🏞｜general-with-images

[pallid ruin]

Okey, paste the second image on top of the zoom one, but it must be in another tab

Something like this

[tawdry escarp]

ok done

[pallid ruin]

Now click both of the images with shift pressed, right click and select "link layers"

[tawdry escarp]

done

[pallid ruin]

Then copy and paste them into the original image now

You selected "convert to smart filters" before copying the image and using the "zoom" on it, right?

[tawdry escarp]

yeah

[pallid ruin]

Okey so now align them and move the new layer to the top

[tawdry escarp]

done

[pallid ruin]

Okey, now with the lasso tool or the eraser you can delete everything except the hair and face

Remember to first click on the first layer one time

If someone knows an easier way to do this please post it lmao

🤣

[tawdry escarp]

yeah this is insane

it's a lot of work for just 1 image lol

and the lighting changes unfortunatley

face lighting looks off

thank you so much for trying to help me

[heady anchor]

just install adetailer extension to avoid all that work,it automatically fixes faces

[tawdry escarp]

thank you!

[tawdry escarp]

this worked SUPER WELL

Thank you so much

[pallid ruin]

I´ll try it too , thanks @heady anchor🤗

[tawdry escarp]

@pallid ruin it basically detects the face after everything is done and re-run SD zoomed in

basically what we were doing before but automatically

[calm nimbus]

I've got this still screen shot from a game. I'd like to increase the resolution, make the foliage more vibrant and detailed, maybe add texture to their clothing, and dehaze. My end goal is to make this wall art for my sons room for xmas. I've looked on replicate, but I can't find a model that suits what I'm trying to do. Could use a little guidance. Thx in advance

[final timber]

hi so im trying to use latent couple and this is what im getting, my prompt for the left one was beach
AND lora:ammit_mk_v1:0.85 ammit_mk, crocodile humanoid, scales, tail, bikini
AND lora:kurama-v2-locon:0.85 kurama, furry, fox, anthro, tail, claws, bikini

what am i doing wrong?

[pure flower]

an exploration with comfyui, controlnet, and purposeful storyboarding intent.

[sterile kiln]

What are you using? AutoPS SD or Stable.Art?

[spark finch]

what is autops sd?

oops lol that image I had pasted a while back I guess or left it on the input line, ignore it

[pallid ruin]

Lol

[spark finch]

it was a pretty awful lot, I was trying out a custom node i made, which is broken as all hell, but it takes text input and grabs a random image from google images of the prompt, as a style to feed to ip adapter...

lovely

yes working as intended... a bit burnt, but probably too high on weight for that ipad

[spark finch]

ah man, so I had the wrong ipadapter selected... and i love mistakes... they're my favorite generations

I don't know. probably should fix the noodles

definitely connected the wrong one

there we go, now it is working better

[spice cairn]

An artistic design I can use for a skin of a Kalashnikov in CS go with the theme of finished redwood and shiny platinum trim

dalle 3 gives me

sd gives me

[pallid ruin]

Yeah sometimes sd doesnt have very good results, but at least is local 😁 and you can train or use a lora about those things that it fails to create

[pallid ruin]

lol

[spark finch]

Remember SD did not give you that, the model you are using give you that

I wager that dalle and mid journey used prompt negotiation and enable certain style formats automatically similar to how we would manually add a Lora or IP adapter images. I don't think that either of those cases are purely the model.

[tawdry escarp]

Huh? Dunno what either of those are

Sorry, I'm really new to this

[spark finch]

magnificent christmas badly merged model....

[karmic gale]

My friend created this, how can i make more of this? As this not doesn't have img2img 🦿

[pallid ruin]

Im stealing that 2nd image for img2img 😁

[spark finch]

They're not hard to generate I am using ipadapter at about .50-.55wt and starting at 0 end at .22 and using various where the wild things are styled art

Just cranked out a hundred and they are all weird af and awesome.

I didnt img2img hrf it yet to clean up the IPA fry it has a little bit of yet though

I used realism or real-world some model like that one of the 10 real something's

I wrote a little back end to grab random images from Google from a style prompt and stick it in the directory and that one came up and did some really goofy stuff

[pallid ruin]

Are you using a1111 or comfyUI? 👀

[spark finch]

Comfy.

Those probably don't have the workflow unless copy paste copies the workflow

I'll throw one with the workflow if you want me to

[pallid ruin]

I dont know why but comfyui ALWAYS uses the --lowvram option with my gpu, I stopped using it because its waay slower than A1111 just because of this

Did the same thing happened to anyone here?

[spark finch]

Really weird

I have 8GB

And it doesn't do that afaik it's fairly quick. As well. A1111 has some weird creep issues with memory leak.

Especially with controlnet. Over time it just gets slower and slower until it goes from like 6it/s to like 20s/it

[pallid ruin]

Ooof its a nightmare

I gotta change the prompt 🤣

[formal kraken]

try adetailer?

[pearl moth]

This looks majestic. having long ass prompts really works

[minor stag]

I don't know how to install it, I want to have it on my computer but I don't know how to do it.

[clear crag]

Hi someone can help me ?
I try to make images with laser eyes same like superman do or homelander in the boys.
But everytime it generate images with glasses like that

[tame tundra]

Let's hope that planet is teraformed 😅 Looks a touch thin atmosphere for survival - bomb image - excellent features.

[wispy nest]

imagin/1 3d

[gaunt heron]

#1072220168534642768

[wispy nest]

||3d text||

https://tenor.com/bAryu.gif

[blissful bramble]

@spark finch ?

[spark finch]

ні...

[blissful bramble]

Sorry

Bad app

[spark finch]

не те, що я хотів сказати

not what I meant..

I was saying to use an SD inpainting model to fix it.

can you not run in cpu only mode? I think if you just inpaint a 512x512, with 10 steps, even a cpu can do it, in an okay amount of time,

[blissful bramble]

I have a bad CPU

And I can't install SD

[spark finch]

whats wrong with the eyes

they look.. funny

[blissful bramble]

It's a sticker

[spark finch]

oh I thought that was something you made lol

and wanted to fix

it's eyes look weird :

like someone forgot to set the right clip layer

[gaunt heron]

No, that's just anime

[serene flint]

ur anime

[serene flint]

ty for the prompt

It even animated cool, what an awesome prompt, thanks a lot sucker!

[blissful bramble]

SteelSeries

[serene flint]

more like steal series amirite

[blissful bramble]

I stole a plane and you? || it’s a joke||

[serene flint]

oh wow what cool stickers who made these

[blissful bramble]

Really ?

[serene flint]

3 of them, yeah

[blissful bramble]

🗿

🫡

🏆🏆

[serene flint]

or i mean, i did

[blissful bramble]

Respect

https://tenor.com/view/respect-hat-cap-kid-gif-7551965

[serene flint]

tyvm

[blissful bramble]

I've always wanted to win ever since I arrived

Sometimes I try to do as much voting as I can

[serene flint]

That's great, I bet you'll win one soon

[blissful bramble]

I didn't expect you to reply to my useless message

[serene flint]

Too bad

[novel panther]

Do you guys knows that how to make the generated image present full body

[wild sorrel]

adding full body worked for 1.5 based models, idk about sdxl

[limpid lichen]

describe shoes and eyes in the same prompt.

[wild sorrel]

that's an option too

[tight fossil]

is taylor swift a jedi??!

[spark finch]

a little bit of interpolation and it isn't so bad from the bot anim

the bot doesn't do too awful.

never really messed with it until now

not exactly what I asked for but... looks ok

[covert girder]

i'm trying to convert a picture to this style, i'm not having too much luck for now, any suggestions?

[south spoke]

You can try image trace feature on Adobe Illustrator. It works similarly to what you're looking for.

@covert girder

[spark finch]

I bought a tesla cybertruck

on Temu

[wispy nest]

img2img is crazy

[spark finch]

I know I put my waifu generation into img to img. is was her cute face with bunny ears, and it gave me this 😦

this was the original, maybe I changed the resolution to much

outpainting can have such weird effects shrug

[covert girder]

I’ll try but from what I can see it picks up too much details

[primal forge]

guys I'm trying to generate an image of a carriage with a driver and one/two horses pulling it, and there's a woman inside. I never managed to get it right once, sometimes the horse is missing a leg, sometimes the coachman isn't there, or the woman is missing. How do you deal with that many specific mandatory details? Manual compositing via photoshop is the only way?

I challenge you to do that without controlnet

[sullen escarp]

supremecoconut: without controlnet, i think the general process is to find a seed that you like that gets most of the way there. by "most of the way there", i mean the horse has the correct number of legs, the scene composition is one that you like, the lighting is vaguely correct, and so on. at that point, you save the image, and then do inpainting on it to fill in the details that are either missing or wrong

i'll have a play around. i used the prompt "horse drawn carriage" with the negative prompt "text, watermark, missing legs", and used SD 1.5 with a seed of 3 using an euler sampler. it gave me this image:

it's not great, but it's enough to inpaint, i think... i'll give it a go

you can mask off areas that need to be inpainted like this:

you can see that i painted a mask over the people in the carriage, and blurred the mask slightly (to improve the quality a bit), encoded the image to a latent with the mask and piped that into a sampler. the sampler is configured with denoising set to 0.9, which essentially means "execute denoising, preserving about 10% of the original image content". only the masked area will be denoised, and the 10% will be mostly made of pixels sampled from the people and the surrounding trees. i used a positive prompt of "trees" and a negative prompt that included "people", meaning that the denoising process essentially removed the people from the image

you can see how you might iterate using this process to add in people as needed, remove objects as needed, correct details, and so on

no controlnet needed!

[serene flint]

#nocontrolnetganggang

[sullen escarp]

i like controlnet, but the request was "no controlnet" 😉

[kindred violet]

can someone help me? i am getting only grey pictures in stable diffusion when i start it up. after a while it mabye fixes itself. And sometimes i get no results. only some weird blurred colors. is there a way to fix it? restarting the ui does nothing

[kindred violet]

happens everytime when i try to use loras

[molten mortar]

Looking for suggestions: In img2img, how can one get the facial expression eactly the same as orig? I am guessing some controlnet model/setting might be the answer Or any other tool perhaps.

[pallid ruin]

Do you have the models in the correct folder? I mean sd models on /models/Stable-diffusion and lora models on /models/lora
Another person here had the same error because the models were in the wrong order

[cinder trellis]

HELLO CAN SOMEONE EXPLAIN me

What's the difference between Latent Noise and Fill

Don't they do the same thing

[kindred violet]

loras are here

and models here

[sonic sail]

Oops? lol

[sterile kiln]

You should find out about "inpainting". What you are trying to do IS inpainting, there are excellent plugins like Stable.Art which allow you to do this very easily.

It's a very old video that I made when I discovered S.A, but it sums it up well. Everything that was added from the base image (below) was generated with Stable.Art (except the dino, which was a photo of a dino toy that I pasted and then inpainted).
Basically, the inpainting is localized img2img.

https://www.youtube.com/watch?v=mtNliPaCNKI

[sullen escarp]

@zealous lily this image was generated in around five seconds on a 1660 TI, so you can at least get something going!

i don't know what those floating red things are, and i suspect i'd prefer not to know 😅

[lapis quarry]

What is the difference between these two?

Other than one saying ‘inpainting’ and the other not saying it.

Do I need to use the regular version with the inpainting one for best results or can the inpainting one also generate the same quality images?

[pallid ruin]

The inpainting version is when you want to add or change things in a photo/image , the normal version is for text2img or img2img

I´ll run sd now and post another example

@lapis quarry

[lapis quarry]

Ah i see.

So i generate an image with it, then change the model to inpainting.

Cool

[tawdry escarp]

thanks!!!

[silver merlin]

Could somebody please explain to me the best way to keep generating one face or person with the same head/hair/eyes/etc., but in different full-body poses?

For example, this image

[placid sundial]

im strugling with the same idea. what are you using?

[silver merlin]

everything the same from the neck and above, but full body and I can eventually change poses

[silver merlin]

Uhh I guess I'm using stable diffusion

web ai or something

webui

[placid sundial]

are you able to get the same face everytime?

[silver merlin]

I'm not sure, probably not

I'm just looking for an easy way to keep the character I want, but generate different images with it

[placid sundial]

yea im trying to figure out the same thing

[silver merlin]

gotcha

oh yea, I have heard that giving it a name, like a really random name helps i think

[placid sundial]

what do you mean by that

like a name of the character?

[silver merlin]

in the positive prompt

give it an extremely unique name

like

if its a boy with brown hair and freckles

name it

joetheboywithbrownhairandfreckles in it's own line

how i understand it is it just helps the generation a bit so it's narrowed down a bit more

idek how to explain

[glossy herald]

Having quite some fun with streaming scenes using SD
https://clips.twitch.tv/DeterminedSpookyGrasshopperDeIlluminati-bJ5OT4ZggsRHS4pP

[warped quail]

how do you do img2img in the discord webui?

[static crow]

oh un autre fr xdd

but keep it in english i don't think it's allowed to talk in another language

[wild sorrel]

huh

[glossy herald]

it is allowed to talk other languages on here, yeah, but frown upon by the community itself if used as main way of communication between us on discord. here, the language isn't the part that is shared really, I thought it would be ok (having moderated here for some time, I would have let it slide)

[static crow]

oh it's allowed here noice

most of the server anre english only

[glossy herald]

not shown in this clip, I also put a draw command that draws the prompt of the viewer, or an avatar reward, that does img2img on your twitch profile picture and uses it as your new avatar in the stream's chat

[static crow]

noice

[glossy herald]

funny useless features but yeah, trying to spread some SD love and awareness :p

[static crow]

👍

[oak dove]

https://tenor.com/bKiKJ.gif

Generate 3D style character photos based on photos

Generate 3D style character photos based on photos ┬─┬ノ( º _ ºノ)

Generate 3D style character photos based on photos

[sterile kiln]

You re welcome

[sterile kiln]

It's beyond comprehension, I've made LoRAs of friends' faces, with 40 photos, the results are much worse than my face with 10 images... I trained in just 1h20 (SDXL), where 8 hours is less for others...
Magic!

[modest furnace]

does anyone know what model this is?

[signal mulch]

Doubtful that the model can be inferred from the image alone, but I suppose some people can take guesses at least

[tawdry escarp]

openpose

[silver merlin]

I've tried that

but

isn't that only for poses?

[tawdry escarp]

do you have a checkpoint?

[silver merlin]

just to reclarify, I would like to know the best way to keep the style and characteristics of a generated realistic person

[silver merlin]

yes, juggernaut

some others ive been trying too

[tawdry escarp]

that's the checkpoint. I've never had any issues re-creating the same girl

with the same checkpoint

[silver merlin]

Oh, so a checkpoint is supposed to be one character (for example)

like

would it be exact, though?

I would want to make my own checkpoint, for example, girl with brown hair, blue eyes

but

[tawdry escarp]

a checkpoint is basically a character trained, yes

[silver merlin]

every generation is the same facial structure, eyes , hair, everything

gotcha

ill look into making my own then

[tawdry escarp]

you can train one but it's not super easy

if I'm not mistaken

Haven't trained one myself

[silver merlin]

if you have an idea of how one could get started, could you point me in the right direction with a video or some instruction?

[tawdry escarp]

I really have no clue as I've never done it, best guess is to youtube search it tbh

[silver merlin]

alright

thank you!

[serene flint]

There's many ways to achieve character consistency, like training a lora on images of them such that it can produce them accurately, or sometimes even prompting can get you there on its own, but from the sound of it I would suggest looking into control net

[silver merlin]

Okay, thank you

[wintry pawn]

please please what is going on?
i have 100% same settings

both shouuld be the same

they are wayy to different

[pallid ruin]

Try using the same seed

[wintry pawn]

same seed ofc

2564348649

try your self

its amgic

nonsense

since sdxl many models adn iamges do not replicate

[pallid ruin]

👀 That´s strange, if you are using the same model it should be the same

🤔

or at least that happens in 1.5

[wintry pawn]

now even extensions can affect

[heady anchor]

if u use xformers and have a different gpu the results wont be the same

[wintry pawn]

ahh forgot the dam xformers

i have nota ctivated

i am using a 4090 and apparently they make things slower and less quality

[rich valley]

I'm trying to use the Krita plugin with a custom ComfyUI server, and the webUI works fine and I can use all the checkpoints but going into the krita plugin it gives me this

im not really sure what I need to download or where to put it

[rich valley]

didn't really figure it out, but it's working now :/

[frigid bay]

Help please

[tawdry escarp]

anyone know why my model's eyes are sometimes hazel and sometimes brown? for that matter, her face shape also changes when she has brown eyes. I want her to have hazel eyes but most of the time she is generated with brown

these two images were generated using the same prompts (((hazel eyes))) and checkpoint:

I'm getting VERY different results in face shape, eye colour and even skin using the same prompts which is weird

[wispy nest]

could be something to do with LoRA you might be using

[tawdry escarp]

I'm not using any Lora, only a checkpoint

[tawdry escarp]

you think it's the checkpoint?

[tawdry escarp]

what upscaler do you guys think looks more realistic?

[wild sorrel]

That's too low res \ denoise to compare.

anyway, depends what you want too, ultrasharp for realistic photo things like this usually isn't great

[tawdry escarp]

what's the best for realistic in your opinion?

[tropic shell]

If you need help, I would check out #🤝｜tech-support

[tulip bronze]

I still need a lot of help.
So far all I can really do is load the software.
Ive played around with donoise and some other stuff to absolutely no success.
I want to create tokens in the art style of the left.

[viral elm]

Help me generate similar images and provide descriptive keywords

[tulip bronze]

how do I generate this so the black background is transparent? (and in png instead of jpg)

[dawn seal]

You can always use Photoshop (or Gimp) to remove the background and save it as a png. That's what I've been doing for my VTT assets.

[steel heath]

very good! very realistic!

?

Wow, I hadn't seen that. thank you

[pallid ruin]

That´s so cute :

I´ll try to generate something like that

[tulip bronze]

Success!
The highlighted token is AI generated!

[pallid ruin]

Nice!

[tulip bronze]

Ignore the size being small. I forgot to upscale it in Foundry VTT

[pallid ruin]

The other enemies/npcs are also AI generated? 👀

[tulip bronze]

Nope. They are original art by a guy. The highlighted one is me trying to replicate his style

[pallid ruin]

Aaa , the style seems very similar, congrats!

[fleet goblet]

Don't mind me.

[sterile kiln]

2070 Super.... Noctua Chromax Edition ! 😄

[blazing cobalt]

dam thats hot

[short fulcrum]

Hey, can anyone direct me over to help finding appropriate Lora or Checkpoints? I'm not sure there is a channel for that

[pallid ruin]

I usually download them from civit.ai

Do you have a certain style in mind?

[short fulcrum]

Yes, a friend asked me to make a few images similar to what they made in BingAI

I can get a lot of styles but the non human is difficult

[pallid ruin]

Oof the site doesn´t load rn

When it comes back i´ll try to search for something similar, does he have the prompt? @short fulcrum

[short fulcrum]

Civtai?

[pallid ruin]

yes

[short fulcrum]

I'll try to get a prompt from then when they log back in

[shell sleet]

Hi! I'm the friend! I don't have the prompt.

Someone else made it for me, it feels like they might've made it using BingAI or something.

Oh i just scrolled up, he already said it was made using Bing

I think it's BingAI, though I have no clue.

[quartz thorn]

prompt: beatiful balkan city at the shore of croatia, dramatic lightning

[sterile kiln]

Hey, could you remember me wath this lens effect please?

[jade meteor]

fisheye lens

[leaden imp]

https://civitai.com/models/228897 a LORA i uploaded to civitai. It is trainied with my own oil paintings and inmitates quick oil painting on canvas

i have little experience with training so would appreciate any feedback

[short fulcrum]

I managed to create a similar one quickly with Bing, atm I'm trying to find a checkpoint and lora that matches the style, even without the bunny nose

https://cdn.discordapp.com/attachments/1026530180812001302/1185916086630236240/OIG.png?ex=6591596c&is=657ee46c&hm=34882ca1bf8c8a0ca8c29e076b101394757b38c207b1904299c329b1466588bc&
https://www.bing.com/images/create/anime2c-inhuman2c-a-woman-with-a-dragon-like-head-an/1-657ee40f3f0e47168b5fe7c15235ceca?id=r8113PDUxFnHJcgoyLr%2B%2Bg%3D%3D&view=detailv2&idpp=genimg&FORM=GCRIDP&ajaxhist=0&ajaxserp=0

The second link to bing is the prompt that was used

[dawn seal]

Using Foundry, I see.

[tulip bronze]

Love it 🙂

I'm using David North tokens. Have bought a sizable chunk of his assets, and will eventually buy all of it. But there are obvs still gaps cos 'players'

[dawn seal]

You might want to look at Baileywiki for map assets. He's got some of the best, and many of them are 3D. https://www.youtube.com/@Baileywiki

[tulip bronze]

Ive been using Dungeondraft with Forgotten Adventure assets.
Made this map using that

[dawn seal]

Baileywiki turned that Galleon into an airship with working propellor. 🙂

[tulip bronze]

Niiiice 🙂Working as in animated in foundry?

[dawn seal]

Yes. They have a lot of animated assets in Foundry, and a lot of Dungeondraft assets as well. I'm a member of the All The Things Patreon tier. Worth every penny.

[tulip bronze]

Animated tiles?

[wild sorrel]

no idea what that is, but I like that style

[shadow ore]

hello

im new user of Stable Diffusion and i need some help please

I used this image as a test model and wrote exactly the same Prompt

I got good results like this, but the quality is low and not the same
What can i do?

[wild sorrel]

do you have same Lora?

[pallid ruin]

Oof I tried the same prompt but couldn´t find a model with that style, sorry

[short fulcrum]

It's not your fault. It's a very interesting style that I haven't found a model that kinda fit with it yet, so I will keep on looking

[pallid ruin]

This is the closest I could get 😁

Maybe there are loras trained in that style

the problem is that civitai has like 15k models 💀

[shadow ore]

yes

[winter totem]

Marilyn or Gigi?

[grave reef]

yeah it's called Img2img

[dawn seal]

Yes.

[winter totem]

take your pick

[dawn seal]

Why pick, when I can have both? 🙂

[winter totem]

model i shot that looks like Gigi

[white pivot]

They made a mistake in releasing SDXL Turbo..... I can create anything I want now...

[signal mulch]

oh no not the lovecraft beer

[wispy spindle]

#🤝｜tech-support

[proper hedge]

ok

[frozen nebula]

Experimenting with noise and latent blending to get wild/grungy/chaotic images... getting some interesting results.

[pallid ruin]

It looks cool

[pallid ruin]

This was AI or a sketch? 👀 Tomorrow i´ll try to do img2img to it 🤗

[wispy nest]

hi

[rare magnet]

Hi everyone,
I want to create an image showing people committing crime, being controlled be demons like puppets. Something on the lines of the inage attached here. Have tried different prompts, have even taken help of interrogator and GPT, but the prompts are not delivering. Can someone help me with prompt or contronet method or any other workflow I can use to generate the image?
Thanks

[serene flint]

HxSVD Workflow 😤

[fading cave]

Please generate an image of a yellow humanoid cat walking on the beach in Caraíva, Bahia, Brazil. The cat should be anthropomorphic, walking upright on two legs, and appear relaxed and joyful, enjoying the scenic beach atmosphere with the ocean and clear blue skies in the background.

[shut sinew]

#1072220168534642768

[grave reef]

is this good or bad (150 batches)

[sterile kiln]

Would you please share your settings with us? It's very time-consuming for me to use SVD, and I often get glitches and such, I don't know how you manage to get such a clean result!

[shut sinew]

Didnt they share like the whole workflow

[sterile kiln]

Yes I just saw that, it's been a long day x)

[silver latch]

New Lora #1186697881390956645

[serene flint]

Yep, if you just load the workflow in comfy it comes will all the parameters set for you. There's also a folder with a bunch of template images that you can drop into your comfy window, but they are for v1, still working on getting some v2 templates up
HxSVD Workflow

[pallid ruin]

Cool!

[wispy nest]

Extraterrestrial Encounter: The Robotic Head and UFO
Prompt made by: mixtral 8x7b model.

Render model: Citadel Eclipse

Thanks @acoustic wasp

Holiday Magic: The North Pole Christmas Tree and Elven Workshop
Prompt made by: mixtral 8x7b model.

Render model: Rev Animated 1.2.2

Santa's Whirlwind Tour: The Aerial Menagerie
Prompt made by: mixtral 8x7b model.

Render model: Citadel Eclipse

[tight fossil]

Hey all! I created a tool to fine-tune stable diffusion to generate AI Christmas images of yourself - try on ugly christmas sweaters with AI! Here's one of tswift https://usephortal.com/christmas

[tawdry magnet]

https://i.pinimg.com/564x/e3/0e/ac/e30eac97695a9ced34e2b963b6b101c7.jpg https://in.pinterest.com/pin/940900547136144196/ Hello guys, i would like to train my local sd like this img to img in img2img option. tried with some models but not getting results. kindly let me know if any easy way or right way. #💬｜general-chat #🏞｜general-with-images #🤝｜tech-support #📝｜prompting-help

[olive rose]

does anyone know of a way i can get stable diffusion to generate content within a space? eg the image attatched, i want to have content generated "within" the depth of the building, like 3d volume looking stuff.

im attempting with depth controlnet but it doesnt really use the sapce , it helps with the shape though

[lucid kelp]

I created this image for my daughter's birthday invitation using another system. I would like to try replicating the style in Stable Diffusion, but I'm having some trouble with the prompt formatting since English is not my native language. Could someone please provide some advice on how to phrase the prompt to achieve a similar style? Specifically, what keywords should I focus on including? Any help would be greatly appreciated!

[sterile kiln]

Hey,
I've used what I found interesting in your workflow, but there's something I don't understand.
When I use your workflow without modifying it (v2), the videos are 16 seconds long, but when I import an image instead of your txt2img>choice menu>video generator, the video is only 2 seconds long! 😦
Do you have an explanation please?

Atm, i'm playing with v1 parts, i can do videos but would know how its possible to create 16 sec videos so fast!

[limpid lichen]

i would approach this with two prompts. You know the image you want but are having trouble masking/ producing it. So here's my idea. Mask only the inside space first and prompt what the internal image should be. Invert that mask and then prompt what the church should be.

you could combine the mask with the depth map on controlnet to give some form to the internals too

[olive rose]

Yea that makes sense, I'll give that a go

[final basin]

Dog

[limpid lichen]

oblivion hd

[serene flint]

Hey, thanks for trying out the workflow! So the ability to use an existing image instead of txt2img is a common request from people, and one of my main objectives to build into the workflow.
I had designed a version of v2 that included this functionality the best I could (the user would still have to manually set the txt2img group to bypass) but then I ended up doing the animation extensions instead, adding that second SVD pass, and RIFE interpolation. The way that this is built, as well as just the use of the Image Chooser node, make changing back and forth from loading an image a pretty difficult task.

But yes for v1 you can just unhook the image chooser, and plug in an image viewer

[modest marsh]

why images are so blurry

how can i fix that

[wispy nest]

Matrix: In a Japanese Era
Prompt made by: mixtral 8x7b model.

Render model: Rev Animated 1.2.2

Burning Oaths: Global Apocalypse.
Prompt made by: mixtral 8x7b model.

Render model: Citadel V10

[wispy nest]

try to hi-res

ChessMaster: The Magical Cat
Prompt made by: mixtral 8x7 model.

Render Model: Citadel V10

[modest marsh]

im using something fooocus not automatic 1111 where could i find it

hi-res

[wispy nest]

change resampling

[modest marsh]

ok thank you

[wispy nest]

since only esrgan\swinir\ldsr are there if im not mistaken

Table King: The Magical Cat
Prompt made by: mixtral 8x7 model.

Render Model: Citadel V10

[wispy nest]

I love this one what the heaven 🤌🏻😂

[pallid ruin]

😁 Thanks, the idea is not mine tho, I img2img this one https://civitai.com/posts/1015545

But now i got another ideas thanks to that post haha

[wispy nest]

Use the same seed though

[viscid tapir]

are $300 3070s a good deal or too good to be true?

i need to upgrade from my old 1070

the budget is $300

my best bet for a worthwile upgrade seems to be a used 3070

[tawdry magnet]

any chance to do like this img2img in local sd? #🏞｜general-with-images

#📝｜prompting-help

@grand walrus

[whole jetty]

2D Animation, personality: [Illustrate a wide shot of a bustling city in China, with tall skyscrapers and busy streets filled with people. The atmosphere should convey a sense of excitement and opportunity, showcasing the vibrant energy of the city]unreal engine, hyper real --q 2 --v 5.2 --ar 16:9.

2D Animation, personality: [Illustrate a split-screen shot, showcasing two contrasting images side by side. On one side, show a student struggling with traditional entrance exams, looking stressed and overwhelmed. On the other side, show a student with a Hong Kong identity studying with ease and a smile, highlighting the benefit of the alternative solution]unreal engine, hyper real --q 2 --v 5.2 --ar 16:9.

[tawdry magnet]

@iron remnant

[frail peak]

Bro is literally growing💨💨 on top of a skyscraper and moving it with an heli 💯💯

[wild sorrel]

Looks like toy heli

[wispy nest]

/imagine, #📝｜prompting-help create a caricature bades on this poker player image

[shut sinew]

#1072220168534642768 read on how to use the bot

[hushed marten]

I don't know why always missing shoes or heels

[wild sorrel]

try another model or higher weight on it

[fair gazelle]

/game_statsfemale runner, woman light skin in sports apparel using smart phone in the city, in the style of light indigo and light emerald, website, web design, hero image, lively facial expressions --ar 16:10 --s 250

[drowsy cloak]

Try using ipadapter, it might help

[iron remnant]

absolutely yes 🙂
just google for "automatic1111 img2img"

[wispy nest]

Increase steps.

[remote matrix]

These are awesome! Could I ask you how you made these?

[wispy nest]

i used the RPG model

and orc lora

[cloud sorrel]

I would really like to know which drugs their sales team use

[limpid lichen]

looks like private server pricing for $100 / month. That's the kind of rack you'd rent for game server hosting, not image generation

you can tell cause it has 64gigs of dedotated wam

[wild sorrel]

game server with 1660 is a bit weird too, what game servers are even using gpu?

[icy badge]

Any solution? I am installing automatic 1111 and I have already downloaded python and git

[limpid lichen]

true the vast majority of game servers don't need a gpu to serve up data. but sometimes the admin likes to play on the server itself to wit it needs a head. more games are going to offload calculations to the gpu in the future too.

also brings other use. there's a big market for travelling professionals. Having a system in the cloud you can remote into anywhere is often ideal .

[limpid lichen]

you need an nvidia card to run cuda versions of torch

[icy badge]

I have rtx 3060 12gb and install last version of cuda

[quick current]

Can I post stuff like this? Can I post more stuff like this?

[sterile kiln]

Did you checked PATH setting on python install?

In any case, you have to restart computer after Python install

[half glacier]

Has anybody here ever encountered this issue? It had been working fine, and all of a sudden all of my preview images in the a1111 LORA browser are showing up just a few pixels in height.

[shut sinew]

Have you tried reloading?

[half glacier]

yeah.. :/ even deleted venv and let it rebuild. Definitely a confusing issue.

Almost wondering if there is a setting somewhere I accidentally ticked.

[shut sinew]

Havent seen that one before, works for me right now

[half glacier]

Is a strange one lol.

[sterile kiln]

t2i-adapter_diffusers_xl_openpose looks working good ! 😮

[hushed marten]

• One asian 20 age woman wearing red chrismas dress and wearing black stockings and red heels standing near one chrismas narrow tree and five chrismas box below the tree, indoor and window at the back of the woman, question answering, smart, kind, energetic, cheerful, creative,with sparkling eyes and a contagious smile, information providing, conversation engaging, wide range of topics, accurate response, helpful responses, knowledgeable, reilable, friendly, intelligent
  Negative {{{sexy}}}, paintings,loli,,big head, sketches, (worst quality:2),(low quality:2),(normal quality:2),lowres,normal quality,((monochrome)),((grayscale)),skin spots,acnes,
  skin blemishes, age spot,glans,nsfw,nipples,extra fingers,((extra arms)),(extra legs),mutated hands,(fused fingers),(too many fingers),(long neck:1.3),
  make me crazy

[limpid lichen]

looks like resolution attention issues. are you using stable diffusion 1.5? it would explain the issues

[hushed marten]

is that upgrade to SD2 is better ?

[limpid lichen]

sdxl would be best. 2 is good too but it was sort of a steping stone to sdxl imo. not much use for it anymore

[hushed marten]

i am afraid some of extensions cant be use

[limpid lichen]

if you want low spec, sd 1.5 is avaiable. If you can run sd2, you can probably use sdxl, so 2 makes a lot less sense as the middle child

SDXL many extensions work out of the box. its been out ofr some time now. things like controlnet or animatediff need their own xl models

[white pivot]

Pharaoh is God. God is Pharaoh !

[half glacier]

In case anybody else ever encounters this issue, it was fixed with settings > extra networks > card width / height. The default is 0.

[ripe cedar]

Anyone use the Stable Diffusion plugin for Photoshop? I'm trying to use the controlnet inpaint option but can't seem to get it to work.

[wary olive]

#✍🏼｜rules-and-tos

[spark finch]

@lone jay I can't show you how to do it without images, so for example let's say you make a superman, he's looking clean, fancy, could probably steal your girlfriend... but... his legs... they're missing...

now to be a real super man, he needs legs, right?

That's where the experts, like me, come in.. and you too, if you learn about outpainting @lone jay then you can, as you asked how to do. Give a image a full body expansion. Don't expect results quite like mine, I am an expert.. but you can practice and maybe oneday,

[lone jay]

yeah

how did you do outpainting in sd

is there a tab for it?

[spark finch]

one day, you can become an expert, a real super man, of the outpainting. see, how easy I make it? do not expect results quite as good as mine, but this is the fullest body photo you will ever see, from a partial portrait. But you can get good like me in no time, with just some practice

or you can click that little drop down in a111 which I assume you're using, since you asked about tabs, and choose "poor man's outpainting" which is there by default I think

[lone jay]

take it easy

one at a time

[spark finch]

A1111 -> click the drop down scripts, select poor man's outpainting

should be in img2img tab

[lone jay]

i only see inpaint

[spark finch]

there's a little drop down listbox..

way down below all that

not a button

drop down they look different, buttons you press, drop downs, tend to drop down...

[lone jay]

yeah i saw it

[spark finch]

it's probably empty

till you selec it

[lone jay]

why is it called like that

[spark finch]

poor man's? probably cos it isn't what the rich and famous use for their outpainting needs. probably just demographic marketing

[lone jay]

now what do i do

[spark finch]

outpaint it.

I dunno, I use the node based in comfyui

I think you probably need to type some numbers giving pixel extension of the outpaint canvas, and some other little numbers for this and that, and then generate

I am not sure what it looks like really, I don't use it

[lone jay]

idk how to use it

its very low res

i will just use adobe generative ai

[spark finch]

that works too

but your best bet is to go to google and search for poor man's outpainting and read a guide, cos it'd be more profecient at teaching you something that I don't really know myself.

[lone jay]

ok

i heard there is sd for videos

[spark finch]

Yea, SDV

It makes cool rubbery fish videos I enjoy those.

especially when they really move around

[lone jay]

you generate custom videos or can i turn my photos to live videos?

[spark finch]

the latter use animatediff

or SDV

[lone jay]

?

[spark finch]

if you want to make a photo into a movie use either of those.

SVD does specificlaly that. really

Animate diff with generate custom videos, I think SVD needs a still image

[lone jay]

yes thats what i mean

make my own images to videos

that would be cool

how can i do it

[spark finch]

SVD

[lone jay]

is it easy to install?

[spark finch]

It's just a model.

https://huggingface.co/stabilityai/stable-video-diffusion-img2vid

[lone jay]

oh

then what do i do

[half glacier]

https://youtu.be/HOVYu2UbgEE?feature=shared

[serene flint]

You can also use my HxSVD workflow to animate from text prompt, and animate text with it, but it doesn't have img2video built in yet https://civitai.com/articles/3355/hxsvd-harrlogosxsvd-txt2video-comfyui-workflow-generate-and-animate-text-with-svd-v2-out-now

[sullen escarp]

anyone got an idea as to how to stop SD introducing unwanted colors? i have the following source image, and i'm trying to introduce cybernetic glowing white eyes. no matter what positive or negative prompts i use, i can only get SD to do colorful eyes.

here's the source image:

here's the kind of image i always end up with:

it always gives blue eyes. if i try to negative prompt away from blue, i get green, or yellow, or red... if i keep extending the negative prompt, in the end the prompt just seems to be ignored and i get various rainbow shades

i've tried heavily weighting terms like "colorless", "white", etc. at some point the sampler just produces overcooked noise

i've tried painting over the eyes in an image editor to pure white. SD just dutifully restores the white area to normal looking human eyes 😆

obviously i can fix this in an image editor, but i'd prefer to get as much of the desired result as possible out of SD prior to editing

[grave reef]

you probably need a Lora

[sullen escarp]

good point

[dry crow]

What are your settings for this?

[sullen escarp]

@dry crow an image piped into a euler sampler at 30 steps with ~0.7 denoise, using the epiCRealism checkpoint. the original image is masked covering the original eyes, but with quite a high feathering (about 64 pixels)

unfortunately i've closed the workflow now, but the positive prompt was something along the lines of white eyes, glowing eyes, cyborg

[ruby quarry]

I love the outcome

[fleet goblet]

OpenDALLE 1.1 is pretty good.

Incredibly consistent imagery.

[pallid ruin]

Is it a new model or its a new AI? 👀 it looks cool

[fleet goblet]

An SDXL model.

It isn't "new" it's just a new version of an already-existing model, OpenDALLE.

I'd say it's pretty good.

This of course is with 2 loras as well

But it CAN do pixel art and voxel art without them

and with removeBG I can just...

there he is

[fleet goblet]

I wanted to stablezero123 him to get a front-facing image

But yeah no I can't get a consistent image at all lol

I can't wait for fully-functional 3D models from AI.

The creativity will be unlimited.

[blazing cobalt]

I need better computer hardware give me some advice bruvs

[wispy nest]

Uhg, now I must download OpenDalle and try it out lol

[sterile kiln]

it sounds crazy, I've never heard of it.

[snow flame]

Pickles

[solid comet]

1

[wispy nest]

Help :(

[plucky copper]

Hi everyone
I encountered a problem in the installation of stable diffusion using auto1111 and no matter what I do, it is not solved
Please guide me if you know where this problem comes from ❤️
I will post the photo of anaconda as well

[high veldt]

请为我创建一个论文的图形摘要，创建图文摘要时需要注意以下几点：

清晰的结构：图文摘要应该有明确的起点和终点，遵循自上而下或自左向右的阅读顺序。

重点视觉元素：突出展示与研究最相关的视觉元素，如关键的发现或实验步骤。

新发现的强调：专注于当前论文的新发现，避免包含不必要的先前研究细节。

避免猜测性特征：图文摘要应基于实证数据和研究结果，避免基于假设或猜测的内容。

简单的标签：使用简洁清晰的标签描述图中元素，以Arial或Times New Roman字体呈现。

避免文字过多：图文摘要应以视觉元素为主，尽量减少文字使用。

清晰的焦点和过程：每幅图像应明确一个要点或展示一个过程，以便观众快速理解。

避免杂乱元素：确保图文摘要中的元素不会分散观众的注意力。

尺寸和分辨率要求：尺寸为长1328像素，宽531像素，分辨率高于300 dpi。

颜色的有效使用：合理运用颜色增强图文摘要的美观性和吸引力。
技术规范遵守：

确保尺寸、分辨率和字体符合您之前提到的规范（例如，Arial或Times New Roman字体，分辨率>300 dpi等）。
颜色的使用：

使用颜色来增强信息的传达，比如不同的化学处理方法可以用不同的颜色表示。

[craggy night]

I keep getting images generated with this artifact banding across the image horizontally, just higher than half way. Tried different resolution and settings but always seems to be there to some extent.. any ideas?

[white pivot]

One of my post was removed from reddit due to "low effort content and ai art"... Ai art bashing is brutal 🤣 Above all they have no idea how long and hard it is to create a lora allowing such details and flexibility...

[spring mist]

generally speaking many mods are out of their mind

[sullen escarp]

that image is 540x612. is that the resolution you're generating the image at?

[craggy night]

No I generated it at 960x1088 in that particular case but I had tried a few different resolutions and all have a similar thing occurring

[sullen escarp]

frequently models tend to prefer you stay at square images, either 512x512 or 1024x1024. often people will upscale and crop to get the actual size they want, or generate multiple images side by side and stitch them together

i'm not saying this is your problem, but i have seen all kinds of artifacts when i strayed from the normal sizes

[snow flame]

All you did was push a button 😅😅

[snow flame]

Banned from a local one for not using Flair 😅😅

[wispy egret]

In my opinion, the biggest difference between Dall-E 3 and Stable Diffusion XL
Is how well it listens to your prompt for dealing with multiple things.
It seems that SD everything has a single "actor/star".
When you try to do more than 1 person or animal, whatever, it doesn't understand.

SDXL prompt:
An HD 4K photo taken by a professional photographer of multi colorful christmas lights at night, the xenomorph alien is wearing a santa claus hat and is sitting at the table banquet of hot food, there is also a large bulldog sitting at the table with the xenomorph during a backyard christmas party celebration event with christmas decorations in the yard and colorful christmas lights on the house and fence, at night

This is me trying to force it over and over to have 2 characters, but it just insists on combining or duplicating or omitting, while Dall-E 3, just did it and i could have greatly simplified the prompt and got the same result.
I didn't actually want them sitting side by side, but it was an attempt to make it understand there was more than 1 focus, but it didn't work
It could be model dependent, I did only use 1 model because it takes a long time to load other models,
but this seems to be an issue I always run into.

So, I hope that the next BIG Stable Diffusion model that comes out, will let you do more things at once, like Dall-E 3
I would much rather have that, than to have an increase in visual quality. if they just go to 2048x2048 images and same level of ignoring prompt focuses, that will be annoying. It may not be possible to have multiple focuses with models that can fit in the VRAM of a GPU. I don't know.

This is done with Dall-E 3.
I gave it a long list of requirements and the only thing it left out, was a sandcastle.
Stable Diffusion XL could never follow this many details.

[heady anchor]

because dalle3 is not a single model like sd

dalle3 does a several steps process like Prompt > LLM for prompt augmentation > Multiple Images > Pick a score > segmentation > control net > image

[wispy egret]

is it possible to do that on an 11gb VRAM video card if you wait long enough?

[heady anchor]

probably not,the LLM would eat all of the vram

[wispy egret]

I guess you are saying the difference is that Dall-E 3 uses basically ChatGPT to help?

[heady anchor]

yea plus other models like the score one,it outputs several images then theres probably another model that scores images based on composition and picks the best one

[nocturne oak]

Wasn't easy; isn't perfect. But it was zero-shot prompted.

Positive was: (xenomorph wearing christmas hat:1.2) sitting across from his (large bulldog:1.05), sitting at banquet table, hot food, backyard christmas party, christmas decorations in yard, colorful christmas lights on house and fence, at night, HD 4K professional photo

Neg was: two xenomorph aliens, two bulldogs, watermark, signature

Took several gens to get there.

Dynavision's output ^

I did add two subjects to a part of the positive as well.

[tropic tide]

can try cutoff(for comfyui), also regional prompting for A1111

[tropic tide]

Damn it's hard to get what you want. I'd say compositing is the way to go, or basically several passes by various means (inpainting, i2i, controlnet, etc). you can't keep up with the powerhouse that is OpenAI or any of cloud based in-house AI image-generation..

[dawn seal]

And this is why I worship Anubis.

[kind bear]

Hello everyone, can you edit your photos in the group?

[cinder crystal]

my dreams keep coming out as videos when i want image

please advise

[tawdry magnet]

this image size is 560 x 1000, i want to resize to 1151x2303 with high resolution. how can i do that?

#📝｜prompting-help

#🏞｜general-with-images

[wispy nest]

图生图

[sterile kiln]

You should set Watermark in negative prompt

[proven snow]

With what? it will always be very hard to make it write text

[proven snow]

Here, 1232x2200

[wild sorrel]

up denoise a bit, it's too pixelated

[honest geyser]

Plese someone help me. Trying to use Dynamic Promt

[grave reef]

#🤝｜tech-support

[uneven kernel]

Need some help, need to ideate on what the background should look like. Is there something in control net I can use to ideate some background ideas?

[grave reef]

IPadapter probably the only thing but even then you still have to give it a refference picture

[uneven kernel]

I see, will check it out

[tawdry magnet]

Can we do this in extra?

[proven snow]

Do what?

[proven snow]

The original image is noisy, I didn't denoise

[tawdry magnet]

May know option for upscale in sd?

[proven snow]

I didn't use SD for that. But you should be able to, just throw it in and see what happens, and mess with the settings until it perfect. Start with just upscaling without a lot of changing so that it comes out the same as the original

I haven't used SD Upscaling much, I don't remember what settings are good

[bronze sky]

Did this eBay listing use inpainting to hide the model number and output voltage of an AC adapter? Lol

[sterile kiln]

what you're trying to do is much easier with X/Y/Z Script

[fair jasper]

new unreal tournament looks pretty good

[pearl loom]

hello i need help , i d like to understand how to define this type of art if i wanted to use dream to make it from scratch
what are gonna be the key words? and what about the type of art , type of design, type of colors u know ...all those things?

this image is from youtube but it is the exact type of art i wanted to understand

[sullen escarp]

@pearl loom try adding references to japanese art periods... "edo period" "muromachi period"

[sterile kiln]

nice images

[desert gulch]

111

[cosmic garden]

/a girl

[pearl loom]

ty my friend, may i also ask what is the type of this image
i mean the type of paint used, the shadows those things, sorry i m not good in english (neither in my native language hahahha to describe what i m searching for)

[dusk obsidian]

Грустная девушка

[nocturne cargo]

first image is the area i want to inpaint in automatic1111 and the second image is the image i have in controlnet, third image is the result im getting. as you can see, there's hints of the logo in there. but how do i make it more clear?

can anyone guide me thru this?

[sullen escarp]

@pearl loom at least some of them are ink paintings. i'm not sure there are any specific terms for the lighting besides "dark background" "shadows" etc. i think you might have difficulty getting something exactly like the image you posted directly, but you might be able to composite a few images together to get something similar

@pearl loom you might get good results with some kind of img2img setup. shove an MS Paint quality image into it marking out the general colours of the regions that you want and use a positive prompt with something like:
japanese period art,(dark background),shadows,skeletons,ink painting,highest quality
and a negative prompt of something like:
text,watermark,monochrome,(light background),frame

just playing around with the SD 1.5 model, i got this:

it's at least interesting, if not quite what you were looking for 😅

[lofty tree]

I wanna die

[nocturne oak]

seems like an overreaction to the situation...consider life

[lofty tree]

consider using inpaint everything extension on a laptop

I made it pretty realistic tho

[brazen ledge]

yall i need help, im trying to inpaint on the car's paint and change it's color and style (like reflective or matte), while maintaining the overall shape. Any recommendations on process?

original color was red ^ prompt was matte blue

[sullen escarp]

@brazen ledge i'd be inclined to do a really rough edit in an image editor. just select (don't need to be precise) the car and shift the hue around to the color that you want, and then use that as the input. i think the problem is that unless you're 100% denoising (which probably won't maintain the shape of the car), those underlying samples are going to be red, and it's going to be quite awkward to convince the sampler to produce blue

others might disagree

[brazen ledge]

wat do u think about controlnets

[sullen escarp]

not sure if it's applicable for this. ipadapter might work for this though, come to think of it. it's pretty good for keeping elements of the image consistent

[brazen ledge]

does it require a refernce image?

im not sure on that part

[sullen escarp]

it's geared towards img2img

https://github.com/tencent-ailab/IP-Adapter

[soft escarp]

I wanted to say that's cool but now I'll sound like a douche for waiting for the video

[pallid ruin]

Yoo it looks amazing

[pallid ruin]

lol im stealing this

[sullen escarp]

heroes of might and cabbage

[serene flint]

This can be done with HxSVD as well

[vestal shale]

would love to know everyone thoughts on this? Is this is even possible? https://javascript.plainenglish.io/instagram-wont-exist-in-5-years-this-is-why-9bed68a710e7

How do you use stable diffusion to get such results?

[nocturne oak]

Link for that article that removes the sign-in wall:
https://archive.is/20230726084458/https://javascript.plainenglish.io/instagram-wont-exist-in-5-years-this-is-why-9bed68a710e7

[vestal shale]

doesn't work

[nocturne oak]

Try this one:
https://removepaywall.com/https://javascript.plainenglish.io/instagram-wont-exist-in-5-years-this-is-why-9bed68a710e7

[vestal shale]

it is just ask for google sign in, quite doable. I can read after google auth

[nocturne oak]

The article is kinda shit...but that second link does work for me.

[vestal shale]

it doesn't

[nocturne oak]

Bottom line is that Insta will remain if it can transform.

[vestal shale]

it shows only 1/44th of the article

[vestal shale]

how do you say that? The article is mind opening. You couldn't even read the article 🙄

[nocturne oak]

How do I say it? Because I read what I needed to, which is what was available to me. The bottom line is that Insta is a cesspool just like every other mainstream social media.

People will gravitate to it if it can remain relevant in any way...but literally anything could cause it to disappear, including AI. However, at the same time, it might not. Why? Because it's all dependent upon people. People are fickle and sometimes unpredictable. In a span of 24 hours, the next thing could be canceled because one person says something stupid.

So...will it stick around? Maybe. It certainly will need to remain relevant. What kind of thing relevancy is in 5 years is anybody's guess.

But that's true for a ton of stuff, not just Insta.

[vestal shale]

The article is just 1/4th what you see in the links you have posted

there are many images

[lofty tree]

I use a1111 or whatever it is and get stuff like this

[lofty tree]

i love using a laptop

[sullen escarp]

it's character forming

or ... destroying. i forget

[white pivot]

Jaffa !

[sterile kiln]

Svd?

[white pivot]

pika lab

[lofty tree]

I HATE DOING THIS ON A LAPTOP

[pallid ruin]

Have you tried with lcm lora and sampling? @lofty tree its faster and the quality loss is not that bad

[sullen escarp]

small thing i put together today that others might find useful:
https://github.com/io7m/com.io7m.visual.openpose_rig

[hazy zenith]

Nude girlfriend

[sterile kiln]

What specs?

[lofty tree]

Some shitty old laptop

It got to 30k seconds and I just closed it

Usually it only takes like 1K

[lofty tree]

Idek what that lcm Lora is

[grave reef]

lcm model's render at 2x speed because it uses half the memory to process

[lofty tree]

How do I get a lcm model

[pallid ruin]

https://huggingface.co/latent-consistency/lcm-lora-sdv1-5

In case you dont get the LCM Test sampling you can install the animated diff extension (but you dont have to use it, it just adds the the lcm sampling)

https://github.com/guoyww/animatediff/

[frosty inlet]

Hey everyone, I encountered an issue while downloading a localized stable diffusion on windows 10

what does this mean?

it also sometimes says port 443 or failure of receiving data from the peer

I figure it's something to do with the proxy settings, but not sure how to access that :/

[pallid ruin]

You get that error while trying to git clone? @frosty inlet

You can try downloading the zip file and extracting it

[frosty inlet]

yup, I was trying to do the git clone

[pallid ruin]

Try this way, maybe it will work

[frosty inlet]

ah, let me try, thank you!!
will come back in a sec

[pallid ruin]

After you extract it in some folder you can download a model and put it in the "models\Stable-diffusion" folder

but if you open the webui.bat it will download the default sd1.5 model

[lofty tree]

thank you man

this took 179 min to generate

[wild sorrel]

179min?

[subtle meadow]

Easy re-wear in Comfy 🙂

[frosty inlet]

@pallid ruin it worked! I was able to download it no problem

only issue now is when I put http://127.0.0.1:7860/ into my web browser, the page just says I can't reach this page

[wild sorrel]

try
localhost:7860/

[pallid ruin]

179 minutes damn, what settings are you using? 👀

[lofty tree]

Nothing special, I’m just on a shitty laptop 😂

[lofty tree]

Yep

[sterile kiln]

You could get some optimsation with some launch arguments, thats why i asked you about your specs.

[sterile kiln]

And not so shitty if you can run SD

[lofty tree]

intel iris xe graphics, 11th gen intel core. idk Its an old work laptop from family

[lofty tree]

barely runs but true haha

[sterile kiln]

This should help to know what GPU you have (take care, at bottom, you can change GPU, because on most of laptop, there is 2 gpus, IGP Intel for low perfs, and Nvidia/AMD ones for high perfs)

https://www.techpowerup.com/download/techpowerup-gpu-z/

[lofty tree]

Tiger Lake GT2 💀

it says multi-gpu is disabled

[sterile kiln]

That explain why you need 179 min to gen instead of 1 minute

[lofty tree]

idk, have i set it to something wrong

--skip-torch-cuda-test --precision full --no-half

those are my commands

I have hres fix on to 1024x1024

and picxreal10 checkpoint

[sterile kiln]

Its an 1.5 based model, with a middle GPU you can run 1024² in 30 seconds

second-hand, you can find 2070 Super models like mine for around 200€ (or even less), which will be more than enough to do much better.

https://www.leboncoin.fr/ordinateurs/2464876598.htm

[lofty tree]

have I done something wrong when installing then?

how Do i check what sd version im on

[sterile kiln]

Nope, but i think one of your args set your A1111 on CPU only

[lofty tree]

which one should I change?

[sterile kiln]

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

Just do a test, if you see a generation set your CPU to 100%, and your GPU not, its your problem

[lofty tree]

u can browse civit.ai

probs find a lot of things like that

then just check what they used

[lofty tree]

yeah cpu is up and not gpu, Is one of my commands doing that?

set CUDA_VISIBLE_DEVICES=0

would that do it?

i looked on github page

nvm it didnt even run

[limpid wren]

how could the models be called? no idea what to look for

[lofty tree]

go to images

[lofty tree]

there will be loras, checkpoints etc

[limpid wren]

oh kk

[sterile kiln]

Never tried to use none CUDA GPUs, i can not tell, but you should remove all not usefull, and use --lowvram, because i'm not sure of what you can do less.
In any way, i'm not sure you can use this shit GPU

I know real dedicated Intel GPU can be used (Intel Arc), but IGP...

[lofty tree]

hmm alr, thank you

[sterile kiln]

Just do a simple test, remove all args, set --lowvram and try a gen

[lofty tree]

ok

i need to keep the --skip-torch-cuda-test tho otherwise it wont start

i need to have the others too i think

Warning: caught exception 'Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx'

says that

[sterile kiln]

At installation instructions, about Intel GPUs :
https://github.com/openvinotoolkit/stable-diffusion-webui/wiki/Installation-on-Intel-Silicon

[lofty tree]

it doesnt show up in my scripts

[sterile kiln]

its an different Git, a different installation

[lofty tree]

do I have to uninstall stable-diffusion-webui and install again?

wait i found it nvm mb

[sterile kiln]

No, just a do a new folder, follow steps to do another install

[lofty tree]

wdym? i create another folder...?

[sterile kiln]

Go out of Stable-Diffusion-webui. Because its standart Automatic1111. This is another git, is another installation.

So go anywhere else on your computer, and install this Intel version

[lofty tree]

oh alr ty

[lofty tree]

do I add the args i had before?

[sterile kiln]

No

[lofty tree]

RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

im gonna do something it says on reddit rq

[lofty tree]

@sterile kiln im having trouble

[pallid ruin]

You have another error? @lofty tree

[lofty tree]

it said RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check so I did what the guy said on reddit, it didnt work, so i tried again. Now my venv isnt re apearing. I am trying everything again

thats what reddit guy said

[pallid ruin]

Add this --skip-torch-cuda-test to the commandline arguments on the webui.bat

to see if it works

it works for me

on automatic1111

it is slow but not that slow like +100 minutes @lofty tree

try this commands

--use-cpu all --precision full --no-half --skip-torch-cuda-test

and use lcm lora

with 8 steps and 1.6 - 1.7 cfg scale

[lofty tree]

whats that mean

nvm it was just a space on one of my args

[sterile kiln]

The objective is to not use CPU, this args do

[pallid ruin]

Aaaa my bad

Its possible to use the integrated gpu? 👀

[lofty tree]

I thought there might be but im not so sure now

[sterile kiln]

I don't think, but i didnt find anything about this on the web

[pallid ruin]

I told him to use the cpu because I tried and I generated a 512x512 in about 2-3 minutes, and with the earlier commands that he had he got a generation in about 170 minutes if I remember correctly 💀

[lofty tree]

180 😦

[pallid ruin]

😭

have you tried again with the CPU commands to see how much time it takes now? @lofty tree

or you got an error or something?

🥺

[lofty tree]

i will soon

its 6:30am and i want go to bed and cry

[pallid ruin]

U.U dont worry it will run faster, also there are other repos for cpuonly, i´ll search them and post them here later

https://tenor.com/view/allex-ano-slay-zee-me-gif-11187196124230634497

[lofty tree]

you are very kind

I'm too tired to sleep anyway

[lofty tree]

btw that thing never showed up on my scripts its sad

ill check back later maybe i did something wrong

[lofty tree]

its still around 18 minutes for a 512x512, 20 step image

and I dont think i have a bad cpu, its 11 gen something

[pallid ruin]

Yeah regular gens are slow, you can use lcm with 8 steps and 1.7 cfg scale, withouth lcm mine took like 6-8 minutes I think, and with it it took 2 mins 40

[limpid lichen]

i always forget that thers a 3rd player in the dgpu market now. i think intel xe cards need the directml

[lofty tree]

what is lcm again?

[limpid lichen]

latent control model (iirc)

its a lora you throw into the mix and allows you to do way faster gens

[limpid lichen]

https://github.com/lshqqytiger/stable-diffusion-webui-directml there is this fork of automatic1111 as well. it's built around directML without needing to custom script it all

[lofty tree]

what does it do/

[limpid lichen]

it's a custom fork of automatic1111. same program, just with directml precoded into it

hurray open source

[lofty tree]

what does that do tho... sorry if im dumb but does it make it faster?

[lofty tree]

why did that never show up for me

[limpid lichen]

directml is the hardware translation layer. instead of cuda. Cuda iis for nvidia GPUs. directML is for everything else.

DML is actually slower than CUDA but.. i mean.. it works if you don't own nvidia hardware at least

[limpid lichen]

the install script likely failed. i don't think thye've updated that side for 1.7 yet

[lofty tree]

if its slower, why would I download it? is it faster since I don't have nividia?

[limpid lichen]

its faster than cpu mode

[lofty tree]

im trying it but its still only using my cpu

[limpid lichen]

still? this should be a whole new installation

[sterile kiln]

Because you probably don't have the same CPU than him

[pallid ruin]

I have a i5 8400, but I even saw someone with a raspberrypi running sd 1.5, and it didn´t took that long

[blissful bramble]

https://youtu.be/Gsj6ToFTGgc?si=QmoqvQUxxtd8ly5V

😱

[sterile kiln]

Yeah, @lofty tree should send us one of those images that took 180 minutes to set up.
@lofty tree
Let me explain: in the images (/output/Txt2img/PNG FILES ONLY), SD saves the generation parameters in the images' EXIF data, so don't send us .jpg files.

[limpid lichen]

3 hour image is probably because the mainbus was being shared by another constant process. Virus scan. Bitcoin miner. Game. Constant interupt requests on the mainbus will turn a 20min operation into a 3 hour op no problem

[sterile kiln]

Or just not enought of Ram

[limpid lichen]

yeah if it has to keep swapping from virtual memory to physical memory too. that's all mainbus traffic

and slow memory

point is, there's plenty of ways cpu mode can suck even more

[sterile kiln]

in France, with 160€ you have a second-hand 2070S

[limpid lichen]

each one? or total?

[lofty tree]

each one

[limpid lichen]

so that's 15 hours there? maybe less because one isn't 1024x1024

[lofty tree]

the one that is just a face was more but i cropped it

but yeah 13-15 hours

[limpid lichen]

[sterile kiln]

50 steps, CPU mode (what CPU do you have?), with just 8 gigs of ram...
As said @limpid lichen, you bottlenecked by your hardware. Windows will use Pagefile.sys and he'll be dragging his dick along, slowly

[limpid lichen]

my money is on an hdd page file

[lofty tree]

11th gen intel i-7

im not tryna drain my bank acc for a good pc/laptop 😭

but yeah i get that i wont be able to generate ones fast, i just felt like i cld get a little faster

[limpid lichen]

the free part of free and open source software is how they get cha

"have all this cool software but you'll need to buy all this"

[limpid lichen]

you can if you get direct ml loaded right and it chews numbers through your intel gpu. i think that's a dedicated one? not the integrated right?

[sterile kiln]

YOu could use Google Colab, you can use it for free, but with limits

[limpid lichen]

my 12th gen i7 's integrated gpu shows up as intel uhd graphics , not xe

[limpid lichen]

yeah but what about the gpu. i saw intel xe ealrier. directml should load up on that

[lofty tree]

is that googles ai image thing?

[sterile kiln]

4 cores 8 threads with 8 gigs of Ram xD
There is no solutions for you laptop

[limpid lichen]

maybe not

[lofty tree]

at least I didn't buy it

[sterile kiln]

it's a miracle you managed to use it at all.

[lofty tree]

lmfao

[sterile kiln]

No, it's a remote provision of GPU power.

[limpid lichen]

what we need is stable diffusion enabled instruction sets on the cpu itself, so we can just use ddr5 memory

comon intel. do something. go hard again

inspired diffusions

[sterile kiln]

Its not possible.

[limpid lichen]

not on current chips. intels doing something though with ML focused instruction sets. vector neural network something or others. avx512 is a big expansion that is untapped yet

of course it woudln't just take that either. someone would have to do what facebook had done with pytorch still. reengineer everything for a fundamentally different architecture

[faint prawn]

The new year's first creation!The traditional British costume that with the British ethnic style and symbolic orange rose head ornament,which shows the genuine love towards life and nature of the British people,is more than beautiful!Hope you all like it!Suscribe me,and I'll show more costumes all over the world!

[wise bolt]

Generate greeting card with New year, with words: “Нехай у цьому - 2024 році подих Господа прожене ворога і принесе в Україну великий Шалом!”

[ripe cedar]

What's the best way to add details to an image without major changes? I know of putting an image in img2img with low denoise with controlnet through a few cycles but wondered if there's another better way

[narrow orchid]

I found this in a tiktok comment section from 2 years ago

[pallid ruin]

so cute 😊

[pallid ruin]

You can use the this lora https://civitai.com/models/82098?modelVersionId=87153 , there are other ones that do the same thing

[ripe cedar]

I do use that but going too much messes it up. That's why I do it in a few cycles but idk

[past lance]

po

[quasi parcel]

Which checkpoint i should use?

[sterile kiln]

None of these

[quasi parcel]

why?

[sterile kiln]

Models (ckpt, or safetensors), are divided into 3 parts, models based on 1.5, then evolved to 2.x, then finally today on SDXL.
I recommend you go to Civit.ai and choose a model based on SDXL. Personally, I use this one a lot: https://civitai.com/models/139562/realvisxl-v30-turbo

A good example of what can do SDXL 😄

[quasi parcel]

Ahahaha

Woah

[sterile kiln]

I recommand you t not take Turbos ones, they are faster but i think with less quality

[quasi parcel]

Where i can find the Model Without turbo

[sterile kiln]

on the same page, just check on top, different models are available

[quasi parcel]

Thanks

[sterile kiln]

what specs do you have?

[quasi parcel]

A GTX 1660 Super but heavily overclocked

[sterile kiln]

Oh, hmmm. If I were you I'd look into launch args. You'll need --medvram, maybe even --lowvram

[quasi parcel]

i use --medvram

Is this without turbo right?

[sterile kiln]

yep, baked VAE is to say that it is integrated. There's absolutely no need for an overhanging VAE.

[quasi parcel]

Wich Settings have you used for this bc i get a not pretty thing

[sterile kiln]

768x1024 > Highrex Fix to 1382x1843 (x1,8) > (Extra tab) upscale x2

[quasi parcel]

With other extension? bc in a lot of video i see that have another setting besides the checkpoint

[sterile kiln]

Some personnal LoRAs, and no extensions.

[quasi parcel]

ah ok

[quasi parcel]

sorry for the stupid question but where i need to set those setttings?

[sterile kiln]

Resolutoin is easy to find, for the hires fix, simply unroll the menu with this name to activate it. Hires FIx is NOT upscaling, it's img2img. Then you need to play with denoising to get closer to the base image or not. The lower the denoise, the closer the image will be to the base image, the higher the denoise, the more creative SD will be.

[quasi parcel]

I have set those but i dont find the upscale and the hires flx

ahhh ok

i saw hires fix

are ok those settings?

[sterile kiln]

yup, it is

but x2 is high take care

[quasi parcel]

ok, now i try

[sterile kiln]

and 0,7 denoising is high too, if you want get closer than original, lower it

[quasi parcel]

ok and the sampling method and upscaler are ok?

[sterile kiln]

i was talking about upscale with x2