sparkle A770

unkempt heath
#

?

charred thistle
#

GPU

unkempt heath
#

Oh

charred thistle
#

I should be able do things much faster after I get that, and probably focus on more SDXL

unkempt heath
#

Is that free

#

Very noob I know 😄

charred thistle
#

the graphics card?

#

no, they're not free

unkempt heath
#

I meant SDXL

charred thistle
#

yes, but SDXL models are trained on 1024x1024, vs standard 512x512 so they take much longer and require a lot more resources

unkempt heath
#

Ohh

#

No potato laptop

charred thistle
#

I can do SDXL on an RX590 but it's 6-7 minutes per image

stray adder
#

I'm also open to sharing access if anyone has a work load that can benefit from a 40g or 80g card

unkempt heath
#

😮

#

Sharing access in what way?

stray adder
#

Running a workload on it directly

#

I can even spin up a VM if you need a closed environment

unkempt heath
#

Ah okay

#

Ty

tender cove
#

... No it's not. Just the website is. 1 sec

#

Bleh. Send me a message if you want an invite

stray adder
#

Done

charred thistle
#

is there an easy way to tell which prompts are not being understood?

iron arch
#

pixel 8 pro can do what sd 0.1 did 10 years ago

#

XDD

#

i bought a p8p... and its so bad.

#

its like a pc but you remove 99% of the features and strength and make it pocketable.

tender cove
#

I prefer tablets in general. Bigger screen is easier to use, almost as portable, and more surface area makes for less throttling so more responsive. I get the cheapest phones I can, and a decent tablet and I'm set.

iron arch
#

i dont know if i should send the p8p back

#

i only bought it cause i wanted free pixel watch 2

tender cove
#

More surface area, better heat dissipation, CPU can run fast for longer

iron arch
#

its not likeu need cpu running fast

#

on a tablet...

tender cove
#

I have a full Linux environment in termux on both my phone and my tablet.

iron arch
#

😒

warm junco
iron arch
warm junco
iron arch
#

when i ll do illegal stuff ill do privacy otherwise irrelevant

iron arch
trim magnet
#

why are toilets private if u aint doing anything illegal

iron arch
#

thats not a rational thing.

#

thats getting used to x.

#

@trim magnet

charred thistle
#

curious, if I did something like (adjective noun:1.0) will this effectively group that as one prompt, but not add additional weight to it rather than processing the words separately?

tribal needle
#

is i possible to invite @radiant meadow to our own server

#

@novel ocean

green void
charred thistle
#

yeah, but I really had to work at convincing myself to go for more vram, because the 8gb A750 is only $190

green void
#

i might get that card too im thinking been eyeing the 3060

#

but honestly all the games side by side run smooth enough in comparision

#

the 3060 usually like 10 fps higher but i just play at 60 fps anyway lol

rain vale
#

Hey guys new here! What do you guys use SD for? I personally use it for video editing been wondering in what other fields people are using it?

forest trout
#

I just generate images for fun. It's just a hobby for me.

potent spire
#

Sometimes i basically photoshop and blend me and my friends or celebrities with those images for fun as well

hasty hornet
#

I stopped using SD since google collab banned web UI's on free accounts

#

prohibited, I should say*

#

thinking my 1050ti too slow, too low vram too, not fun

pale latch
#

end prohibition

covert igloo
#

Guys need help

#

I want to create a QR code with my on portrait in the background

#

Any ideas ????

jolly verge
#

does anyone know if comfy ui can work with 8gb ram?

vale steeple
charred thistle
#

@jolly vergeram or VRAM?

#

right now I'm doing SD1.5 stuff with controlnet and my system total is using 7.85G RAM and less than 6gb VRAM

void valley
#

Hi all! Is anyone here using an RTX 3060 12GB? I'm thinking of getting one for myself. But I can't find any info on what is the largest resolution it can generate images at in Stable Diffusion.

charred thistle
#

from what I understand the 3060 12gb is pretty solid at AI for the price, it was on my list of possibilities until I decided on the one I bought last night

#

@void valley also, it's not so much what resolution it can handle, but how long it's going to take, and how long you want to wait. I can do 1024x1024 on an RX590 easily, but it's going to take my 7 minutes per image

void valley
#

Quite honestly I just want to hut the 1080p. I can wait. I am running cpu on one machine and a gtx 1650 on the other right now. I mostly just run them locally and control them from the phone while doing other things. So waiting is not a problem.

#

I also have an rx570 8gb. But as far as I checked it wasn't supported.

charred thistle
#

not officially, but there are solutions

void valley
#

I saw one guy setting it up undel ubuntu. But quite honestly I am not that experienced with linux.
Do you run your 590 under windows?

charred thistle
#

and i'm not sure if the solutions work on Windows, maybe

#

if you can run docker containers on Windows then it should work

void valley
#

Yeah. But at that point I might as well just boot up linux from another ssd and learn that.

#

Whichever has easier to follow tutorials I guess.

#

And how well does your 590 work?

#

What speeds do you get on 512x512?

charred thistle
#

without LoRAs I can do 512x512 in like 44 seconds

#

since I am using LoRAs right now and doing 512x768 it's taking me about twice that per image

void valley
#

Yeah. Any chance you remember the it/s speeds on the 512x512 or the 512x768?

charred thistle
#

but a guy that goes buys ulyssessrr has docker containers for a version of ROCm and torch that work with the rx590 that I use to run comfyui from

void valley
#

My gtx 1650 gets me like 6sec/it on 512x512 and around 15sec/it on 512x768. I am wondering if it would even worth the hassle to set the rx570 up.

charred thistle
#

i'm doing about 3.90s/it

#

it may be a pain in the ass figuring out how to use docker containers to do it, but if you attempt it, go for the torch docker and run that with comfyui, automatic1111 is unusable for me

void valley
#

Thank you.

charred thistle
#

and that is part of the reason why I didn't buy AMD again for a new GPU

abstract dust
#

hey guys, i am working on using stable diffusion sdxl model for inference purpose for generating images using some prompts, What is the hardware requirement for fine tuning sdxl? Thanks

proven swallow
#

Hi I need a help when trying some prompts with words kids , young the stablity ai api shows some error like banned words or invalid prompts Can I get list of banned words to try some valid prompt?

trail mortar
cerulean zealot
#

Hi guys, now "path to ckpt" no longer exists in stable diffusion automatic 1111... it has been replaced by path_to_model or model_link.
It doesn't work to use the path "/content/gdrive/MyDrive/Fast-Dreambooth/Sessions/"blabla"/"filename".ckpt... does anyone know how to do it now?
I'm using google colab

reef epoch
#

@wise stratus its been awhile 😔, entertain us pls

smoky zinc
#

Anyone gotten SDXL ControlNet inpainting to work in OpenOutpaint? It doesn’t seem to be masking properly. It’s just generating a completely new pic over the original

charred thistle
#

i probably won't touch SDXL again until next week

#

it takes about as long to do an SDXL image on my GPU working properly as it did taking a standard image to finish using CPU only

#

I'll have an A770 next week

fervent thunder
#

where do I go for help with a problem im currently experiencing with comfyUI?

#

trying to use the sdxl refiner but the node just gets bypassed

jolly verge
charred thistle
#

SD1.5, easy for 8gb VRAM

jolly verge
#

So is it minimum 12gb for confyui?

charred thistle
#

especially if you use comfyui which seems to use fewer system resources than automatic1111

vast ingot
#

nah, comfy can run under 12 too

charred thistle
#

no, comfyui seems to use fewer resources than automatic1111

vast ingot
#

it has less overhead than automatic, true, little lighter

jolly verge
#

Ah ok.

charred thistle
#

you can even do SDXL on 8gb with Comfy if you don't mind waiting 5+ minutes per image at 1024x1024

vast ingot
#

I love that comfy has become popular with time, it's one of the best tools around imo

charred thistle
#

yeah, it seems confusing at first, but there's so much more you can do with it without having to have someone else rework a UI for you first

jolly verge
#

I saw a video saying you can use a portrait image to make images that have poses. Can I use base Comfy? Or do I have to get Loras or other things?

charred thistle
#

are you talking about using controlnet to copy a pose?

vast ingot
#

comfy has everything you need as far as code go for those things, but you may need models too. for what you describe, you may need an openpose preprocessor (to extract the pose from the base image) and a controlnet model using that pose if I understand well what you describe (not sure what that video was showing).

#

I used comfy for such comvoluted things to be honnest, almost anything is possible in that

jolly verge
#

Let me see if I can get the video

#

Can we post links?

vast ingot
#

like generating an image using 4 different models, merging the latents, ...

#

I think we can ? not sure, this is the general channel, it may be blocked, but it's mostly possible to post link in this discord yes

charred thistle
#

yeah, I set up controlnet in comfy.. just needs to load image node into a midas depth map node and feed that into an "apply controlnet" node, and that controls the output

jolly verge
#

Oh it’s not showing up

charred thistle
#

I need to figure out how to do backgrounds and foreground subjects separately in general, comfyui might be the easiest way to do it because I could probably use a depth map to detect the character which I could use as a mask for an img2img after generating a background

vast ingot
jolly verge
#

Ah ok thanks

#

Ok I posted it in there

vast ingot
#

yep, this uses controlnet mostly, so you'll need to download some controlnet models, but that's about it

charred thistle
#

piecing together nodes to do custom things automatically isn't really something other UIs can do, but comfy can.. in fact, I'm about to try to set that up right now

jolly verge
#

What about certain features of an image? Like if I want the exact eyes of a character in one image to be the eyes of a different character? Is that possible? Or am I gonna have to have both images and just keep regenerating it?

vast ingot
#

you're talking photoshop territory there imo.
You can use masks, to let the AI alter only the eyes, but you won't just sum the eyes of another character from another picture, you'll generate new eyes

charred thistle
#

basically, if you apply a mask of an image over where a character is at, you can apply a background to said character with a separate set of prompts.. so what I am attempting to do is use a depth map to detect where the character actually is, convert that to a mask, and use that

jolly verge
#

One more thing. I have both a desktop pc and a MacBook. Has anyone run comfy on a MacBook? Or is just not enough power to run it?

vast ingot
jolly verge
#

ok, thanks for the help!

vast ingot
#

example of controlnet (using a 4 leaf clover as base image) and 4 different prompts for 4 different elements, using different models too. Comfy is quite the best for such things
#🏞｜general-with-images message

#

well, 5 prompts if we include the total picture one that is used for the last few steps

charred thistle
#

right now where I am at is just converting a depth to a mask so I can do everything around the foreground subjects

fervent thunder
#

I tried to generate this scene in SDXL. I got the background and the beam, but all other elements were ignored.

charred thistle
#

whelp.. I got a workflow that automatically deletes and recreates backgrounds

#

generate the subject character first, that output image goes into a depth map, depth map gets converted to a mask, mask gets inverted then goes to a vae encoder for inpainting, which sets the latent image for a new ksampler node which has it's own positive text prompt for the background

#

i just wish there was a simpler way to straight up do the foreground and background separately without having to do subject detection

vast ingot
#

well, by definition, foreground and background is that, separating the subject from the rest.... hard to imagine one without the other

charred thistle
#

yeah, but it's difficult to actually describe multiple subjects in one prompt and not having terms get mixed

vast ingot
#

that's also a problem even without foreground/background. but there is an extension for that too !

#

this is meant to limit the impact of tokens to specific group of words

fervent thunder
#

I think that the way to avoid confusion is to have some mathematical terms or physics included in the prompts. Like vectors, coordinates, etc.

vast ingot
#

you can also target the prompt to specific parts of the picture, with area conditionning

#

not sure if it's in the main branch now

charred thistle
#

i may try some of those.. IMO, how it is right now is a bit of a weak point with AI

vast ingot
#

right now, things like comfy UI letting you code even custom nodes quite easily, are tools to build dedicated pipelines for specific tasks
You can spend some hours on a workflow that will do wonders, and call it from the command line, integrating all this in other process for different purpose
such tools like comfyUI are the main glue that will put together more advanced, specialized and user friendly tools imo

#

what's the state of the field currently on LLMs, open source models in particular ?

charred thistle
#

working this sort of stuff out though.. the problem solving, this is actually fun to me

vast ingot
#

yep, I get that 🙂

#

I spent so much more hours making workflows, labelling, adding rerouting for clarity, groupping, making it more powerful and modular, .... than using the workflow once it's done

#

one strong thing that I loved in the "targeted conditionning" concept (that lets you set where a specific conditionning is applied) is that it also works with controlnet, meaning you can use multiple controlnets with different models on different part of the picture if needed

#

it can also be quite the bother when scaling the picture, since the conditionning needs scaling too

charred thistle
#

right now my workflow looks like a spaghetti monster

#

and the output for "santa clause" in one prompt, and "the surface of mars" on another, comes out looking like the corpse of santa clause half buried in the sand on mars

charred thistle
#

i probably shouldn't even bother with this until I get my new GPU and can do renders a lot quicker, which lets me test results and adjust much quicker

jolly verge
#

I finally got comfy running on my macbook

charred thistle
#

how well does it run on that?

jolly verge
#

seems to run pretty well. It took like 30 seconds each image for a batch of 4

charred thistle
#

512x512 i'm assuming?

jolly verge
#

yeah, I can try with different one

charred thistle
#

i wonder... is it possible to have multiple checkpoints going at once?

#

that's be cool for simultaneously comparing checkpoints

untold hill
#

Hi guys, was hoping to get clarification on hosting my own stable diffusion server..

I'm new to developing with StableDiffusion and wanted clarification after trying to do my own research, feel free to flame me if I didn't RTFM. I want to deploy my own web application publicly which uses SD in the backend with an already-trained model downloaded from CivitAI.

What is the cheapest cloud server / service I can use to host a server containing my Stable Diffusion model that generates content based on prompts. I don't want to store any generated content other than user information for logging in and authentication purposes.

Has anyone used this service? https://console.cloud.google.com/marketplace/product/techlatest-public/techlatest-stable-diffusion?q=search&referrer=search&hl=en&project=ai-sandbox-398320

I'm wondering how this works vs doing it manually. Any and all advice appreciated.

gilded trout
#

what does this mean on the stable diffusion webui page "Your space is on error, check its status on hf.co"

charred thistle
#

that makes no sense to me

#

curious.. when you do prompts, it it best to do more important prompts first?

gilded trout
#

anyone?

plucky shoal
#

i think it is not official sd webpage.
Stable diffusion webui, is local installation, wouldnt be surprised if hf.co is some scam page.

fervent thunder
#

Hello peep!

soft sparrow
#

is there a /describe feature here? if not where can i use it. I have seen it somewhere

#

i know midjourney has one but its paid

fervent thunder
fair spear
vast pollen
#

Bing AI no longer alows the word 'demon' in prompts

#

or nuclear bomb

#

I wish the model Bing AI uses was downloadable because their promp blocking is garbage

iron arch
#

id like to buy a SD phone

hasty hornet
#

thinking wot why demons

soft sparrow
#

to describe the image

soft sparrow
fervent thunder
charred thistle
#

i'm personally never using a1111 again

tough island
#

hey everyone

#

would this be the chat that i could post a question?

#

oh i think i found it!

charred thistle
#

I'm actually surprised how much of a difference denoise makes, it's almost like having a completely different model

small vigil
#

Hey all. this is a rando question - has anyone yet to see a consistent approach of a trained model of a realistic character been used to inpaint into realistic photographs? Basically to inject them into scenes. I've seen a few use cases but nothing solid. Wondering if it's too far a leap (especially if the photographic style/scene/lighting) changes too much across photographs

#

Thinking of using a posable model to create a depth map to match into the scene and inpaint in. but it's been tricky so far, not sure it's a rabbit hole worth exploring quite yet

fervent thunder
charred thistle
#

it was nothing but problems, besides comfy is more versatile

fervent thunder
#

i also tried comfyui but coulnd't settle with its interface

charred thistle
#

I like visual scripting, something that's cool about it is if something fails, you can actually see where the failure is immediately because you can see the lines of code being executed in order

#

which was great when I was making stuff in Unreal Engine

#

less syntax sensitive than traditional coding

neat cypress
charred thistle
#

so what was the point of SD2.1 then?... it's here and almost immediately surpassed?

halcyon escarp
#

hi

#

hi guys

#

:)

past idol
#

I can't seem to generate images

fervent thunder
past idol
#

O shit nvm

charred thistle
#

unless SD3 is trained on higher resolution, and replaces both SDXL and SD2.1

#

what resolution is DALL-E 3 trained at?

#

seems to be 1024x1024, same as SDXL

hasty hornet
#

as far as I remember SDXL was intended to be part of something better thinking

charred thistle
#

perhaps.. and if they do successfully clone DALL-E 3, I see it as being an SDXL replacement, not a standard SD replacement

#

so, maybe SDXL2.0? Would be nice if the improvements could also be applied to standard

pale latch
#

sdxl 1.1

charred thistle
#

by the sounds of it, SDXL2.0.. I've seen what DALL-E 3 can do, it's a game changer for many, if stable diffusion can keep up, that'd be amazing

pale latch
#

game is the same. Game changer is like when shaq started slam dunking so hard that it broke the hoops. game changer is like when google maps came out and reloading webpages was less important.

game is even worse with dall3 since its a restrict license from a service

#

dalle 3 is over hyped. its not that great. people can use sdxl to greater effect if they learn

#

"it understands prompts better" but thats all you can do is prompt it. sdxl has countless other ways of controlling and affecting an image since it's open. also, what's good is understanding it if it can't be used for anything except looking at for personla use only

fervent thunder
#

bing is censoring just too much. I want to create explosions, war, blood but not real graphic content. It's fantasy art such as vampires, destruction caused by fireballs, magic, etc

pale latch
charred thistle
#

@fervent thunderfor sure... I actually got a temporary ban from Bing because I was asking it for body horror stuff

#

specifically, someone tearing out their own eyes, like in event horizon

cunning girder
#

I've been using DreamStudio to try to generate images of a scorpion and it's just not working, it keeps giving me some sort of shrimp/lobster looking thing, any tips?

Prompts I've tried:
A golden scorpion with eight legs and two big crab claws and a long stinger tail

A golden scorpion with a long, tall, tail with a stinger curved up with red eyes on a leaf with a dark green background

hasty hornet
#

thinking go higher weight on scorpion

hasty hornet
#

I'd get rid of "with" , "and" , "a" too

potent spire
#

@pale latch the way the AI understands the prompts and how you can interact with it via ChatGPT is indeed mind blowing

#

But you dont have all the customization that Stable Diffusion or Firefly offer

#

But people might use more than just one of em like i di

#

di

#

do*

hasty hornet
#

daboo dee daboo da

grand glen
#

generating images is like gambling, just need to generate 1 more time till i can get the perfect image

vast ingot
#

the best strategy is to put all your AI points into luck. A good luck of the seed can drasticaly reduce the total count of images you need to make

#

that, or reducing your perception score, so you don't see the imperfections

hasty hornet
lilac rune
#

I made a open source AI Painting bot😂

fresh ivy
#

What does "DM A" and "DM B" mean?

stray folio
#

Hey all . Does anyone here interested in smaller docker image for amd rocm pytorch . The official amd pytorch image i dowloaded was 35-50GB . Now i was able to go down to a 10-15GB base image just by changing to fedora without anaconda . And i havent done anything funky yet

#

Tested with stable diffusion webui and comfyui . Both works nice so far. with just basic installation

warm junco
hushed ferry
#

can anyone tell me what this means: "Warning Image-2-Image: disabled" ... happens whenever i check the image 2 image box

stray folio
warm junco
soft sparrow
#

does anyone have a really long prompt for stable dif? I need to prove a point about them existing

stray folio
#

I must take back my statement it probably would not work as amd hasnt resolved wsl2 rocm as of yet

solar condor
#

yo guys, i tried to use dreambooth on my mac and here what i have it doesnt work.. when creating the model

#

Duration: 00:00:00
Error completing request
Arguments: ('CCCarlitosStyleTTTattoo', 'anything-v3-fp16-pruned.safetensors', '', False, '', '', False, False, '') {}
Traceback (most recent call last):
File "/Users/carlos-albertorios-flores/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/utils/utils.py", line 273, in f
res = func(*args, **kwargs)
File "/Users/carlos-albertorios-flores/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/ui_functions.py", line 984, in create_model
result = extract_checkpoint(new_model_name=new_model_name,
File "/Users/carlos-albertorios-flores/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/sd_to_diff.py", line 151, in extract_checkpoint
original_config_file = get_config_file(train_unfrozen, model_type)
File "/Users/carlos-albertorios-flores/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/sd_to_diff.py", line 82, in get_config_file
model_version_name = model_versions[model_type]
KeyError: ''

#

Traceback (most recent call last):
File "/Users/carlos-albertorios-flores/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/routes.py", line 422, in run_predict
output = await app.get_blocks().process_api(
File "/Users/carlos-albertorios-flores/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/blocks.py", line 1326, in process_api
data = self.postprocess_data(fn_index, result["prediction"], state)
File "/Users/carlos-albertorios-flores/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/blocks.py", line 1229, in postprocess_data
self.validate_outputs(fn_index, predictions) # type: ignore
File "/Users/carlos-albertorios-flores/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/blocks.py", line 1204, in validate_outputs
raise ValueError(
ValueError: An event handler (f) didn't receive enough output values (needed: 10, received: 3).
Wanted outputs:
[dropdown, html, html, html, html, html, html, html, slider, html]
Received outputs:
[None, "", "<div class='error'>KeyError: ''</div>"]

charred thistle
#

i think after I get my new GPU I'll probably just be doing SDXL exclusively, since according to benchmarks I'll be able to do 1024x1024 stuff faster than I can do SD1.5 512x512 now

#

also, something I've really began to appreciate with comfyui is the ability to have a bunch of different text prompts placed off to the side somewhere that I can plug in and swap around as needed

brave jetty
#

Hi! I'd like to find an equivalent of " Stable Audio " that I could be install on my PC. The same for " Suno AI " that allows people to generate songs (music + text + voice)

worldly bough
#

Not sure if this is the right place to ask. I'm trying Fooocus for the first time, instead of Automatic1111. Fooocus seems to only let me use 5 loras at once though. Is there a way to add more lines for loras?

Also just realizing that I don't know how to use embeddings in Fooocus. Or whatever the files that go into the "embeddings" folder are called.

runic merlin
#

guys I have a questiooon, can I change de name of the models and loras? Because ufff is quite a mess and im always not remembering the one im looking for and im not sure if it can affect
Or if even I can use Subfolders?

stark spindle
#

why do all my outputs look so ugly

#

how do i fix this

#

like why is it so pixelated and stuff

#

also my computer doesnt seem to be using much resources at all

#

task manager tells me nothing is really doing any work

runic merlin
#

u have wrong vae

#

like if u have xl vaeo u need xl model

#

but if u mix them

#

caput

stark spindle
#

oh

runic merlin
#

I mean that can be 1 of the reasons

stark spindle
#

alright, ill try finding another VAE and model

runic merlin
#

yeah

#

u will end with few of everything

#

I have 80 GB of models and loras

#

😦

stark spindle
#

jesus christ

#

alright lol

pale latch
# warm junco Does the docker image works on windows ?

nope. accessing amd gpu through wsl2 i thought was going to be easy too when i was using AMD. Apparantly microsoft has tried to work with them to get their drivers up to speed for WSL2 but amd's window driver engineering team is kind of, well, you know

#

the VM needs the host's drivers to support pass through and AMD isn't there

warm junco
pale latch
#

Not till 8000 series probably

#

if i had amd and was trying to do this machine learning stuff, i'd switch to nvidia. in fact, that's exactly what happened. Bought amd gpus exclusively for nearly 20 years.

#

they were cheapper and could play games still. the limitations weren't that hard

warm junco
pale latch
#

still no xformers builds though

#

though i suppose sdp works on amd i think

fervent thunder
#

are blocked images accessible to whoever moderates the AI?

#

"an angel impalling a dragon" was blocked. But "an angel slashing a dragon" was not.

pale latch
#

impalement is too sexual

charred thistle
#

i'm using an RX590 with ROCm on Linux right now, it works fine

#

but AMD did drop official ROCm support on my card and it looks like they're trying to tighten up which GPUs are allowed to make use of it

upbeat tide
#

Anyone got a prompt that creates a clean visual representation of a neural network? Or else a Lora or some embeddings? I'm lost and can't create one in SD (Automatic1111 webUI),.. any help appreciated, thanks. 🙂

#

I tried Img2img and it just copies the existing one I got from the interwebz,.. hardly original or accurate.

fervent thunder
#

How i can generated images ?

surreal perchBOT
#

@fervent thunder

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is a public bot on the server that generates images available as a research beta for SDXL, you can find the current status of the bot in #1047610792226340935. There are plenty of ways to use Stable Diffusion such as the official https://dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware - check out #1080946152318443610⁠ for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

fervent thunder
#

I'm asking about this server, there are 1 - 10 bots

woeful marlin
#

Been trying to find a good place to buy a used 3090, tried ebay got burned on a bad card (return accepted luckily so didn't lose any money but still) are there any reliable ways to buy a good used card or is it just take your chances on ebay and keep trying till you find a good one?

fervent thunder
#

buy a new one, less problems

woeful marlin
#

Sadly I need both kidneys

slender dirge
woeful marlin
#

Related question how much better is a 3090 than a 4080, in terms of training and generating large images (speed not as important to me) is the 3090 the clear winner?

fervent thunder
#

(

dry sparrow
#

hi, kind guys: why I can't access midjourney's channel?

#

It just can't click in, but it's ok for other channels

clever venture
#

I wonder when will people make ai less chaotic

naive bear
#

All I wish I had was some sort of pose curation AI, or at least a model that understood any type of pose and not just the typical basic poses.

#

Even throwing a bunch of ControlNets together with Depth, OpenPose, and Canny, it doesn't make it possible for current SD models to make the poses the characters are in if the poses and angles are so foreign to it's training data.

tribal zenith
#

Hello.. is there a mobile app for illusion diffusion ai?

umbral spade
#

What's the difference between Checkpoints and Loras?

solar condor
#

hey do any of you know about dreambooth on mac ??

stark hornet
#

if i want to use comfyui with a v1.5 model, what node do i need to take ? "CLIP Text Encode (Prompt)" ?

plucky shoal
#

if you have default Workflow. Try just load 1.5 model and it should work

charred thistle
#

i do need to organize comfy more, and make use of grouping so I can turn entire clusters of stuff on and off depending on my needs

solar condor
#

guyss how can i train SD with my style on a mac, dreambooth doesnt seem to work..

queen goblet
#

Hi. Is there a support team here? I'm not sure if there is a network error or if my image outputs/prompts have caused me to be locked out?

plucky shoal
#

Try probably ask in tech support or prompting help but be more specific.

queen goblet
dim oxide
#

hey people, i'm searching for a working discord bot to use with my stand alone auto1111 installation... there is literally zero bots that work out of the box or i'm crazy? everything either trying to download their own model or some shenanigans ...

violet fern
#

Someone know if its possible to somehow detect faulty images, where body proportions are off or other errors in the image? Like, to generate a lot of content and the filter them sort for the best results.

charred thistle
#

@violet fernthere are some tools for fixing faces after generation, but, when it comes to faulty images, the go-to solution to them is inpainting

violet fern
#

I mean more alike a process where the AI detect IF it has errors, not to improve them.

fallen loom
#

Hey I am brand new to all of this so I’m sure I’m missing a lot, I have everything loaded in but when trying to edit Image to image with paint in to replace I painted area I get basically nonsense instead of the AI looking at the rest of the picture and replacing the effected area with something that also makes sense, am I missing something obvious?

charred thistle
#

probably not as well as you can detect errors with a quick glance

violet fern
#

So it is somehow possible?

vast ingot
#

you can find extensions for automatic that do aesthetic scoring

violet fern
#

Thanks thats what i was looking for!

fervent thunder
#

is there an explanation for bizarre mistakes such as drawing the armor but without the body? I wanted an armored dragon flying, somehow the AI made the armor with no dragon. An empty armor flying. what?

potent spire
#

@fervent thunder did you try generating the dragon first and then the armor?

#

inpainting

strange karma
#

hola

#

soy nuevo aqui

fervent thunder
#

how do I do inpainting?

grand sail
#

I have returned

lost relic
#

Hi I was thinking about using midjourney or stable diffusion for comics and I was wondering which is better?

charred thistle
#

well stable diffusion gives you more control and there's no cost or limitations

weary dagger
#

Are there any promo codes available just now?

sly eagle
#

Hello! do you know what is the best way to train a art style to use it for games, comics, or other media? with Lora or checkpoint?

idle umbra
#

Hi guys, how do I generate a new image based on an already existing image? For example, how can I upload my own photo and generate it in a different art style?

sly eagle
idle umbra
#

Is it not possible to just upload the image and then tell AI to recreate that image, but in a different art style?

#

Midjourney has it

sly eagle
#

I don´t know if there´s other way

idle umbra
#

Never heard of controlnet, what it is?

#

Is it usable in here? Or just in the PC version of Stable Diffusion?

sly eagle
#

in the pc version

sly eagle
charred thistle
#

heh.. love jacking up skinny prompts, along with muscular prompts, and creating very buff skeletons

#

nice, my A770 will be here day after tomorrow

cedar gulch
#

Someone know how to create a realistic Disney poster?

thorn stirrup
#

hey how do i make a negative prompt?

neat cypress
charred thistle
#

nice... just found out there's a node that automatically sends a finished image over to the input of my "inpainting" workspace... more automation, yay

rough sail
#

The more I see DALL-E 3, the more I cannot wait for the day SD replaces CLIP and begins to close the gap. I mean, SD has the big advantage of being completely uncensored and open source yes, but closing the gap where prompt adherence is concerned would be the greatest thing that could possibly ever happen to SD.

ornate flame
#

Yeah, hopefully SD3 is released before year's end

barren sentinel
#

reminder to not use bing create since ye it great but limited and you cant exemple use cigar or anything similar :C

gritty dust
#

any tips on the dataset of pictures i need for making an embedding of myself. i started just asking my girlfriend to take a picture of me in 1:1 ratio every time we go out to have a diverse set of different backgrounds clothes and lighting in my data set, but not sure how many picture of close up etc i should do and how varied i should make facial expressions and angles.

potent spire
#

And if D3 in the meantime will get some updates as well

vocal hearth
#

I downloaded stable diffusion and there's an error later that shows in the code that says "stable diffusion failed to load" what can be the solution?

vocal hearth
still glacier
vocal hearth
#

Okay thanks..

#

I'll post a better problem statement

midnight crag
warm junco
midnight crag
warm junco
#

Maybe reinstall chrome but save your passwords and stuff

midnight crag
#

yeah, I will try that. Thanks for the help, 😉

warm junco
#

No Problem 🙂

bleak matrix
#

Good morning, everyone!

#

How are we all this gorgeous morning?

fervent thunder
#

You

bleak matrix
#

Good! What are you up to?

forest trout
rose stirrup
#

anyone knows of any lora to put the kimetsu no yaiba uniforms in a character?

fervent thunder
#

can someone generate this prompt cuz bing image generator doesn't work ? "four skeletons in a gas station smoking sigar in fisheye view"

charred thistle
#

hmm... I need to do more optimizations with ComfyUI... what I'd like to do is have a series of "preview image" nodes that populate sequentially, so I can keep my last 4 generated images on screen, or separate them all when I do a batch of four

neat silo
#

hey, just curious, when people use all these ratios in their prompt, what are they doing?

#

eg. (ornamental engravings:1.2)

#

what does the ratio do?

drowsy tundra
#

Hello anyone there? I need a charity outpainting

#

For a youtube banner

forest trout
# neat silo what does the ratio do?

That number applies weight to the text inside the parenthesis. A 1.2 means they want the model to emphasize those tokens 1.2 times the amount of the other tokens.

karmic path
#

how to I img2img in a bot channel?

charred thistle
#

nice.. just polished up by auto background workflow

south veldt
# orchid spade did you got an answer?

well, it seems the whole Clipdrop community got an "answer" of some sort. Clipdrop updated their pricing plans, and SDXL image generating, along with Stable Doodle and Background Replacer are crossed out in the Free plan meaning no more free generating with Clipdrop. These functions are now behind the paywall

charred thistle
#

I'm thinking I might even just create a custom comfyui node which consists of a group of nodes I have linked together that can perform a single, automated task

bleak matrix
#

Good morning, everyone! How are we all this lovely day?

potent spire
#

hello Sunny, im fine besides of pain in my right hand. How about you? 😄

fervent thunder
#

what is the best stable diffusion model

#

and if i dont have the requirements can I still do it

karmic flame
#

What's your VRAM fam?

karmic flame
fervent thunder
karmic flame
#

Google what your graphics card does

#

it should list the VRAM

fervent thunder
#

ok

vast ingot
fervent thunder
#

ok

fervent thunder
#

@karmic flame

karmic flame
#

I have that and it's more than sufficient (except for SDXL which is the latest version)

fervent thunder
#

is sdxl the best one?

karmic flame
#

Depends. If you can't run it easily it's not going to be the best for you. But yes it is better in general.

fervent thunder
#

alr which one should I take then

karmic flame
#

My suggestion: Google how to install automatic1111 GUI -> look at civitai dot com -> search for checkpoints (also called models) -> find one whose art you like -> make sure it's compatible with SD 1.5 or 2.1 -> google how to place the file and how to generate images with automatic1111

fervent thunder
#

oh ok

karmic flame
#

there's a lot to cover but for now try to get a model running

fervent thunder
#

alr

fallen mason
#

whats the best stable diffusion model to download rn?

vast ingot
neat silo
fallen mason
#

or does it like not matter so much?

vast ingot
#

I'll be honest, I'm coming back after a long break and haven't ran it yet, I'm downloading it too currently, and I don't know the requirements, I have 24GB, but I think I eared it could also run on a lot less

fallen mason
#

holy monkeys 24??

neat silo
#

of what? ram?

#

or vram

fallen mason
#

vram

neat silo
#

jesus christ

fallen mason
#

like gpu ram

neat silo
#

that's insane

#

what card do you have

vast ingot
#

I'm on 3090TI

neat silo
#

damn

fallen mason
forest trout
#

SDXL is still the latest official model.

fallen mason
#

it only got 8 though

neat silo
#

I'm running a modest GTX 1660

vast ingot
#

yeah, the range of VRAM is quite big on the 30XX series

fallen mason
#

3X AS MUCH

neat silo
#

lmao

fallen mason
#

goodness

vast ingot
#

price too though

forest trout
#

I've got a 3090.

neat silo
#

I've only got 6 gigs of vram lmao

vast ingot
#

I got that one specificaly for training

fallen mason
#

i have a prebuilt system

#

so idk if u can upgrade those

neat silo
#

I hate prebuilts, they feel icky idk why

fallen mason
vast ingot
neat silo
#

free?

vast ingot
#

most of it being trained from art on this very discord

neat silo
#

oh i guess if you're actually training models then 24 gigs might be reasonable

fallen mason
# neat silo free?

yuh my laptop broke so they gave me a voucher of £1200 so i got a desktop instead

neat silo
#

but for generation that is waayy overkill

neat silo
fallen mason
#

IKR

#

what do u even need 24 gb for anyways?

#

most things u can even do dont require that many gb's do they?

vast ingot
neat silo
#

mhm

fallen mason
#

whats the difference between normal stable diffusion and sdxl?

vast ingot
#

the size, number of parameters. How it was trained... a lot internally
From the user point of vue : the quality of the results, and the requirements

fallen mason
#

will i have to download different models than the ones i was using before?

#

that was a very time inducing process

vast ingot
#

it's the main difference, yes. there is one model for SD 2.1, and one model for SDXL
hell, there are hundreds or thousands of models if you're into that

fallen mason
#

mmmmmmm

#

yk what its probably worth it considering the difference in quality

neat silo
#

quick question, what is a pickletensor file?

forest trout
#

Some models will require a second file with the same name that will determine some internal settings that need to be applied to use that model.

fallen mason
#

i think the last model i used was 1.4 so im hoping alot is different

warped terrace
#

hi guys , anyone can help me if i stream ?

#

would be much appreciate

vast ingot
warped terrace
#

i just installed stable diffusion

#

my first pictures generated with model are so ugly !

#

i guess something is wrong

vast ingot
#

sorry I can't, but welcome around

warped terrace
#

Thank you .

grizzled harness
#

What does "ugly" mean?

neat silo
#

does that mean this file is potentially unsafe? I mean I got it from civitai so idk

forest trout
#

If it ends in .safetensors you are ok

vast ingot
#

civitAI runs a basic unpickle test, to prevent from virus. people can also report dangerous files. They do not guarantee anything though
This indication just means it's an old file usually, not necessary dangerous
but if there is a safetensor alternative, take that one for sure

forest trout
#

If it ends in .ckpt it's probably fine, but you should try to find a .safetensors version.

vast ingot
#

there are also spaces online that can convert a ckpt to safetensor, but this is usually more for the people that do make the models

#

it would require you to upload the model for that

neat silo
stray dawn
#

Not sure if this is the place to ask but is there a list of "Negative prompts" that are used regularly to get good results?

fervent thunder
#

yo

#

where do i ask quiestions

charred thistle
#

whelp, UPS just dropped off my new GPU

latent lichen
#

has controlnet been removed from the bots?

warped terrace
#

if anyone can pm me , i need help with stable diffusion , would be much appreciated

olive mica
#

Hello everyone, is there an option for inpainting available here?

midnight crag
#

Hi guys, is there a specific place thread to ask about comfyui?

lavish lake
fervent thunder
#

Does CPU matter at all for stable diffusion

charred thistle
#

@fervent thunder not if you're using GPU to do all the work

fervent thunder
#

for IT/s i mean

charred thistle
#

what's your GPU and are you using that for SD?

fervent thunder
#

1080ti rn but i'm getting a 4090 soon

#

like, next saturday

charred thistle
#

then I don't think a change in CPU is going to net you any noticeable difference

fervent thunder
#

really only care about stable diffusion performance as games dont benefit from 16 cores vs 8 and since i'm playing on 1440p a faster cpu really doesn't do anything

#

looking for a reason if its even worth changing to am5 yet

#

1500€ for a new base is a lot of money compared to additionaly getting the cheapest 4090

fervent thunder
#

i want to train loras

#

something thats a pain in the ass on my 1080ti, takes over 2 hours to do a proper one

quiet storm
#

gm frens ive arrived

fervent thunder
charred thistle
#

well now, seems I stepped in some shit... installing comfyui for intel arc

quiet storm
#

so who should i follow on yt for learning?

#

like actual boyos no click baiters

charred thistle
#

that might depend on if you use automatic1111, comfyui, etc

quiet storm
#

i dont know what any of that means

#

i got stable diffusion? installed

#

ok i think i found my senpai, Olivio Sarikas

charred thistle
#

stable diffusion is a command line tool, if you want a graphical interface, you use something like automatic1111 webui, or comfyui

oblique summit
#

hey can someone help me understand how to update controlnet

warm junco
charred thistle
#

I broke my operating system

#

But maybe it is for the best

oblique cipher
#

Has anyone used animatediff, please let me know. Why I've installed animatediff like in the picture but comfyui still doesn't recognize it??

fervent thunder
#

Hello)

charred thistle
#

nice.. thing I got torch 2.0 working with intel arc optimizations now

charred thistle
#

wow... full 20-step 512x768 images in 9 seconds

#

4 seconds per with DDIM

#

vs over 40 seconds before.. so yeah, 10x speed upgrade 😄

lavish lake
charred thistle
#

dude... 1.30s/it for 896x1152 SDXL

magic falcon
#

Does SDXL + Controlnets (IP-Adapter, depth etc) work when using TensorRT in A1111?

robust tide
#

Hello, I just installed SD last night on my 1080Ti

#

Is it okay to generate images of my favourite kpop idol?

#

It would just be for personal use, but the law is confusing

lilac basalt
#

hello, any pointers on where I can find negative embeddings for sd2.1 ?

charred thistle
lilac basalt
charred thistle
#

You can filter results for 2 only, but it's also newer so there won't be as much stuff for it over all

lilac basalt
#

alright, thanks

#

i guess i'll follow instructions to create one myself

compact nimbus
#

what are the like licensce terms of images you make?

fervent thunder
#

Is there any way to enable NSFW images?

My friend wants to know.

warm junco
fervent thunder
#

No I mean on the platform of the photo generation

fleet wren
#

Who has an RTX3070 please respond.

plucky shoal
#

i have @fleet wren

noble bronze
warped terrace
#

guys , is it possible to turn a picture of myself into an anime character ? i mean exactly the same picture

woven forum
#

yes image to image

frigid depot
#

Can someone help with this?

NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

#

How do i use the --no-half commandline? Where do i enter it?

stable pawn
#

I feel stupid again, I have to reinstall Automatic 1111, but after a stroke I had a a year ago things sometime fall out of my memory, this time it was something about argument lines, I remember "--xformers --autolaunch" but then a huge blank; I have a Rather okay Intel with a 3090 how shall my "set COMMANDLINE_ARGS=" look like?

karmic pumice
karmic pumice
stable pawn
karmic pumice
#

it's like long prompts, it's mostly useless to have tons of args if everythings working fine

stable pawn
warped terrace
#

@stable pawnwhen i lauch it , in the console command it says xformers not found , running without it

remote glade
#

yo I have a question

#

Does mov2mov have a size or length limit?

charred thistle
#

getting Intel Arc optimizations working on Arch Linux was actually extremely simple

fervent thunder
#

guys does anyone know how to change camera position. Im trying to get a shot from underneath so you have a crazy perspective

charred thistle
#

could try low angle, maybe "shot from below"

fervent thunder
#

lemme try

#

doesnt really work, maybe its because i have a pretty large prompt

charred thistle
#

you could try adding perentheses around the prompt to give it more weight, like ((something))

fervent thunder
#

i have 3

#

thanks anyway

fallen mason
#

how long is a pic using sdxl gonna take on average to generate?

#

i got 8gb vram and it taking years

warm junco
warm junco
fallen mason
#

right yeeee i got no half vae on but i thought xformers made it worse so i got rid of it

white light
white light
warm junco
fallen mason
#

sowwy

fickle quest
#

Hi i need help I just installed stable diffusion with realistic vision but it only generates two or more people instead of one when i put a "complex prompt" it doesn't matter if I tell it that there is only one character alone.. I don't understand

floral umbra
#

Holy fuckin shit! I just tripled my generation speed!

#

Nvidia finally released their tensorRT extension, and i went from 7.36 IT's to 20!

fallen mason
#

is lora file supposed to end in safetensors?

quiet storm
iron pollen
#

The constant generation of waifus on this discord is beyond unfathomable levels of cringe

slate dust
fallen mason
# slate dust yes

for some reason when i downloaded a lora file thing and i put it in the folder, it didnt appear

jolly briar
#

Hey everyone, does anyone know if I can use my own z-depth image in Control Net or other plugins? I'm looking to generate some textures on top of my 3D scene, but the z-depth produced by Stable isn't suitable for my needs

hasty hornet
slate dust
quiet storm
#

any good content creators out there? to learn and keep up with SD?

crimson marlin
#

Does anyone know a good model for making anime images more shonen-like (creating characters like ichigo n shit) and not hentai like every single anime checkpoint i find on civit?

hard elk
#

Hi, would you know which is the best model driven by microscopic images? I would like to generate some thank you

storm relic
#

Good day! Could anyone tell me what is the best way to start making AI generated portraits using real people images? How to make face super realistic?

charred thistle
#

man.. now I have to go through the task of rebuilding my SD workflow

golden veldt
#

who wanna start with dropshipping?

grave stag
#

This still only works on Nvidia and CPUs right? Also I know with at least my 3060 ti it runs out of Vram so likely anything with 8GB of ram will not preform better so if I had a 3070 ti or a 4060 ti (not to mention the slower buss speed that also makes things worse) it won't do any better and be faster right? Then for the 3080s would I fill Vram using a 3080 12gb or would there be a possibility of better performance with a 3080 ti?

#

Does this logic and process even make sense?

charred thistle
#

@grave stagthat would be false. Earlier this week I was using an RX590 AMD, now I'm using an Intel Arc series

grave stag
# charred thistle <@975848561440878642>that would be false. Earlier this week I was using an RX59...

Did they update it or something? I thought it used to only run on Nvidia? Or at least nobody would say to buy a GPU other then Nvidia for this. If I consider an upgrade from my 3060 ti (though I might stick with Nvidia) would an equivalent AMD card preform as well as one from Nvidia? I think AMD is cheaper on the used market? That and ark might have the cheapest price for high Vram cards but there is no upgrade from my 3060 ti that can be Intel.

charred thistle
#

The RTX3060, if it has at least 8gb is actually pretty good for SD, the 4060 didn't provide that much of an upgrade over the 3060... I bought an Intel card with 16gb VRAM and I'm liking it

grave stag
#

I have the ti card, I could buy an ark but then it be a downgrade for anything that doesn't need more then 8GB ram (I think there is no arc card better then a 3060 ti right?) So then that's AMD or Nvidia and does it run better on Nvida vs AMD?

grave stag
#

Especially since I also use the PC for gaming.

grizzled harness
#

Nice sale on frames at Michaels right now.

#

Not sure if anyone prints/frames any generations. 🙂

charred thistle
#

@grave staghow much VRAM do you have now? And where do you feel it's lacking in SD?

grave stag
# charred thistle <@975848561440878642>how much VRAM do you have now? And where do you feel it's ...

I make a few images (not that I know how to write image prompts well enough to get results I want and like) and I spike to 100% Vram usage. I tried just using this to learn by starting to make the exact image shown in the examples and I can't even match that or the anime character right but that said it takes like double digit min for like 4 images if I turn up the complexity and resolution any amount its annoying.

charred thistle
#

@grave stag Automatic1111 or Comfy?

boreal dew
#

What's the major difference between A1111 and Comfy?

#

I've been hearing COmfy is faster?

charred thistle
#

Comfy doesn't eat up a bunch of VRAM

#

I was doing 1024x1024 SDXL on an RX590 with 8GB VRAM

boreal dew
#

WHat does it eat then? There's gotta be a caveat in it somewhere

grave stag
charred thistle
#

or it's just more efficient of a UI

grave stag
grave stag
charred thistle
#

it just uses nodes, pretty easy to figure out, and more versatile

boreal dew
#

Huh... Might give it a shot on my laptop. My laptop has a 1660ti in it, and it's slow

#

I want speed

charred thistle
#

my Intel Arc GPU now does 512x768 images at 20 steps in 4 seconds 😛

boreal dew
#

WHen I move from this 3060 back to my laptop, I'm gonna be having one hell of a time

grave stag
boreal dew
#

Also, I know this is more of a question for tech support, but I really didn't get any definitive answers, does animatediff take a long time to render videos on a 3060 or is there something wrong with my get up?

#

I feel like it shouldn't be taking an hour and a half

#

with 100-130s/it
I should mention google isn't helping me, I keep getting answers on how to animate irl things

charred thistle
#

@grave stagA770 has 16GB VRAM

#

but it's not really the VRAM making it that much faster, it's a lot of the other stuff, like having dedicated AI cores

grave stag
grizzled harness
#

Random question -- in auto1111, can the x/y/z plot be used to render across different negative prompts?

grave stag
charred thistle
#

they'd both be pretty close until you start doing stuff that actually uses more vram, like larger batch sizes

grave stag
#

Idk I might get back into this with the GPU I have then again it doesn't make sense to have a 3060 ti desktop and a 4060 laptop might as well be the same.

All I want is to learn how to use this to make OCs for RP

charred thistle
#

yeah, try ComfyUI first, it might be exactly what you need

grave stag
#

How many batches is using a lot of Vram?

obsidian adder
#

Is Img2Img possible using these bots? Or just on the website

grave stag
charred thistle
#

when I was testing it out, I think Comfy was using an addition 800-1000mb per image added to the batch

grave stag
#

So 1gb per image so 6 images and I'm out of ram?

charred thistle
#

if you were doing them all at once maybe

#

sequentially there wouldn't be a limit

grave stag
#

How long you think it take to go from prompt to 10 good resolution images based on prompt?

charred thistle
#

i wish I had a tool that let me see how much VRAM I was using making these 1920x1080 SDXL images

grave stag
charred thistle
#

nah

#

there's a tool to check intel GPU usage but it doesn't seem to understand values for hte Arc GPU

grave stag
#

Its been a while since I've used this and I don't even have this set up on my laptop, how long you think it take to learn how to make OCs so I can stop paying people for refs?

#

@charred thistle think you can teach me how to use this again?

charred thistle
#

comfyui should start out with a basic setup needed to generate stuff... you need to have checkpoints installed in the models/checkpoints folder to load up.. one box for positive prompts, one for negative, the tool that processes it all in the middle and an image output

grave stag
#

Also if I am upgrading from 3060 ti for this and just it doesn't make sense to have a 3060 ti and 4060 laptop what GPUs should I be looking at? Will AMD preform as good as Nvidia?

charred thistle
#

it doesn't seem to be like there's much to be gained going from a 3060 to a 4060

#

AMD should get good performance if AMD supports ROCm with that card, which for new ones they will

#

if you have comfyui up and running, the github page for comfyanonymous/ComfyUI_examples has a bunch of example workflows for doing a bunch of tasks, like inpainting, upscaling, how to use Lora

grave stag
grave stag
charred thistle
#

you would need to have everything exact if you want a match.. meaning the checkpoint, any loras used, both prompts, image size, cfg, number of steps, the seed, etc

grave stag
grave stag
# charred thistle you would need to have everything exact if you want a match.. meaning the checkp...

If you want to help me out and help me learn how to recreate the images and make my own OCs that's great but unless I find a teacher and one that can really sit down with me not multi task and sometimes take hours to respond, this just be to complicated for me to learn and dive into. Even though it seems so promising to be able to make my own OCs for my rps and not just paying people a small fee to make the better or more important ones on discord.

charred thistle
#

I'm still fairly new myself.. about to try merging concepts from two different images and see what happens

grave stag
#

You seemed knowledgeable about this with how much you know for GPU specs and what to run for this and such?

trim magnet
grave stag
trim magnet
#

if u have a 3060 ti i think thats enough for now if u only gen images

boreal dew
#

Okay, so before I start the conversion to comfyui, is there anything I should know about it other than the fact it looks confusing

#

liek how much faster is it compared to A1111, is it slower than A1111, is it bad, is it good

grave stag
# trim magnet if u have a 3060 ti i think thats enough for now if u only gen images

I mean a better card can speed things up and I'd keep this card for a while like maybe years down the road and it doesn't make sense to have a laptop basically as powerful as my desktop, if I can get a 3080 12gb at a good price I might upgrade to it. Thing is 3080 12gb be basically the same price as a 3080 ti that be just under a 3090 that be I wish I could say just under a 3090 ti but no that be way more expensive.

trim magnet
boreal dew
#

Is it lighter or heavier than A1111?

grave stag
#

Idk I'll look for 3080 12gb deals, they might as well be 3080 tis in price though witch is strange and that said 3080 tis at least be a little less then 3090s even if it be pretty close and if I'd buy a 3080 ti I'd buy a 3090.

boreal dew
#

that should've been my first question lol

trim magnet
trim magnet
#

but some extensions that are in auto1111 cant be used in comfy

boreal dew
#

That's fine, I don't have many extensions to begin with.

#

I just want to be able to animate.

#

and it not take an hour and a half.

grave stag
trim magnet
#

and yes they have rtx and dlss

grave stag
trim magnet
#

yea its probably very simmilar to a 3060 ti in gaming and yes they have all the encoders,it has less cores because its turing and 30 series is ampere

grave stag
trim magnet
grave stag
trim magnet
#

you would only buy a quadro if u want lots of vram at reasonable price,if u have 600$ or more just buy an used 3090

grave stag
trim magnet
grave stag
grave stag
trim magnet
#

yea if i had 500 id go for a RTX A4000 16G

#

have u tried with an used 2080ti 11gb that one i have seen recently as low as 300$

grave stag
#

I have $400-$450 if that I'd like to go under $400 if I can. I thought since the price a few months ago was this price now I'd be able to get a 3080 12gb for this but I can't I can't seem to get them even though other cards like the 3060 ti seem to have gone down in used price by $50-100.

grave stag
#

Its very much a good value at that price but I doubt it be better then my card if my card goes for $250 on eBay.

#

If it is better it won't be that much better.

#

I just want to get some card with more Vram that is actually better so it doesn't seem like my laptop and desktop have the same GPU and I'd have reason to switch to my desktop for use.

trim magnet
grave stag
#

You said a 5000 be as good as a 3070 and this be a 4000 so it be worse then my card other then some A.I. stuff due to the Vram.

trim magnet
#

the a4000 has the same specs as the 3070 ti only difference is that 3070 ti has gddr6x vram and the a4000 has gdrr6

#

and the extra 8gb vram of course

charred thistle
#

I'm happy with my 16GB A770 I got for under $300

#

just sitting here doing 1920x1080 sized SDXL renders

grave stag
#

Also what does the x mean for vram anyways?

trim magnet
#

slighty faster in vram intensive tasks

#

the 5000 is similar in gaming performance to a 3060 ti but overall the 3060 ti its better because its newer generation so more cuda cores, u would only buy the 5000 if u get it cheap at around 300$ and if u want it for mostly ai

#

the 4000 is better because its from the same generation as 3000 series but its more expensive so u would have to camp ebay until u can get one for like 400$

boreal dew
#

Okay this may be a really dumb question, but I have to ask... can I run comfyui off a flashdrive?

#

a 3.0 flashdrive in a 3.0 slot

#

it seems to be running, like loading in the cmd window, but it seems to be slow.

#

I'm curious if this is because this is a first time boot.

#

Okay! It works great actually.

fervent thunder
#

the name is "murder" and art is a dead person with a sword

#

do I have to use local processing then?

remote glade
#

Hi uhh

#

Can i get the invite link for stable dreamer bot

#

So i can add in my server

chilly cedar
#

Hello!

charred thistle
#

Nice, I think I might have actually aided in a new guide being updated for using Intel Arc GPUs with SD. It seems to be much simplified and the Intel optimizations run on Python 3.11 that really only entails a copy and paste into a command line to apply and should work for any Linux distro

glossy mountain
#

Hi guys, is bombs or dynamites a forbidden word for the bot right now? Trying to make a Joker, but seems like there's never a bomb.

charred thistle
#

"bot" runs locally and I'm not sure if it having any censorship, but the way models are trained is they need to be fed images in order to understand certain things, and that doesn't sound like something a lot of people train for

glossy mountain
#

Sorry, i meant the sdxl beta bot accessible through discord.

charred thistle
#

aah.. well then what I said is even more true, since newer models are likely going to have less polish to them

robust tide
#

Can you trade time for higher quality images?

#

If I was willing to wait an hour for an image to generate, could I have an absurdly high quality one?

warm junco
#

Quality is not defined trough steps or time

robust tide
warm junco
#

100 steps wont give you better quality

warm junco
#

Steps are = how often should something change in the image

robust tide
#

I'm getting way better outputs with 75 than 25

#

Much more photorealistic asian girls

bleak matrix
#

Good morning, everyone! How are we all today?

robust tide
#

It's so repeatable, I'm sure it makes a difference

robust tide
bleak matrix
robust tide
#

I can't think of any business use for SD or anything else I can use it for other than producing pics of asian girls. It's been an interesting way to harness the power of my GPU now that it's not mining ethereum.

#

I'd be very interested to know if anyone has used SD for their business.

robust tide
vital void
fathom hatch
#

hey to all.anyone using automatic ui on apple silicon ? how is the performance ? i did some digging and i seen people recommending diffusion bee and trying it now but i also wonder if there is any software to run sdxl on apple silicon ?

minor smelt
#

I am so intrigued by this community and technology. Considering trying to run and install on a VM with NVIDIA Quadro M4000 -- good idea, bad idea? If reasonable, is it best to follow the automatic guide to set up?

remote glade
#

I feel like you will make sussy stuff

woven forum
#

Amongus 80085

fervent thunder
#

Hi guys -- do you all have a recipe for installing stable diffusion on Ubuntu 22.04? I see a bunch online, prefer to use yours if there is one so I get it right?

woven forum
#

depends on which UI you want

fervent thunder
#

something fairly standard I guess -- didnt know there were different options

woven forum
#

install conda first

#

I'd go with comfy, easy to install, very extensible.

fervent thunder
#

works for me

woven forum
#

conda should be straight forward to install on ubuntu.

fervent thunder
#

conda sorta like pip?

woven forum
#

No it's more like docker.
You can make it have certain environments with specific packages.

fervent thunder
#

ohhh -- ok

#

my ubuntu box is normally locked away from the web, but accessable on the interior network -- so I hooked it to the web and am doing the normal upgrades right now

#

so conda, then comfy?

woven forum
#

Yeah

fervent thunder
#

does comfy install all the background code needed and the models and such?

woven forum
#

comfy you basically just clone the repo, activate the conda environment, then do pip3 install -r requirements in the comfyui folder

#

then you run main.py and connect to it through a browser.

#

You can make it a server with --listen and connect from another computer

fervent thunder
#

thats ideal

#

i might have a few questioins -- working on it

minor smelt
dusty tree
#

do any of you know any good ai tools for translation?

charred thistle
#

Hmm, I have my Intel card set up to use ComfyUI with the tutorial shown on the ComfyUI Github, benchmarking it to compare to an earlier benchmark I took of a simpler process to see if all the hassle of installing and using "intel for python" and other optimizations are worth it

quick yew
#

tried to install roop for webui and now it doesn't work even without the roop extension

#

😦

charred thistle
#

roop?

quick yew
#

an extension for face swapping

#

now my webui doesn't start

charred thistle
#

does it start if you remove roop?