🏞｜general-with-images

[cobalt pollen]

AI now able to do bad photo shop jobs, excellent

[nimble geyser]

let's see yours, wally

we're waiting patiently

[cobalt pollen]

wait a few months until i have my new laptop xD

[sterile kiln]

It's a double saber, lit on one side only 😉

https://tenor.com/view/think-smart-gif-22411928

[cobalt pollen]

needs some fishes eating other fishes early 2000s gameplay

[nimble geyser]

perhaps one day

[prisma stone]

Anyone have a prompts that I can generate this character?

My models and my prompts cannot correctly generate an image of him

And my friend requested me this too

I don't wanna fail them

[cobalt pollen]

have you tried textual inversion?

[stoic pivot]

One fun thing is to use "dressed in ?????? suit", here is a man dressed in "tight unicorn suit", "zombie suit" and a "deer suit".

[mystic marlin]

For Baldur's Gate fans. Shadowheart herself. lol

[noble zephyr]

Been using SD for a little less than a week now, thoughts?

[obtuse palm]

Comparing 3 different upscaling models.. which one you think did the best?
you may need to download and zoom in to see difference

[stoic pivot]

This can be one of those old RTX ON/OFF jokes (Second Life - img2img_AI)

[nocturne oak]

Smooth areas look better in realistic rescaler, detailed areas are enhanced by ultrasharp. But swinir seems to be a good compromise of both.

[prisma stone]

Nope

What is that?

[heavy moth]

too basic honestly

[noble zephyr]

What would you do to improve it?

[heavy moth]

I would make it not too basic

[noble zephyr]

Yeah but how? Add more detail?

[heavy moth]

🤷‍♂️

this is generated with one word; "fantasy"

[jagged garden]

Hi guys, my models have often a nose piercing.. a very ugly one. it never looks good. Even though i don't write anything about piercings in the prompts.. How can i get rid of this?

[proud dagger]

probably remove the mile long negative prompt, and use the negative prompt box to remove things you don't want? 0.o

ps [] de-emphasizes a word, () emphasizes, I assume you mixed that up because de-emphasis of 5 is basically just an ignored word. I don't think you'd be allowed to post the image if you hadn't messed that up though.

[jagged garden]

wait.. can you see what prompts i typed in?

[proud dagger]

yeah

it's stored in the image metadata

[jagged garden]

lol.. but long negative prompts are good. stop lying about that.

[jagged garden]

what does that mean.. i don't understand 😢

[late adder]

#1072220168534642768 the tarot in the looking glass, behind it a mirror maze from which the senior arcana of the tarot look out

[split rover]

Truly a work of art

[prisma stone]

Nothing in Stable Diffusion cannot generate 2 people hugging

[wispy nest]

Can you explain this

And why you did it

[obtuse palm]

Just a comma difference.. if you put at the end of the last prompt

[fallen dock]

looks oddly familiar

[nimble geyser]

Blip and clip

[wispy nest]

How comma this happen

Is she like

With the comma looking more at the viewer and has warmer lighting?

I'm too stupid to see the difference

[obtuse palm]

the comma at last doesn't do a big change but it definitely does a subtle change

i like her forarm band though.. using the comma

you can experiment yourself.. you'll get two different results using it so why not!

[fallen dock]

one with last comma looks better, lol other one has artifacting in the hair, lol

[obtuse palm]

yeah comma makes it better sometimes, and sometimes worse

[wispy nest]

That's literally something from those "spot the difference" games

Haha

[fallen dock]

interesting username, lol

[wispy nest]

[obtuse palm]

it's fun to experiment

[wispy nest]

I usually put comma at the end

[obtuse palm]

comma has a special value within the prompt i guess

[fallen dock]

However, I've never used commas to separate Lora model references

Since I unusually put the reference at the end anyways

[wispy nest]

I always do

It looks like something I don't need to

But it gets on me if I don't separate everything with a comma lmao

[uncut zinc]

How can I take a 2d image and make it 3D rendering? I've tried img2img with low denoising (0.1-0.5) and it is barely changing the style. The higher I go, the more other things change but the style is not changing much. I really need the details of the image to stay together as much as possible and just become 3D instead of a flat 2D image.

[fallen dock]

Try using a different Checkpoint?

Maybe merge some models?

[obtuse palm]

need help

if an image is generated full body for others but only half body for me for all same settings then what could be the reason?

[nocturne oak]

There is some other setting that differs. Could be nearly anything.

[heady anchor]

or maybe they used a controlnet pose

[nocturne oak]

I mean...nearly anything at all can cause differences. Different seed, different CFG, different prompt, different VAE, different anything.

[obtuse palm]

they only used text2img+hires.. not sure if controlnet was involved or no
Here's the data:
1girl, photo of nude model 18 y.o, tanned caucasian 18 yo, situated near Kilimanjaro in Tanzania, final fantasy still, extremely high quality RAW photograph, detailed background, intricate, Exquisite details and textures, highly detailed, ultra detailed photograph, warm lighting, 4k, sharp focus, high resolution, detailed skin, detailed eyes, 8k uhd, dslr, high quality, film grain, Fujifilm XT3,
Negative prompt: FastNegativeV2
Steps: 20, Size: 512x768, Seed: 2680399189, Model: aniverse_V12, Version: v1.4.1, Sampler: DPM++ 2M Karras, CFG scale: 7, Model hash: 4393bb4b9b, Hires steps: 20, Hires upscale: 2, Hires upscaler: R-ESRGAN 4x+, ADetailer model: face_yolov8n.pt, ADetailer prompt: "photo of a blonde girl, (film grain)", ADetailer version: 23.7.11, Denoising strength: 0.53, ADetailer mask blur: 4, ADetailer confidence: 0.3, ADetailer dilate/erode: 4, ADetailer inpaint padding: 0, ADetailer negative prompt: BadDream, ADetailer denoising strength: 0.25, ADetailer inpaint only masked: True

[nocturne oak]

If there are different versions of that model, you might want to try a different version.

Unfortunately, the information showing on Civit is A) not complete and B) not always correct.

[obtuse palm]

i used v12 and v13 and all are coming the same way for me.. half of the body

[nocturne oak]

Sure, but they could easily have done something else that isn't explained in the generation data summary.

They may have used ComfyUI and you might be using A1111. Without the proper care, the images will be different.

Hell, I've got images on Civit with wrong generation data simply because Civit doesn't have the 3M samplers available in the dropdown.

[obtuse palm]

ooh so those examples are pretty misleading.. i see

[nocturne oak]

I'm not saying that specific generation data is wrong, I'm only saying that it's possible.

[obtuse palm]

disappointing huhh

thought i could learn from them but no i guess

i'll have to scratch things out myself

[nocturne oak]

You have to understand how this stuff works. Basically, the rectangle created with the dimensions you give it is filled with random pixels...the latent noise. The sampler & schedule will then essentially "infer" the image via the prompts over a series of steps. (I'm simplifying here just to speed this part along.) So anything that changes any part of the latent noise, the calculations, (or anything, really) can change your image in either dramatic or subtle ways, depending upon what changed and how.

[obtuse palm]

could be a reason with different automatic version? like he used 1.4.1 and i'm on 1.6.0

[nocturne oak]

I haven't toyed around with differences in A1111 versions to see if something like that changed generation output, but I'd assume there's some possibility of that, yeah.

[obtuse palm]

probably.. who knows..

[obtuse palm]

which looks more correct?

[obtuse palm]

and these two

pompt was: (best quality, masterpiece, colorful, dynamic angle, highest detailed)upper body photo, full body photo, fashion photography of cute girl with iridiscent pink hair, in lush beach, gold bikini, (intricate details, hyperdetailed:1.15), detailed, moonlight passing through hair, (official art, extreme detailed, highest detailed)

i think second sets of pics followed the prompt more accurately

but as there's body upper body and full body prompts given so i'm not sure which is correctly done

[blazing kettle]

i cannot even tell the difference im not gonna lie

ill say

number 1

because the extra hair

[fallen dock]

nah it's the hand in the shadow

[south falcon]

how to

[kind quartz]

did you install some extension recently?

[dusky tree]

Why posex ui not showing ?

[fallen dock]

first off, you don't even have "Send this image to ControlNet" checked, then make sure ControlNet is even enabled before hand, and then make sure you have "canvas-zoom" extension.

[grave sparrow]

Hi, visual maestros! I'm working on a project that requires 3D camera tracking, replacing a floor, and inserting a few objects into the scene. I tried using stable diffusion/controlnet and applied EBSynth for animation, but the outcome wasn’t satisfactory. I'm considering using After Effects with Mocha for planar tracking but am unsure about masking around columns. Could you recommend a quick and effective approach or software solution for this? Thanks!

[fallen dock]

at that point just 3d model something in Blender, lol

then you'll get exact reflections, lmao

[fallen dock]

took about 10 minutes to make mate

Also SD can benefit from 3D modeling as well. If ya don't want to make just a 3D animation and bother creating all the textures, UV maps, etc., depth/normal images can be used to help with animations for SD to render. (Using ControlNET of course)

[grave sparrow]

Wow! Thank you for the advice! I'm comfortable with 3ds Max, but I'm not up to date on making a 3D scene from the source footage these days. I've tried using the aeTransfer_v2.1 script to transfer the scene into 3ds Max, but the camera doesn't seem to match what's in After Effects

Here's what the scene looks like in After Effects and in 3ds Max https://drive.google.com/file/d/14W1sDn0lIWIPT-_uAVKiEbkDPyBPQiOU/view?usp=sharing

[grave sparrow]

Seeing that you managed to recreate something similar in about 10 minutes is impressive! Would you be willing to assist me in building the 3D scene? I'd be more than happy to compensate for your expertise and time 🙏

[fallen dock]

Unfortunately I've not ever used 3ds Max, I have used Maya and Substance Painter but for a short time back in community college, lol. Pretty much all paid/subscription based 3D software I don't own myself. In summary what I would do is create a 3d scene very similar to the real footage like what I showed you, render a normal pass (just a normal render) to get similar floor reflections, lighting, etc, then another pass rendering the floor white and everything else pitch black to use for alpha channels (so you don't have to manually mask the pillars over the ground again). That's just coming from a mostly blender user's point of view (like me) but that's probably what I would do. I can't help since I've already got alot of other projects going on but yeey.

Same kinda technique used for fluid simulations on real people and real scenes.

[grave sparrow]

Thank you for sharing you ideas! Initially, I wanted to leverage the Tokyojab technique with Stable Diffusion Controlnet and EBsynth, generating 4-8 keyframes for EBSynth. Post achieving a desirable result with EBSynth, the aim was to derive masks for the columns, possibly through ControlNet Segmentation preprocessor or Mocha's planar tracking. Your mention of rendering a normal pass for accurate reflections and lighting, followed by another for alpha channels, was also something I've been considering. While I have my concerns about closely replicating the original arch while constructing it in 3D, your input gives me some direction... but still - I'm currently concerned about how close I can go to replicating the original arch in 3D

[sterile kiln]

Hey guys,

Will one day it be possible to use LoRA made on 1.5, in SDXL?

[limpid lichen]

nope probably never.

[limpid lichen]

lets simply it. imaging like, sd1.5 is a bunch of random dots on a grid that make up one pattern. and you make a new pattern by adding dots to that grid. so when you overlay your dots, your intended pattern shows up. but now put your dots over another entirely different set of dots. it'll be meaning less in that context

loras are basically a set of numbers that change the original model's numbers. they only work if they match up to the same architecture

[fallen dock]

brub

[limpid lichen]

its supposed to be spock an will smith but i think theres a mixup

[fallen dock]

Icecold what's with the Star Trek distorted dance spam, lol

[limpid lichen]

this nails it. dead on.

[limpid lichen]

i just discovered visioncraft software. ui for animatediff

[fallen dock]

[limpid lichen]

takes about a minute for 3 seconds. i can use any 1.5 model

opening his n64 on christmas morning

he's so stoked

[limpid lichen]

@heady anchor this ones for you

i think hes turning japanese though

[heady anchor]

he looks like one of those chinese generals from those tv shows 🤣

[earnest solar]

Nice to know that the bot is still as bad at composition as ever.

[limpid lichen]

he do be ripped in that one

i put japanese into the negative. run it again!

[limpid lichen]

poly180 why are you spamming mini cars? i keed i keed

annnnd thats all

[fallen dock]

[somber mulch]

I wonder if anybody here has managed to generate a grown-up photo of Jenna Ortega as she looks like now. I guess it's because of the training data but the images major AI models generate of Jenna Ortega are of her as a kid, mainly because that's the only images they have of her

Something like this

[fallen dock]

Just look on civitai lol

[wispy nest]

The AI keeps generating light poles or something similar to alien light sources rather than a city in the distance

[next totem]

@kind quartz the settings im using right now, and the prompts

you can uh... see some of the nightmare fuel im getting out of these on the right

I matched the image generation resolution to the original image as well

im also gonna make note that I am fairly new to messing with stable diffusion as well

[kind quartz]

try lighter denoise strenght or go higher to change image very

[next totem]

lemme go to .3 like you suggested

the lora im using is meant to emulate the renaissance style as well

oh hey what do ya know, moving the denoise down made it almost perfectly recreate the image.... thats progress, it doesnt look renaissance though

[kind quartz]

wouldnt lean on lora. O.3 change image just slightly

[next totem]

Should I get rid of the lora in this case?

the checkpoint im uisng is SD V1.4 full

actually this aint half bad

tbh

didnt realize adjusting the denoise value would have such a drastic effect

[kind quartz]

i would and in img2img description would write renaissance painting or renaissance still life 🙂

[next totem]

you mean up here? or is there another place i can add words for

Im glad you didnt question the thing im trying to make

[kind quartz]

what about do not describe banana? It is in image

[next totem]

I did that at first cuz i was getting women instead of bananas xD that was before i messed with the denoising though

Ill try taking the word banana out and see how it does

[kind quartz]

you have some image putting in img2img.
just put it as image there

[next totem]

Wait you mean just type image in the prompts?

[kind quartz]

no put image in img2img tab. What picture are you putting there? Banana?

[next totem]

this

I addded the description on top of that though because originally even with this image there i was getting a person instead xD

[kind quartz]

yes so no needed describe banana

[next totem]

oh damn

took that out

i think i have a winner

im crying holy shit

[kind quartz]

😄

[next totem]

Thank you for the help i appreciate, ill continue tweaking the denoising and stuff ot see if i can get something i lioke more but this is really good

[kind quartz]

see you!

[next totem]

Yall are here creating beautiful ai art and here i am like

banana

[prisma stone]

There must be something wrong

why his head is still near the edge at the top?

[sterile temple]

does swapping the from and to inputs make a difference?

[prisma stone]

I don't know

[slender prairie]

This should be the same as the ATJ LoRA example pic on civitai - but it looks completely different although all parameters and the model are the same. why?

[languid cairn]

一个中国仙女

[naive copper]

maybe it's not loading the lora

screenshot the settings

[naive copper]

I'm not really familiar with how those "styles" work, but that's the one unusual thing I'm seeing

are there any example images that have simple prompts you can try to recreate

no additional loras or styles

[slender prairie]

I just can generate some simple pictures like a cat with a hat or a mushroom with eyes - but I don't understand what's wrong here. I've deleted the first installation of SD, cleaned the cache and er-installed it - but... no changes

[naive copper]

Don't know to be honest, looks alright other than those "styles" and also the lora appears twice in the generation info for some reason

perhaps one of the "styles" loads the lora a second time

or something weird like that

[slender prairie]

Ok, thank you...

[real star]

/iagin a car pink

[olive egret]

https://www.youtube.com/watch?v=72qmtr41iWU

[kind quartz]

@dense tapir A1111 feature, but feeling not safe with it very.

[dense tapir]

how so?

I see by hand is still needed but the brunt of the labour it did

[kind quartz]

it is automaticaly imagine b&w images, extension, but something must be done in webui-user and i dont believe it so much.
That torn must be done manualy. This is only colorizing 🙂

[dense tapir]

I am not much into colourizing old b.w images

[kind quartz]

yes but i think it is working very well, will try coloring video today

[nimble geyser]

I do believe combining controlnet with clipvision and ipadapter will vastly decrease labor time in these sorts of endeavors

guys, you need to check out the stable audio website. type in an image prompt or something. be impressed and terrified at the same time

[fallen dock]

cool post it in #🎵｜stable-audio not here, lol

[nimble geyser]

is this channel called image chat, or general chat? if I were to post in the channel that's called "stable audio" would it not be a bit off topic? this isn't general image chat. it's general chat with images. and I don't see a multimodal channel. so I guess I'll just have to not post anything else with more than one style of media

thanks for the pedantry

[fallen dock]

https://tenor.com/view/emoji-text-bubble-gif-25744751

I mean they made a #🎵｜stable-audio channel for a reason but idk bro you do you.....

bro had a whole paragraph ready

[nimble geyser]

I was just having fun with something new and thought I'd share. I didn't know where to share. it's people like you that suck joy out of the world

[fallen dock]

Sounds like you did a good job of that yourself, I also noticed you deleted that comment, nice

Anyways, moving on

[nimble geyser]

well I mean, I didn't engage with you. how about you do some self reflecting and stop worrying about me

[limpid lichen]

kind of a shit ass attitude tbh

treating chatrooms like a meangirls highschool situation is just so not fetch

[inland summit]

Left or right, what you think is the best for a ring-shaped architecture on Mars, that can turn in rotation for an artificial gravity ?

[limpid lichen]

mars has gravity, just over 1/3 of earth. so if you are 50kg, you'd be about 15-20kg there. Also centrifugal force creates a force perpendicular to the rotational axis, not top down like gravity would expect to work. but i like the one on the left one better.

Realistic mars colonies are likely going to be similar to total recal. Excavated underground, sealed and pressurized. Domes like those would be for bringing in sunlight or entry points. Rotation may work for logistical reasons like a train switch yard

[limpid lichen]

i love the colors on this. using animatediff?

[wispy nest]

Yeah

[boreal falcon]

@wispy nest fascinating tech but man are they garbage compared to other stuff

[limpid lichen]

if you mean runway ml's gen2, yup. or pikalabs. yeup. But these are open models that can run on a home GPU.

Anyone can train a motion model too. Give it 6 more months and they'll be the standard

[boreal falcon]

standard for what? 😄

for general animating not at all

for some fun play around for a niche of consumers eventually

[fallen dock]

I do 3d animation and I can tell ya there is no way it's replacing it that soon.

[boreal falcon]

that soon to say the least

it doesnt even replace 2D static artwork

let alone 3D and especially animating

simply not comparable to animating in Maya, 3ds Max, Blender or in 2D animating as well

[heady anchor]

what kind of 2d static artwork?

[boreal falcon]

kinda all kinds

anime art is at biggest risk tho imo

but even there

the thing with anime just like with comic art is that you can hide and thats what is actually something artists (including pros) do is to hide hands or other parts where someone might not be that sufficient or just spares the time by hiding those

you dont have to draw and paint full body with every part of it on the canvas

which comes in handy for people within AI art community that like to generate anime stuff

but yet, there are still major issues with AI art even in that sector

[heady anchor]

as someone who used to draw i dont hide hands if they come out deformed i just fix whatever deformity is on the pic by drawing on top of it

[boreal falcon]

but there are those who hide em if the situation allows it ofc

i mean artists as well as people who use exclusively generative AI

[heady anchor]

yea if u generate and post a lot of images is better to hide them,its faster instead of spending like 30 mins per image

[boreal falcon]

yeah

unless you can work on them by "hand"

Photoshop or whatever

most people in the community cant likely

which is part of the reason why i say 2D isnt and is hardly going to be replaced either at the current state

[wispy nest]

do you know if i works to draw some ketc to use as controlnet?

I noticed that when there are mountains, negating mountains with negative prompt doesn't work

[wispy nest]

that womans head tho

[chilly reef]

hey which file i have to dopwnload

[nimble geyser]

second from the bottom

base with 0.9vae

[noble zephyr]

Just want to say that I find it highly impressive that SD was able to spit out this image without inpainting or after the fact editing

[nimble geyser]

https://tenor.com/view/the-universe-tim-and-eric-mind-blown-mind-blown-meme-mind-explosion-mind-explosion-meme-gif-18002878

[wispy nest]

Lol people coming here just to argue if it will replace animation

[boreal falcon]

Nah, but it got to that topic after someone claimed that generative AI animating will become standard (in general animating?) In 6 months 😄

[brazen sand]

yes I was thinking that, like it only works in low gravity right? and also the perpendicular thing. But the thing is it wouldn't even work on not low gravity, so it doesn't matter how it is built. But it would make more sense perperdicular. https://youtu.be/1wJQ5UrAsIY?si=K48EnDK6mas1nCSV

I mean if they put you on a giant wheel on earth, you will just drop, right?

Bah, it could work but it would be completely unsafe

[limpid lichen]

go to a fair that has a gravitron ride. same principle

when it spins up it pins you to the wall

[brazen sand]

oh I see

[wispy nest]

He didn't say that

Standart for this tech. He is not saying it will replace all animation

[boreal falcon]

Dunno, if thats the case then its different. Although i see someone else being the standard: Wonder Dynamics

That one is already going to be bridged with Autodesk Maya for example

I believe tho it wont be cheap to use

[chilly reef]

hey how to vhange brush color to black

change*

[noble zephyr]

well you're on the inpaint tab, you should be on the sketch tab

[cyan snow]

managed to use IPA while maintaining AIT =]

also figured out how to make IPA's blend function more refined

[wispy nest]

Sure. But it's insane that you can even do this stuff considering where we were just 1-2 years ago.

While now you can already make short gifs on your home PC

[cyan snow]

DUDE

duckball!

[weak loom]

Yeah that's pretty close to the sort of stuff I've been able to generate

I've switched over to a different ratio to avoid the bulky look (which works a bit)

[kind quartz]

still mech, i think you cant get a robot 🙂

[fringe spindle]

@kind quartz https://imgur.com/vvJPXW6

[kind quartz]

yes, Stable-diffusion, there should be models @fringe spindle

[fringe spindle]

(I put the web link in that folder too, in case I lose where I found this)

[kind quartz]

ckpt is model, suggesting you download file with safetensor extension, it is safer

[fringe spindle]

Er... wut? lol

[kind quartz]

those are on civitai.com or huggingface
mmnt i will send you some link for 2.0 model.

[fringe spindle]

Thank you

[kind quartz]

https://civitai.com/models/4384/dreamshaper

try this

[boreal falcon]

Yes, which is why i said the tech is still fascinating

And also not everyone will want to play around with Blender and co.

So the specific niche of users will find a use definitelly in this one

[kind quartz]

@fringe spindle when dl, put it in folder models/stable diffusion. And choose it from dropdown menu in top left on first screen.

[fringe spindle]

Thank you - going to do that now.

[kind quartz]

also check then what you have in VAE folder

[fringe spindle]

Where is the VAE folder?

[inner pagoda]

Hi all, I'm new to the discord, but have been putting hundreds of hours into building LORAs, and exploring machine learning. Love what I've seen here, and I want to start sharing too.

[fringe spindle]

@kind quartz Found the VAE folder - it's empty

[kind quartz]

where are models
hm you will have to find most probably 1,5 VAE
Try generate something

[fringe spindle]

I re-did the cat looking at mouse image usinf the model you suggested - it's much better

[kind quartz]

if you have good gpu, try 24 or 30 steps, if you have weaker put sampling method to EulerA

[kind quartz]

it is 8 or 6GB?

[fringe spindle]

Does this help answer that?

[kind quartz]

it looks o.k.

[fringe spindle]

What should go into VAE?

[kind quartz]

vae file, but it looks it is working without it. May be baked in model, but it wasnt written on model.
If you realy need put something in it, try ask in tech support, i dont know where it is.

[fringe spindle]

OK, Thank you for all your help - I really appreciate it!

[kind quartz]

np, @fringe spindle have to go sleep now 😄 good night!

[fringe spindle]

G'nite!

[wispy nest]

What is deepbooru

[wispy nest]

Dreamshaper has baked in vae

[fringe spindle]

I'm very new at this. I've just installed SD today. Do I need A1111, and if so, why?

[dry crow]

A1111 is the webui you use to create images.

It uses Stable Diffusion

[fringe spindle]

When would one use A1111 in preference to SD?

Also, what do "headLeaf" and "arstation" do?

[dry crow]

SD is the core mechanic to generate images. A1111, comfyui, invokeai, are webuis (tools) to use the SD mechanic.

Artstation is a tag used to get images that are likely similar to the ones on the artstation site.

Headleaf no clue

[fringe spindle]

Thank you. But I'm still not clear on when I would use a webui as opposed to SD itself?

[dry crow]

SD don't have a user interface. Its a Programm running purely as python scripts.
The webuis are user friendly Programms we use to generate images. In the background still works SD

Your using SD by using a webui

[fringe spindle]

Understood. Which webui would it be that I'm using? (Sorry for being so ignorant). I'm using what popped up as a browser window during the SD install process.

[nocturne oak]

Automatic1111 (aka A1111)

[fringe spindle]

Thanks.

[fringe spindle]

Is there any form of lexicon of what SD can understand in prompts? I'm thinking that I might use a term familiar to me but which means nothing to SD?

[kind quartz]

In A1111 go to img2img and interrogate a image, it should make you txt files with styles, artists and most common words or similar.
I think SD knows very much "words".

[urban siren]

Can anyone recognize model used for this? It's used in Tokio Lo-Fi Healing YT channel

[fringe spindle]

@kind quartz Thank you - that was interesting. This was the image I used:

CLIP returned this:
a train is traveling over a bridge in the mountains and hills of scotland, with steam pouring out of the top, Andrew Henderson, award-winning photograph, a storybook illustration, magical realism

Deepbooru returned this:
building, field, forest, grass, ground vehicle, nature, no humans, outdoors, road, scenery, tree

CLIP sounds better, I think?

I have to wonder though - how did CLIP know it's Scotland? To me it looks like it could be any hilly or mountainous part of the world.

[kind quartz]

@fringe spindle you should have generated 3 text files, somewhere in A1111 folder, with what i say before.

[fringe spindle]

The text I copied above appeared in a prompt box in the img2img window. I believe I'm using A1111 but only because someone told me that I was - I haven't seen anything that actually says so. I don't see any A1111 folder anywhere here. This is the folder structure I have:

[kind quartz]

no i mean text files that should be generated using clip, check interrogate folder

should be 3 files flavor something, authors and styles or medium.

[fringe spindle]

Yes, I have the following, which look like they contain the lexicon I was wanting before - but don't seem related to the specific image I was interrogating?

[kind quartz]

i think just small fraction 😄 but yes, can be inspiration.

it is not, it is generated when you interrogate any image.

[fringe spindle]

Oh I see. If I interrogate more images, will I get more content in "Interrogate"?

[kind quartz]

no, thats the point 🙂

[fringe spindle]

Er, sorry -- what is the point?

[kind quartz]

it is generated just once, sorry my english is very bad, never had it in school, too old, so sorry if i am confusing. Thing is interrogate how many pictures, those files will remine same

[fringe spindle]

Thank you (and your English is very good - it's my inexperience that is the problem!).

So, is there no way of getting more of that content?

[kind quartz]

you can check authors with pictures or tags used. On some pages.

[fringe spindle]

OK. Thank your for your patience 🙂

[kind quartz]

https://www.urania.ai/top-sd-artists

https://proximacentaurib.notion.site/e28a4f8d97724f14a784a538b8589e7d?v=ab624266c6a44413b42a6c57a41d828c

@fringe spindle happy to help 🙂

[dry crow]

Maybe thats done with the flat2danimerge model

[sterile ether]

I am trying to get an art style I made in MidJurney in SD. is there ANY way to figure that out? Any help would be great.

[frozen nimbus]

train a lora

[sterile ether]

Yeah I was thinking that... just it comes out as horror when I tried. I maybe need a better Lora method.

[kind quartz]

or you can try interrogate image to get some idea, can give you combination of artist and so @sterile ether

[sterile ether]

interrogate as in what?

lost me on that one.

[kind quartz]

in img2img, clip interrogation. Upload image here in A1111 and click on it, it gives you description of image. Probably i wasnt understand well.

[sterile ether]

Ohhhh

[wind turret]

@fossil kayak hello. Please can you share how you come up with this prompt?

[fringe spindle]

I was looking for the "Tiling" checkbox (and others). Has it been removed from the main window? I see it's in Settings, but I'd prefer it to be in the main window, for easier "as and when" access. Is there something I can do to get to appear like in this screenshot:

My SD window looks like this:

[kind quartz]

if you have it in settings, try turn it on in quicksettings. It should appear on first page on top.

[fringe spindle]

Er... quicksettings? I don't see that...

[kind quartz]

in settings show all, i probably misspeled name of it, mmnt, going to run A1111 it starts with quick, so search in all settings for quick

yes it is called quicksettings, search for it 🙂

[fringe spindle]

I must be looking in the wrong place - I'm in the Settings tab's window, but I don't see a search box?

[kind quartz]

show all. and then ctrl+f quicksettings should get here in screen i posted

only not sure if tile is there

it seems it isnt there

[fringe spindle]

OK, found it - I'm sure I looked there before... but anyway 🙂

[kind quartz]

supposed when you find it in settings, it should be here 😦

[fringe spindle]

OK, so I have quicksettngs now... what do I do?

[kind quartz]

it isnt there. It is total fail to remove it and keep in settings only.
I have not find it there. There are only tile for resizers...

[fringe spindle]

Er... but I found "tiling" in quicksettings... or have I got the wrong thing?

[kind quartz]

You have right, i couldnt find it by typing 🙂

[fringe spindle]

I have another question, please, about NG deepnegative?

[kind quartz]

NG deepnegative? I dont know what is it

[fringe spindle]

It's an embedding, supposed to help prevent malformed images

[kind quartz]

aha, and what you need then? How to use it?

[fringe spindle]

You put "NG DeepNegative" in the negative prompt box, and it should prevent many malformations, bad anatomy, hands with 6 fingers, etc. It's on CivitAI.

[kind quartz]

i checked it now. And what help you need. I think there isnt universal solution to solve all bad anatomies

[fringe spindle]

No, not universal, but they say it should help - it's been trained of bad anatomy, so putting it in nagative prompt basically says," Avoid these".

[kind quartz]

I have another question, please, about NG deepnegative?

and whats the question 🙂

[fringe spindle]

The file is called "ng_deepnegative_v1_75t.pt", and I've seen it referred to in other AI image generators, but only as "NG DeepNegative" - not the whole filename. So I was wondering, is it sufficient to use only "NG DeepNegative", or must I use "ng_deepnegative_v1_75t"?

[kind quartz]

ng_deepnegative_v1_75t
on page where you downloaded it, there is on right trigger words which should be you as well mentioned

[late echo]

Add different flowers and make her in princess style

[fringe spindle]

Ah thanks, I missed that 🙂

It's in uppercase there, but if you click the copy icon, it's puts a lowercase version on your clipboard - I suppose it doesn't matter? But, it seems you need the whole file name ("ng_deepnegative_v1_75t")

[kind quartz]

i think what is in copy matter, or it is not case sensitive

yes should be whole

[fringe spindle]

Thanks 🙂

[kind quartz]

no problem, hope it all works well for you!

[fringe spindle]

👍

[earnest solar]

You ever had a prompt be misread but in a cool way?

[earnest solar]

what's uh.... going on here?

Nothing in my prompt asked for the dnd style thing

Masterpiece, best quality, a large but old man, Darth Purgatus Dark Lord of The Sith, fantasy, Star Wars, detailed Sith yellow eyes, detailed red skin, villain, Sith Pureblood, short white hair, wearing detailed full cybernetic armour, sci-fi, wearing an eyepatch,

Negatives: Vader, Maul, wings, shield, horse, riding an animal, weapons, ((exposed skin, Big Breasts, Big Breasts, NSFW, lewd, risque, nude, sexual, skimpy clothes,)) hat, laser, lightsaber, lowres, bad anatomy, bad hands, bad faces, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, bad eyes, ugly eyes, dead eyes, blur, helmet, username, monochrome, face, showing face, gold, double, two characters,

[vapid tendon]

I'm going to cry... I've had 20 regenerations, this is the best but it has 3 legs (spoiler because of suggestive tone)

[heady ice]

how it work?

can someone help?

[nimble geyser]

I don't know

[shadow light]

hello everyone, sorry to bother, just want to ask for your advice..

i really want to copy the lighting of this photo because the beauty of the skin really stands out but i can't get it

i have tried = softbox lighting, overcast, outdoor lighting, frontal light, front light, dawn, neutral light, natural light, early morning,

but the light still appears to be coming from left side, but i can see a little improvement in "overcast"

[nimble geyser]

just make stuff like this instead

[shadow light]

can you tell me what prompt for lighting did you use?

[nimble geyser]

oh I don't use prompts anymore. sorry

[shadow light]

ok, thank you

[nimble geyser]

in all seriousness, I really don't know how to achieve what you're trying to do or I'd tell you

I can't believe how clean these all are

[shadow light]

yeah, first time i saw something like this on youtube, i really thought that, it was a real girl and i was like "what a goddess"

[nimble geyser]

I just make weird stuff so not sure how to make regular stuff

[snow pewter]

what do top p, nucleus, and top k do in this context?

[snow pewter]

has anyone trained using BLIP2 AND/OR WD14 captions without editing them? I wonder how the results were

[prisma stone]

There should be some way that I can fix hands :(

[tacit laurel]

These are slightly oldschool looking but they are part of a recent project!! This is actually an incredibly quiet and chill discord isn't it!! it makes sense that is !! 🙂 🙂 So i will try not to post too many, but i have found many different shapes that it is good at making!!

ok!! i will find a new project for next time but i had forgotten about these!! I think what's interesting about Stable is that, most of these would take many months to make in a CGI program and, i know i personally couldn't yet have visualized these types of scenes. But now that i have seen them who knows!! That is a sometimes overlooked good aspect of art visual light synths!! (which ai art programs are, they make light shapes like a synth keyboard or computer makes sound shapes!! and they correspond to emotional shapes in their infinitude!!)

[dry crow]

Hey you, make sure to use a highres resolution than 512x300 for example 512x768
Also use some negative tags like mutated, multiple limbs, blurry, lowres,

[vapid tendon]

It's an image for a bot on SillyTavern, it's a test more than anything

[dry crow]

Ahh okay

[snow pewter]

hey rid of all the negatives. Inpaint, maybe using controlnet especially if you had some idea where you wanted his hand to be.

[prisma stone]

Idk how to :(

[snow pewter]

I assume you know how to get rid of the negatives. That might already fix your image.

For controlnet, easiest way is to probably download someone's controlnet template. You got manager? You know how to install extensions?

[fringe spindle]

Noob question, please. I'm a new user. I have a few models (absolutereality, cyberrealistic, dreamshaper, juggernautXL, rpg). I've read good things about SDXL - what can it do that the others don't? Or, when would I use SDXL in preference to any of the others?

[kind quartz]

it is trained on higher resolution. So with no SDXL your generate at size 512x512 or 768x768, while with SDXL it is trained on 1024x1024 and other 1Mpx resolutions.
Thats just one thing

[fringe spindle]

Thanks @kind quartz

[kind quartz]

If you have gpu powerful enough, sdxl is superior.

[fringe spindle]

What happens if I close the SD browser window, and the cmd window? How do I run it again?

I've been too afraid to turn my PC off!

[kind quartz]

Should be run browser and then (cmd window) <- run webui-user.bat

[solid sedge]

im not able to get stable diffusion to do anything. currently im trying to get img to img to do anything but it doesnt work.

after i click generate nothing happens but it claims to be complete

any help would be greatly appreated

[umbral lake]

https://cdn.discordapp.com/attachments/487337208836128778/1072496689438535721/mgs2ai.mov

[snow pewter]

but most of the newer 1.5 models are trained on 1024x1024 images...

and SDXL takes way longer to generate a 1024x1024 than SD1.5 does... so I wonder what the actual benefit is... if there's something inherently better about the architecture.

I still get better results from 1.5 mixes, but maybe in the future when SDXL has better models, that'll change.

[kind quartz]

there are different advantages as well. You can go way higher and generate full hd wallpaper or even more.

[snow pewter]

is the benefit just about pixels? I make 2048x2048 on 1.5, but generally I prefer not to. So in that case I have no use for SDXL?

[limpid lichen]

yuou can make 2048 images withn 1.5, but it's thinking about them like they're 512 images and just filling in the details with it's 512x attention span.

SDXL has more parameters to understand prompting context better, plus a 1024x base attention. More fine grain details since the model has a higher native attention.

https://arxiv.org/pdf/2307.01952.pdf here's the technical report on it. It goes over a lot of the changes and what the goals are

[wispy nest]

https://www.youtube.com/watch?v=qgLme4eHJvk&lc=Ugw5-6-l3QVpqsFS3z54AaABAg.9umjm6A4ygh9uzsIQpWlve this is in Thailand I think. Great to use as controlnet

[undone otter]

making Vergil, might make the lawnchair meme and try the white monster one

[kind quartz]

from where are those almost perfect numbers from? CN model?

[wispy nest]

Qr module

It's controlnet I think

Haven't used it yet, it's cool

[kind quartz]

it is for old sd now only

[kind quartz]

Thank you

[kind quartz]

it even generated names for car 🙂

[open parcel]

hello! I am new in this discord, can anyone help me with this new version of AU1111? thanks in advance!

I tried to update it by editing the code with

git pull
and turns out it just says error 1 and does not do anything after I type any key

[kind quartz]

ask in #🤝｜tech-support with reference to this picture.

[open parcel]

thank you, sorry for interrupting!

[kind quartz]

not problem at all @open parcel

[fathom mural]

Sorry for my ignorance but what tools offers that node interface?

[prisma stone]

Oh I used the original one

with something called https://github.com/WASasquatch/was-node-suite-comfyui

[fathom mural]

thanks

[kind quartz]

Mondrian images are fun to play with CN

[cloud wedge]

hello guys
i need help finding R-ESRGAN-4x+ upscaler, where can i get it? i searched everywhere

[nocturne oak]

That's one of the standard official models:
https://upscale.wiki/wiki/Official_Research_Models#ESRGAN_Arch

[kind quartz]

@nocturne oak they are moving slowly to different place.
https://openmodeldb.info/
But you are right. They seems to be stay on both places

[kind quartz]

was really hard to hide him at least on this level, still very easy...

[lone viper]

Hello, can anyone make Medieval version of this image?
Will be much appreciated

[wispy nest]

anyone know anything about animatediff

it only does these

idk why

when i disconnect the animatediff model injector it just makes normal images

[swift plover]

Anyone have any ideas on how I can prompt SD for blog article photos?

adding "blog article photo" is surprisingly not bad, but I need a prompt more solid/scalable

[wispy nest]

what qualities does a blog article foto have

[swift plover]

No text or humans since those are distored

Just something that can accurately describe the blog title I suppose

[kind quartz]

probably 10 steps and gif can be issue? Gif has limited color, hasnt? Try different format if same @wispy nest

[wispy nest]

i did try other samplers and steps, doesnt work

i gave up on it

other people have the same bug without solution

or apparent cause

[kind quartz]

i mean gif, try png if you can

o.k. didnt knows that

your model has baked vae, you dont need load vae imho.
Can you share this workflow?

[kind quartz]

practicing to become golf resorts architect 😄

[wicked yew]

sd-webui-discord Support setting with ui, multi image gen, retry and delete image, multi ControlNet now! This is a opensource bot !

[obtuse palm]

any idea how to fix this part?

[wispy nest]

I'd like to carefully inspect the full image for research purposes in my DM

And to know what's wrong

[glacial agate]

i want to understand outpaint workflow. currently i do this very ugly way

1/ i got my original picture

2/ i want to outpaint on the left, so i copy/paste it in Paint in a large empty area, and i screenshot a new picture with a empty space that i want to outpaint

3/ Finally, i put this new img into a1111/i2i/inpaint and mask the white area and prompt

but is there any better workflow ? 1/ it is annoying 2/ i doubt about quality, i mean, i just do a screenshot from paint

[kind quartz]

there is openoutpaint for A1111. I like it is very creative, but not sure if for your purposes. If you want try it, then in webui user bat add --api in commandargs line.

[muted mulch]

animatediff

[obtuse palm]

looks smooth asf

though there's certain issue always there.. like in second image his right foot glitch

[muted mulch]

Yes interpolation is made with Rife, not perfect yet

[scenic umbra]

did you fixed this? I'd try another sampler and scheduler, some combinantions produce garbage

dpm 2M, 3M with Karras works good

also eulers

[boreal falcon]

A lot of tbh

For fun its usable tho

[wispy nest]

i'll reinstall it and try later, thanks for the tip

[wispy nest]

nope

[wispy nest]

it's vaaled

oh it seems to work with the base model

just not with any fine tune

[kind quartz]

old yard 😦

[cinder scroll]

What ai model is that generates images like that

[kind quartz]

it is model for sd for controlnet, called qrmonster, it is on huggingface

it isnt classical AI model, working with any 1.5 ai models, not with sdxl now

[cinder scroll]

Link to huggingface?

[kind quartz]

oh, i would have to search it as well. huggingface qrmonster. thats all

https://huggingface.co/monster-labs/control_v1p_sd15_qrcode_monster/tree/main

should be 723MB file + yaml

[cinder scroll]

What about one of the model pages?

[kind quartz]

?

Sorry my english i very weak
you will have to install controlnet, and model any 1,5 you can search on civitai.com, seems bit more readable.

[cinder scroll]

This is one of them https://huggingface.co/spaces/AP123/IllusionDiffusion

[kind quartz]

no, i am using jugernaut final

[cinder scroll]

What?

Is it on huggingface as well?

[kind quartz]

i think this isnt 1.5 model. Illusion diffusion i dont know what it is

[cinder scroll]

Oh

[charred light]

I'm get this error, any idea what I should do? Whats "ModelPatcher"

[wispy nest]

Can i somehow save all saved images with the seed for each pic?

[wispy nest]

If you throw the images in the PNG info tab you'll see the seed

They are automatically saved with all generation info

Most of the information

Only as pngs?

maybe idk if they have to be PNG to have info

I'm noob

[crystal delta]

FAQ: I’m missing credits/having issues processing payments. What do I do?

If your account balance is incorrectly showing 0, please wait 2-3 minutes to see if the system auto-corrects. If not, please confirm with your bank to ensure the transaction completed successfully & then fill out a request using the following form on DreamStudio so the team can investigate it further: https://beta.dreamstudio.ai/support

[kind quartz]

i think it is A1111 working only.

[kind quartz]

@wispy nest or simply give seed number to name of image in A1111. Not sure how in Comfui, probably similar

[timid yew]

Hey can someone recommend a good laptop that will run stable diffusion smoothly and fast enough?

Budget is 1k

[earnest solar]

Honestly, any "gaming" laptop in that price range should be 'good enough' for your purposes.

[charred light]

You would need a Laptop with at least a nvidia 2060 rtx graphics card @timid yew

[peak hemlock]

how would i make images like these. What is the process?

[kind quartz]

controlnet, 1.5 model, and QRMonster model for CN. And A1111

[peak hemlock]

thank you so much.

also does it work with words too? i saw some images of dogs that made a word if you squint.

[kind quartz]

yes but you have to made those words

[peak hemlock]

ok thank you for your help!👍

[kind quartz]

no problem 🙂

[gloomy field]

I love those

[wispy nest]

They are getting old

[blissful cairn]

Most time I've spent on a single thing in a while. Took forever to upscale, too 😦

https://sd.efreakbnc.net/gallery/media/original/photo_a_dirty_old_hobo__fashion_model_1_1___strutting_single_file_down_a_pink__catwalk_at_a_fashion_show___dirty_garbage_in_ (6).png.webp

[vernal arrow]

How did you generate that? Is that img2img with a given text and a small noise?

[kind quartz]

it is with high denoiser. Premade manualy TEXT and in SD controlnet in A1111 used qrMonster modul from hugging face. There soon will be similar i hope for SDXL

[vernal arrow]

Ohhh, cool!

[kind quartz]

@wispy nest can you read it? I always move with something crucial and cant get back...

[kind quartz]

what a large shaft thing on top. Glass ball probably?

[sterile kiln]

I asked him for a magician's staff with an orb at the top, he went a little overboard xD

[kind quartz]

🙂

still easy but at least quite different

[wispy nest]

i cannot maybe im just dumb

[kind quartz]

it is not letters but numbers 🙂 same as in bottom one @wispy nest

[wispy nest]

oh lmao then yes

[kind quartz]

🙂

[wispy nest]

im trying to figure out how to make it more visible with xl because .5 seems too faint and the entire image gets messed up at even .51

maybe should use this instead of the other

[kind quartz]

i got too much numbers and always forgot some get back and i am stuck.
Strength used to be about 0,7 in rare cases up to 1 in old SD

there is one value 0-2 set to 1, which is best, probably this strength.

[wispy nest]

yeah i put strength to 2 and it was just noise, seems like 1 looks like noise too

[rain gazelle]

total accident but I think it's fun 🙂

[limpid lichen]

https://imgur.com/QwoPSJO motion modules have really invigorated sd1.5. this is cyberrealism and i got open pose on a few key frames pulled from a dancing gif

same prompt new gen

[limpid lichen]

i love this shader style ❤️

[limpid lichen]

objects in mirror are closer than they appear

[wispy nest]

do anyone knows if there's like a way to publish manga with AI?

made with *

[rich token]

https://sg-public-api.hoyoverse.com/event/social_sea_share/redirectUrl?key=m20230831hy16658qo0&lang=zh-tw&game_biz=hk4e_global&title_key=cake_share_title&url=https%3A%2F%2Fact.hoyoverse.com%2Fys%2Fevent%2Fe20230927celebration%2Findex.html%3Fgame_biz%3Dhk4e_global%26hyl_presentation_style%3Dfullscreen%26hyl_landscape%3Dtrue%26hyl_auth_required%3Dtrue%26hyl_hide_status_bar%3Dtrue%26utm_source%3Dshare%26utm_medium%3Dhoyolab%26utm_campaign%3Dapp&img_key=cake_share_img_6&desc_key=cake_share_desc&timeStamp=5652914

[spring aurora]

/icecream images

#🏞｜general-with-images give me high quality mango ice cream images

[kind quartz]

@fringe shell but it is not much top down as well worthless...

[fringe shell]

worth less rather than worthless 😀 what ckpt are you using?

[kind quartz]

Jugernautfinal. safetensor

[kind quartz]

you can play in img2img to get some new fresh original idea. 🙂 Second image is original rendered. And first image is based on it, just it should look like bedroom in 18th century. I think same things can be done for exterior, been image guided <- if proper word. You need controlnet and some models, canny, depth, MLSD and as well probably QRmonster, which can boost imaginery very @fringe shell

[kind quartz]

@fringe shell there are first two your prompts if it is what you are looking for?

[undone kite]

I'm so impressed with your work bro.. really great. I'm trying to catch up to your skills but, I'm stuck on the first steps. with my version of SD only producing static or grey squares. any suggestion on where to search for the problem. I get no errors on the terminal

[kind quartz]

oh i dont know. Try post errors in #🤝｜tech-support
my skills are basically 0. If you mean animation, non static images it is not me 🙂 if hidden images, i played with it curently

@undone kite picture you are refering to is isometric livingroom i believe original one. Luck and time is needed.

[torpid rampart]

I need some advise, Im trying to create an image with ONE hallway in the center but all the time im getting thid:

suguestions?

[sterile temple]

center, symmetrical?

[torpid rampart]

I just added that in the last one

my promt: one centered symmetric straight long hallway with neon blue volumetric lights, aesthetic, mysterious, videogame, modern, neon, minimalist, liminal, backrooms, empty panels
Negative prompt: people, persons, silhouettes, multiple hallways
Steps: 78, Sampler: DPM++ SDE, CFG scale: 23, Seed: 4119121761, Size: 1024x576, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.6.0

I want something like that

[sterile temple]

a tunnel with neon lights, runway, purple and blue neon, violet and aqua neon lights, neon ambience

[astral patrol]

guys i just installed stable diffusion today but its only generate this blank picture what should i do?

[kind quartz]

can you post whole screen? of A1111?
What is your gpu, any error in cli? That black icon most right on bottom

@torpid rampart how about neonpunk theme or neon noir theme?

[torpid rampart]

tunel is good idea

[torpid rampart]

I will add this

[astral patrol]

here btw i have a 1660ti gpu and i didnt have any errors

[kind quartz]

and can you show what you have in Webui-user.bat? specialy in commandlineargs

[sterile temple]

luffy sounds like a lora and not a checkpoint

[astral patrol]

i have no idea how this thing work i just installed it an hour ago lmao

[sterile temple]

checkpoint is the main file you use to make images, and a lora is like a style or character you can apply to get a particular look

[kind quartz]

yes dl SD checkpoint and put it in models/stable diffusion folder

[astral patrol]

well y'know its was the model's name i downloaded from civitai but whatever i generate it give me a blank picture

can you help me?

[kind quartz]

download for example non XL version of model called dreamshaper, on civitai.com. Make sure it is labeled checkpoint and size above or equal to 1,99GB

i realized i have instead of neon origami 😄

[astral patrol]

https://civitai.com/models/4384/dreamshaper this one?

[kind quartz]

if model 1.5 then o.k.

[astral patrol]

yeah its s.d 1.5

[kind quartz]

so yes

[sterile kiln]

This model is very well known, I have used it forever

[astral patrol]

btw i have 1660ti gpu is that good ?

[kind quartz]

i think you will be able, but not suggesting you SDXL therefore. Should work

[sterile kiln]

Enough for 1.5 models yeah

[astral patrol]

ok thanks

[sterile kiln]

You will need to use the --medvram or --lowvram arguments

An example of what i did, i have a 2070 Super (8 GB Vram)

set COMMANDLINE_ARGS=--medvram --xformers --api --port 7861 --always-batch-cond-uncond --ckpt-dir "D:\SD\MODELS"

[kind quartz]

@sterile kiln and if you put model in original folder, does it sees it or only in path you specify?

this is closer @torpid rampart but not much 😦

[sterile kiln]

I did not try

[kind quartz]

o.k. np

[torpid rampart]

thankssss

[kind quartz]

it wasnt for you 🙂 @torpid rampart

[torpid rampart]

yeah hahaha there was 2 messages together

i didt click the correct one

[kind quartz]

yes classic me 🙂

[torpid rampart]

anyway im aslo istalling some models xd

installing*

Is possible to make small gifs/videos with automatic 111 UI?

[sterile kiln]

It can help
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings

[sterile kiln]

A1111 is dotted with extensions, some of which allow it. But be careful, this is the kind of thing that can require a lot of GPU performance

[torpid rampart]

Hahaha my laptop can get toasted?

[sterile kiln]

It can end in a CUDA crash or be very, very long

[kind quartz]

or out of memory as well

[torpid rampart]

I have 16Ram memory (15 avaibable) and Nvidia GForce Rtx 3060 plus intel core 9

u think it will survive?

Literally I need to make something like a like an still images with a little movment

[sterile kiln]

It's the Vram that counts, your ram will only be used a little

[torpid rampart]

Also im seeing that there is checkpoint and checkpoint XL

Can I put anyway Checkpoint xl models in stable diffusion folder? Or needs to be another folder

[sterile kiln]

There are 3 types of models. Those based on 1.5, those based on 2 and 2.1. Then SDXL

[torpid rampart]

so

I cant use them on Automatic111?

[sterile kiln]

Extensions do not work on all models.
1.5 is less demanding, but is the one with the most extension, and which allows the most for those who have little Vram, but SDXL has fabulous results, but requires much more Vram, therefore takes longer.
I advise you to avoid 2.x

[sterile kiln]

yes

[torpid rampart]

ok

[astral patrol]

thanks that dreamshaper thing worked do you know have any method i can make it faster to generate?

[sterile kiln]

How many Vram on your 3060?

[astral patrol]

its a 6gb 1660ti

[sterile kiln]

Oh, i melted you guys x)

I thought some people were saying they could use SDXL with this card

[kind quartz]

less steps, hurts quality, medvram instead of lowvram, xformers you have, Sampler EulerA or try uniPC could work on lower steps 🙂

[astral patrol]

thanks

how many steps is enough?

[kind quartz]

and resolution 512x768 or other way round or 512x512 or 768x768 no higher

not enough steps, but let say for your HW 20-32?

[torpid rampart]

xd

[astral patrol]

btw what is CFG scale?

[kind quartz]

cfg scale is how it follow your prompt and using ai creativity.

suggested say 5-15, play with it to see.

[astral patrol]

ok

[sterile kiln]

The CFG Scale is the strength with which SD will listen to your prompt, the more CFG scale there is, the closer it will be to your prompt, with potentially more quality, but if you use too much, SD loses its creativity, and it often becomes very ugly. Personally I play between 7 and 12

ah Bernix was faster x)

[astral patrol]

then i will let it be on 7

[astral patrol]

actually yours was longer and more detailed....

[sterile kiln]

10 is often the happy medium

[torpid rampart]

hehehehe

now me

How do I install an extension

https://github.com/guoyww/AnimateDiff/

[kind quartz]

in extension tab.

[torpid rampart]

I downloaded it and put it on the extension folder

yeah yeah

and I have an error

[astral patrol]

is it possible to generate more than one pic at once?

[torpid rampart]

animatediff.github.io/

[torpid rampart]

how i get that link

im using this: URL for extension's git repository

[kind quartz]

@sterile kiln i saw you type, so i didnt write. Then i dont see you typing so i wrote it, but you for sure write it more precisely

[torpid rampart]

animatediff.github.io/animatediff.github.io/

ah ok

avaibable and not from url

[sterile kiln]

of course. and greatly better

[astral patrol]

but how?

just spam the generate button?

[sterile kiln]

https://raw.githubusercontent.com/AUTOMATIC1111/stable-diffusion-webui-extensions/master/index.json

[kind quartz]

not for your hw i think, you should render more in series, not in parallel

[torpid rampart]

mmmmmm

[astral patrol]

okay....

[torpid rampart]

but this link is for animateddiff?

[sterile kiln]

To simply generate multiples ones with the same prompt, just use Batch count

[astral patrol]

oh

[sterile kiln]

Nope, its the list of extensions

[torpid rampart]

oh pff xd

im more confused now

[sterile kiln]

Oh, srorry, you're right

[kind quartz]

Not use batch size! @astral patrol

[astral patrol]

ok

[kind quartz]

just batch count

[sterile kiln]

you just have to click on load from, most extensions are listed there, but there may be some that are not listed

[torpid rampart]

ok found it

any other extension you recomend?

Now im here

[astral patrol]

what smapling method is good?

[sterile kiln]

Euler A, DPM++ Karras, DDIM, UniPC

[sterile kiln]

ControlNet with models
Stable.Art (Photoshop plugin with A1111 implementation)
MultiDiffusion (use Tiled VAE Fonction can save lot of Vram)
Posex

Here's what you can do to better understand your models

[torpid rampart]

mmmm and if I do more landscape stuff

or scenes

[sterile kiln]

In one go, A1111 generates everything and does the fresco himself
(X/Y/Z plot)

[torpid rampart]

not really a portrait person xd

[sterile kiln]

It is not the content that is interesting, but the methodology.
How to get what you want most easily

Do a X/Y/Z grid can help you to know if your Ckpt model, or LoRA model support more or less of CFG or else

[torpid rampart]

ok

Thanks a lot

[astral patrol]

how much vram do you use ?

[sterile kiln]

8

If you have CUDA crashes or out of memory, you can use the Tiled VAE I talked about above to exceed your max Vram, but the generation will take longer. Use only if you have crashes.

And first you probably need to put --medvram or even --lowvram. Otherwise your tests will be biased

[kind quartz]

or lower resolution

lowvram for 6GB isnt needed, lowvram slows down generation very

[astral patrol]

what if it use all the vram?

[kind quartz]

--medvram is must for you

[sterile kiln]

I think too, he should use --medvram

Once you have found the prompt, the CFG scale and the ideal seed, you keep the previous seed by clicking on the recycle button, then you use the HighresFix to generate the same thing bigger.

This is how with low resolutions, we can achieve very beautiful results in very high resolution

(an old one with 1.5 based model)

4128x4128

[kind quartz]

i would prefer resizing in extratab. I have always issues in hires, i must missed something crucial 😄
for example latent isnt lossless, other needs plenty of VRAM. But probably only my problems.

[sterile kiln]

Resize in Extra tab is upscaling, HighresFix is Img2img + upscaling, a complete difference. HighresFix will add details, create another one, upscale don't

[kind quartz]

i know

but it will not create exact same image? I will test it again iam in A1111
what method do you suggest @sterile kiln latent?

[sterile kiln]

4x-UltraSharp or ESRGAN4x

[astral patrol]

is there a way to save the prompt when i launch the webui?

[sterile kiln]

Depend on what Denoising Str you set, lower is Desnoise, closer from the original image it will be

[kind quartz]

o.k.
png info, keep pictures in png and you keep prompts that it was created it with, or you can use style save it in first page @astral patrol

yes denoise i know very played with it last few days 🙂

[sterile kiln]

Or you can simply drag&drop an precedent generated image in prompt field, and then clic on blue arrow, it will apply your settings of the image

have to do it on a clean tab

[astral patrol]

i used that --medvram and this happened wtf

[kind quartz]

can you show your webui-user.bat?

[sterile kiln]

and copy paste error on https://pastebin.com/ and share link here, better to read

[astral patrol]

i used that --disable-nan-check and got fixed

[kind quartz]

hm, this is bad solution. Ask in tech support.
Have you put that --medvram in commandlinearg =

[astral patrol]

got fixed by itself lmao btw its it possible to use multiple loras at the same time?

[kind quartz]

It just hide error messages, not solve them. So it should work before as well

[sterile kiln]

Its possible,
but be sure to use LoRA compatible with your ckpt models. If you have LoRA 1.5, don't use SDXL models and vice versa.
This generally happens when you keep the same prompt by just changing the ckpt.
Because A1111 does not allow loading LoRA incompatible with its current ckpt (the graphical interface)

[astral patrol]

nah i deleted that --disable-nan-check from the bat because my generates got black then i run the webui again and it got fixed somehow

i think i got lucky

[kind quartz]

maybe

[torpid rampart]

AAHH I dont know what I did

and now im having an error

in CMD

So web UI is not opening

[kind quartz]

can you post what do you have in webui-user.bat?

[torpid rampart]

where is that? Like in CMD?

[kind quartz]

in root of stable diffusion, the file you double click to run SD

[torpid rampart]

ah so yeah CMD

MMM

One second

[sterile kiln]

and copy paste error on https://pastebin.com/ and share link here, better to read

To do it, select all text of cmd by a right clic, (not ctrl+c, Ctrl+c is kick process in a terminal)

[kind quartz]

sorry i am bit slow 😄

[torpid rampart]

hehe

maybe is ModuleNotFoundError: No module named 'tqdm.auto'

[sterile kiln]

Another great Grid

[torpid rampart]

but idk what does that mean

[sterile kiln]

Same answer, just use pastebin, as i said before

[kind quartz]

me too, probably some extension you installed?

[torpid rampart]

Yeah for sure thats the problem becaus eit was working before

but I took all of them out of the extension folder

[kind quartz]

it is not cleanes way. How about cmd?

[torpid rampart]

https://pastebin.com/xXXakRHs

I took all the extensions out and still nothing

[kind quartz]

show what you have in webui-user.bat 🙂

or you will have to reinstall probably. Delete VENV folder and run webui-user.bat -> it will install again

[sterile kiln]

How long have you installed A1111?

[torpid rampart]

how long?

today

xd

[sterile kiln]

ok

[torpid rampart]

and I was just testing extensions and this happend

[sterile kiln]

What extension did you installed?

[torpid rampart]

pff

wait

[torpid rampart]

where is VENV folder I dont see it?

[sterile kiln]

Stable Diffusion\stable-diffusion-webui\venv

[kind quartz]

all seems to be related to video = demanding
in root. It is folder, but ask in #🤝｜tech-support if good idea?

[sterile kiln]

You should remove suspicious extensions you installed recently. You can just remove folders

[kind quartz]

as it cant find python, not sure it will work

probably you deleted more things

[torpid rampart]

I delete everything of the folder

[sterile kiln]

don't

there is some default extensions

[torpid rampart]

of extensions I mean

[kind quartz]

and do you installed it with git or with zip file?

[torpid rampart]

I just move it to other folder outside in my desktop

zip

automatic installation

[kind quartz]

hm i thought so, do it again, and with git

[sterile kiln]

well

[kind quartz]

zip is BAD way

[torpid rampart]

ok

[kind quartz]

go with a1111 installation for windows nvidia if you have those

[torpid rampart]

I did this one

Is the zip one you mean right?

and you are saying to do it this way right --->

[kind quartz]

yes i think so. Use method with git. It has many advantages and less troubles.

[torpid rampart]

ok

[torpid rampart]

and to unistall completley is only deleting the folder?

[kind quartz]

yes

when installing python, check add to path!!!

[torpid rampart]

oki thankkkks for all the help guys

[frozen hound]

Hello everyone. I'm having an issue of my results fading when I do the run. May I post for help here?

[sterile kiln]

send one of these images here

[frozen hound]

Starts with left image and after a few frames becomes realistic again

[frozen hound]

starts with left and then fades to realistic after a few frames

[sterile kiln]

No, send the result image, i need EXIF data

not a screenshot

[frozen hound]

File Name : POS New Lora 2(0)_000074.png
Directory : /Users/jacksondoran/Downloads
File Size : 771 kB
File Modification Date/Time : 2023:09:28 11:08:20-05:00
File Access Date/Time : 2023:09:28 11:11:36-05:00
File Inode Change Date/Time : 2023:09:28 11:11:34-05:00
File Permissions : -rw-r--r--
File Type : PNG
File Type Extension : png
MIME Type : image/png
Image Width : 1280
Image Height : 704
Bit Depth : 8
Color Type : RGB
Compression : Deflate/Inflate
Filter : Adaptive
Interlace : Noninterlaced
Image Size : 1280x704
Megapixels : 0.901

I feel like this is not what you're looking for. Sorry very new to this

[sterile kiln]

just send the image here

[frozen hound]

I copy the png in here and it comes as screenshot

[sterile kiln]

Go in stable-diffusion-webui\outputs\txt2img-images, and drag&drop your image

[jagged garden]

Hi, i have a question. My teeth looks sometimes like this, way to often in my opinion... Is there a fix for this? in the prompt, or negative promt, or after detailer? Something like bad-hands-5 for example..? Any solution?

[late root]

I've been experimenting to improve AI-upscaling with sd.
FreeU helps a lot to add fine details:
Left: original 128x128;
middle: ESRGAN NMKD 8x upscale model;
Right: Details added with stable diffusion.

workflow and test pic:

[mild tusk]

i made cursed while i was trying to get something something roblox

[sterile temple]

what beautiful eyes! 😄

[mild tusk]

gordon freeman be like in HL3

[mild tusk]

haha

two goblin fighting

ok that is amazing

ok this one have made a m16 pretty good

ok a last one i dont want to spam

[rain gazelle]

It's Christmastime right?

[cursive turret]

just merged my first model, what yall think

[heady anchor]

😍

[wanton nimbus]

is the stable diffusion bot entirely free?

[short hamlet]

I made a mistake.

I meant to generate 150. I think I generated 1500. This has taken 7 days thus far.

[short hamlet]

Also, yes it is

both local and online

Local is much harder to set up, but is definitely far more superior in the fact you can change model

[short hamlet]

Turns out, nope! It's just a beefy model. Furrydiffusion.

[wild sorrel]

what are you even doing

this looks like image mining at this point 7 days, damn

[empty flower]

why is it stuck like this?

i finished first round on training and would like to continue with second round

but it stucks on waiting...

[wanton nimbus]

well i used local and it did not like that lmao

[short hamlet]

Your computer might not be able to handle local

[wanton nimbus]

it's a beefy pc so im sure it's not that

[short hamlet]

You could've set it up wrong

Or the model you're using is just outbeefing your pc

[obtuse palm]

can someone suggest me something that will target and remove the text like thing appears on the top right without modifying anything else in the image?

[stoic pivot]

Mix something cute and something fun and throw in some puppies; what can go wrong?

The prompt was: "retro circus poster showing a baby clown and a small puppy dog standing in a colorful candy landscape, wearing clown outfit and knee-socks, a circus tent in the background"
Model; "Stable-diffusion-xl-base-1.0"

[wanton nimbus]

i followed the auto1111 steps

[limpid lichen]

what was it you didn't like?

[wanton nimbus]

the images had fucked up colors