🏞｜general-with-images

[quasi parcel]

yes sorry i have put 1.8

[sterile kiln]

and you should select a different upscaler methode, i use x4 ultra sharp, the best to me

[quasi parcel]

I have not the x4 Ultra Sharp

[sterile kiln]

the first step is to search for the right seed. If you click on the recycling logo, you'll keep the seed of the image generated just before, but if you click on the dice, the seed returns to -1, so it's random. So first close the Hires Fix tab to find an interesting seed, then once you've found one you can keep the seed and regenerate with Hires Fix.

[quasi parcel]

ok

regarding the sampling method and steps?

[sterile kiln]

this one maybe
https://openmodeldb.info/models/4x-UltraSharp

in stable-diffusion-webui\models\ESRGAN

[sterile kiln]

I personally use between 30 and 50 steps

[quasi parcel]

now i try

[sterile kiln]

another thing to know, open a new tab in your web browser with SD, if you drag&drop an image in the prompt field (then press the blue arrow under the "generate" button), it will import the settings used for the image in question. So try this one, it'll import my settings.

[quasi parcel]

cool

[limpid lichen]

i find that chrome supports drag and drop and pasting images into fields so muhc better than firefox. i browser main with firefox but i load the various webui's into chrome

[fallen blade]

Has anyone managed to generate 2d characters for games at different angles consistently?

[limpid lichen]

i'd start with 3d prototypes and then apply a character lora to them. there is a cool thing for 2.1 if i remember though.

https://civitai.com/models/3036/charturner-character-turnaround-helper-for-15-and-21 @fallen blade 1.5 too!

https://civitai.com/models/104099/reference-sheet-model-sheet things like this too

https://civitai.com/models/86207/character-head-concept-poses-openpose-3d-for-depth-maps tool kits like this too

[fallen blade]

@limpid lichen ty so much ill take a look at all when i arrive home

[bitter dragon]

guys I was looking at a photo on civitai and I see "30 nodes" in workflow, what does it refer to?****

[sullen escarp]

@bitter dragon 30 nodes in the comfyui setup they used to produce the image: https://comfyanonymous.github.io/ComfyUI_examples/area_composition/workflow_night_evening_day_morning.png

[bitter dragon]

ooh thank you. And for you, which one is better? a1111 or comfy?

[sullen escarp]

i prefer comfyui, but i have a software engineering background. i like the way the configuration can be saved, and the explicit nature of it (so you have some idea as to what's going on underneath)

i really dislike graphical node-based environments, though. i wish they provided a text-based interface. you spend about 70% of your time dragging nodes and wires around (this is a problem common to all node-based environments, blender being the other big one that everyone knows)

[bitter dragon]

okay thank you!

[faint prawn]

What's up guys your Chengdu guy,Last name Dai,is back to Discord!The traditional German skirt dirndl, with beautiful design and pattern style, is capable and smart;A simple and refined headdress that shines on the girl's head;A brilliant emerald necklace lights up the silent forest between the mountains and rivers.

[wispy nest]

how can we fix this pls tell

[crystal pollen]

remove any/all embeddings in your prompts that are not for the checkpoint used : SDXL embeddings don't always work with SD1.5 checkpoints and can error out, vice versa

[wispy nest]

nvm i got it fixed

[wispy nest]

but i dont understand this whatsoever

[crystal pollen]

can you screenshot the workflow before the error shows

[wispy nest]

hey

can you tell me

how do is search for a node in my workspace

whats the shortcut

[crystal pollen]

doubleclick empty space

[wispy nest]

and how do i add

ultimate sd upscale

[crystal pollen]

https://github.com/ltdrdata/ComfyUI-Manager add this to your custom nodes folder and restart comfyUI .. it adds a manager button letting you install missing nodes and install tons of nodes from within comfyUI

[wispy nest]

https://huggingface.co/FacehugmanIII/4x_foolhardy_Remacri/tree/main

i downloaded this

where do i put it

[crystal pollen]

Models/Upscale models

[wispy nest]

ok i did it but will i loose my image when i restart it

i hope not

[crystal pollen]

you will upon restart

[wispy nest]

yeh i did

and i still do not have the ultimate SD upscale

lets talk in dms

[exotic temple]

Hey I was just wondering, what would be the closest art style to the one used in the APB Reloaded loading screens? It's some offshoot of graffiti, but I don't know if there's a more specific name or not.

https://youtu.be/_fQyEN5lJRw?si=IqMu8HTSIgrnAGmD

[winged forum]

Hi, does anyone know how I can use more power from my graphics card to generate in comfyui. it seems to only use 20% to 30% or less and I have a 4060ti 16gb.

[grave reef]

that's normal, mine only uses 4% lmao

if you click the gpu and look at the bottom you will see it's using all it's VRAM tho

[tepid relic]

/Prompt：A beautiful sunset，carton

[wild sorrel]

where's carton tho

[wispy nest]

is proper style transfer a thing yet? like turning a photo/3D render into a consistent art style but more or less keeping all details the same, the PhotoLab mobile app can do it (and says it's "AI", not sure to what extent) so there must be a way to do it locally, this is one of the PhotoLab presets but one day it could just disappear and I want my stuff to be consistent, the app requires an internet connection too so I can't just store it somewhere forever if it gets removed from the app

[faint prawn]

Yo,hello guys in Discord here and I'm Dai,the panda city's good boy！Today I'll show you the Dutch costume！

[kind kayak]

蓝色毛衣小女孩

[elder cobalt]

Hello there! I'm curious about the cool image-making magic the bot uses – which stable diffusion model is it? Have there been any other models before this one? I stumbled upon a dataset called DiffusionDB with prompts and images that apparently came from this Discord server. I'd love to know which model or models were behind creating those images! Thanks😊

[shut sinew]

Please dont ask the same question in multiple channels

[elder cobalt]

ok sorry

[fallen blade]

Playing around with resizing old 2D sprites from the game Ragnarok Online following the instructions at https://www.reddit.com/r/StableDiffusion/comments/13f7jy7/i_remaster_street_fighter_old_sprites_using/. I'm not able to bring out as much detail in the upscaled images. Any ideas on how to add more detail to the upscaled models?

[sterile kiln]

increase denoising str

and use SDXL models, not 1.5

[fallen blade]

Can i add sdxl with aaaa1? Thank you btw 🙂

[sterile kiln]

Obviously, its highly recommanded

[fallen blade]

will try it then 🙂

[uncut otter]

Anyone know why my images are finished like this as if they were interrupted halfway, I used sd before and the output was fine but now that I installed it again it does this, I used realistic vision v6, DMP++ 2, 512x512, 50 sampling steps

[sullen escarp]

@uncut otter what noise schedule? any LoRAs used? i've seen this happen, but it's never been the same root cause twice

[uncut otter]

No LoRAs used and if you mean denoising strenght it is 0.7, I also have an AMD GPU so I had to make some workarounds in order for it to run but when I used it before it worked

[sullen escarp]

for noise schedule, i mean that there may be a setting such as "normal", "karras", "exponential", etc

[uncut otter]

Oh I used SDE

[sullen escarp]

i think those kinds of visual distortions you're seeing is where the system is constantly overshooting a solution when it reaches the end of the step count. if you watch the generation, you might see it kind of bouncing between what looks like two differently distorted faces

i'd see if the problem still occurs with a mindlessly simple setting like an euler sampler and a normal noise schedule

and 50 steps might be a bit much. you could probably get away with half that

[uncut otter]

Thank you I'll give some of those things a go and see how it turns out

[lofty yew]

Anyone else hungry for a new sdxl official update like a 1.1 or whatever it will be called.

Fine tunes are getting better but I feel like they are all pretty similar

[crystal pollen]

@sweet thistle

this is with the default untrained NCNN upscaler for photo's by this program (4x upscale)

[sweet thistle]

pretty good
how hard does it hallucinate details, or does that depend mostly on settings?

[crystal pollen]

it just upscales..no hallucinations as in.. it doesn't try to detect things that isn't there

[sweet thistle]

dang awesome thank you

[crystal pollen]

https://pika.art/video/fa5d010e-42bc-4d09-9d46-b37448b18bab the girl with the magic pancakeplate 😄

[fiery marten]

2 ┬─┬ノ( º _ ºノ)

[limpid lichen]

#🌶｜off-topic

[blissful bramble]

Thx

[torn swift]

Two Chinese people are celebrate thier 50's birthday.

[sterile kiln]

Jeez, looks so realistic, especially background

HAHA, its a real picture, i'm so dumb...

[shut sinew]

You never know here but yeah, pretty sure its a real one

[sterile kiln]

With this name...

[blissful bramble]

And background

[naive thistle]

jailbreaked dalle 3

[crystal pollen]

lol trying out the comfyUI handrefiner thing using the whole meshgraphormer model 😄 results in baby hands sometimes

and seems to only detect 1 hand at a time so somehow needs a extra pass

[pallid ruin]

motorbike made of pumkins

[lofty ivy]

ive been wondering, and trying with various levels of success

but could i give Stable Diffusion A1111 a sketch and ask it to finish the drawing for me? like what if i gave it something like this and asked it to ink and colour it in comic fashion, with a sick model or lora base:

ive given it drawings that ive inked and coloured and gotten great results btw

[shut sinew]

All you need for that is ControlNet

And some playing around and tweaking

[lofty ivy]

ControlNet? waht that, can you tell me more?

[shut sinew]

https://github.com/Mikubill/sd-webui-controlnet

[lofty ivy]

thnx, ill check it out

just found a sick civitai file but its not a checkpoint or lora, its EMBEDDING? where do i install and use it?
https://civitai.com/models/214976/comic-book-pencils

[pallid ruin]

You have to download it and put it in the embeddings folder of automatic1111, then when you open sd you can select it on the textual inversion button

[lofty ivy]

thank you duuude. youtube and chatgpt werent helping at all

[pallid ruin]

np 🤗

[frank timber]

How do i generate with this style

[ancient agate]

@worthy shore

[turbid dune]

Hey guys, do you have any idea on how to make images in the same style than that ? What model to use, etc... The "kinda realistic but you can tell it's ai" style

For exemple, I tried recreating the pirate one but I only have results similar at this render : 😐

[pallid ruin]

Im sure a lot of models are able to recreate that "style" but Im not sure what was the prompt of that images :S

[turbid dune]

Well, it was the same prompt for both the pirate images

[pallid ruin]

I´ll look dreamsharper page too see if some images match that style

Aaa, do you have the prompt? so i´ll try it on that models

[turbid dune]

Yes, of

[turbid dune]

(highly detailed:1.2),(best quality:1.2),8k,sharp focus,(subsurface scattering:1.1) (depicting a character reminiscent of a youthful Will Turner from Pirates of the Caribbean:1.2),(detailed face and eyes:1.2),capturing the character in a sunlit tropical port with a ship turning its back in the distance, (mixed emotions:1.2), (warm sunlight:1.2), (featuring a captain attire akin to Will Turner's:1.2), (including a wider shot in an American plan style:1.2), (cinematic lighting:1.2), highly detailed, (Dungeons and Dragons:1.3),(Elden Ring style:1.3), (enhanced hyperrealism:1.2), RPG, artstation-worthy, smooth, beautiful, and sharp focus, incorporating two pistols and a rapier, with the character exuding a more youthful appearance and a friendly demeanor, drawing inspiration from renowned artists like artgerm and Greg Rutkowski.
Here you go

The first "good" image was made using this discord server's bot and the second was made using my local stabble diffusion and the revAnimated model

May be if I try with the SDXL model I'll get simillar results

[pallid ruin]

this is dreamsharper 8

i gotta try a newer model, this was updated like 5 months ago

[turbid dune]

This already looks a lot cleaner

[sterile kiln]

🙂

[turbid dune]

Which model did you use ?

[white pivot]

There can be only one RA !

[velvet nacelle]

hello, does anyone know why my stable difusion is white? i need a version with the blue one

i need this function

[sterile kiln]

Just enable dark mode in settings.
And, this "function" is Depth from ControlNet extension

[wild sorrel]

🫣 no white theme warning 🫢

[turbid dune]

Guys, any ideas why my renders looks like this ? What settings could I change/tweak ?
I'm using DreamShaperXL without Lora, DPM++ SDE Karras (I got same result with Euleur A)

While using the SDXL got fine results like this with the same prompt 🙄

I'm gonna loose my mind

[dry crow]

SDXL models got trained on a 1024x1024 resolution, going for 512x512 will make them look like that

[turbid dune]

Can't believe it was just that...

Thx

The render looks way better now

Also, would it be possible to take the face from the first pic and kind of "copy paste it" on the second picture, or something similaire to the second picture ? What would be the working process to do that ?
Or at least get the pose and camera angle from the second pics with face from pic one

[wild sorrel]

did 3 images, but other ones kinda meh

[sage ibex]

What happens when I let my five year old daughter control the prompts

[pallid ruin]

https://tenor.com/view/scout-gif-25786536

Sorry I had to do it XDDD

[velvet nacelle]

yeah i thought i had a different version. turns it was just browser night mode plus some plugins

[pallid ruin]

Could you download controlnet extension and the depth model?

This is the link to the extension https://github.com/Mikubill/sd-webui-controlnet

[velvet nacelle]

yeah already did

[pallid ruin]

Here are the small version of the controlnet models, I think they are similar to the normal ones in quality but each one only weighs about 700mb https://huggingface.co/webui/ControlNet-modules-safetensors/tree/main

[velvet nacelle]

what is controlnet for? there are like 8 files of 1.4 gigs

[pallid ruin]

There are different models, in that case he used the depth map model, so you can import a depthmap image and "fill" it with the image that you prompt

Btw you can use the link that I posted, I forgot to say that the "t2iadapter" are the smallers model, you can use that

Its 155mb each one

I´ll try generating something with it and show it in a video

[pure rampart]

mig23 ai art. needs improvement. idk why it has those symbols there

[pallid ruin]

The one from the video is this one, the depth model, in the controlnet github there are images that compare the different models

here´s some example

[deft walrus]

you mean a depth map? That's in ControlNet, select enable, pick an image to use to set your map, click depth, then select the preprocessor with midas or one of the others, then the model will say 'depth' in it. Click the 'run preprocessor' exploding thing between the preprocessor and the model. Watch a video for details but that should get you started.

[pallid ruin]

@frosty inlet it should look like this

If it doesnt work remove it, it may be another problem

[analog dust]

Zara Haida

[analog nymph]

what's the GPU requirement of this model

[sterile kiln]

Yeah, scribble is really nice :
prompt :
A woman

[sterile kiln]

ZZtop bro, pilote just have good music tastes

[flat heart]

can anyone tell me if this image is ai generated? there is a big discussion about it and i’m saying that it’s ai generated while others are countering me

[rain gazelle]

certainly suspicious... found the thread on twitter about it and the artist "left the psd proof at work" or something

[flat heart]

yes they said that they can’t pull up the layers until they get the file from work.. which leads me to believe that they’ll try to trace or fake a WIP.

these bits are what made me question the authenticity of their art

[rain gazelle]

and people that say "I will never draw again" cause other suspicion.

too extreme... an artist is not going to just stop

someone faking it could

[flat heart]

yeah they’re trying really hard to defend themselves 😭 all she has to do is admit she’s wrong and correct it.. because otherwise she is a great artist.. i don’t understand

[rain gazelle]

I'm really not sure, not an expert on identifying 😐

trying to find older art... only see one so far and it is very different

this person is definitely a k-pop fan, that's for sure

I found only 1 other image of their other art, really... there's a lot of k-pop retweets to scroll through to get to any content. 😛

this one

[flat heart]

interesting.. i understand style changes a lot but hmm

[dry crow]

@wispy nest the Face i used for IP-Adapter

[rain gazelle]

Let me know how it all plays out 🙂

[dry crow]

with End Step set to 1 (no change in expression)
and with End Step 0.7

[flat heart]

i gotcha!

[wispy nest]

thanks will try it

[dry crow]

Np, you need the IP-Adapter Plus Fave full Controlnet model

The other ones are good too. But thats Made for faces only

[gray inlet]

Thank you, looks great :) I played with Faceswaplab a bit and it suited my needs very well

But will have a look into the IP-Adapter stuff for comparison

[frosty inlet]

oof, feeling a bit lost. I was following this tutorial using counterfeit for checkpoint and vae, to create anime girls, added depth and open-pose and the quality is crap. anybody know why? Dall-e does a much better job surprsingly

[tall goblet]

improve cloth quality and bike make fancy

[wispy nest]

@wispy nest

[rocky hawk]

Didnt know if alrdy mention, look closer at the eyes, even if not full ai it uses ai and then maybe PS, eye is too blur and even more notice compare to any of their other images
Artist can change style but they won't make it worst

[quasi parcel]

Why i dont have preprocess imagess tab?

Can someone tell me?

[topaz prairie]

@wild sorrel

[wild sorrel]

oh, I see, images wasn't even loading for me without vpn lol

[topaz prairie]

bizarre

[wild sorrel]

idk , maybe older images

[topaz prairie]

so probably they used a different scale before

[wild sorrel]

Are you going to generate using dreamstudio or other webuis?

[topaz prairie]

ultimately probably comfy but I want to work my way through them so I understand everything better

[wild sorrel]

not sure about comfy, but in automatic1111 - usually going above 1.7 isn't good idea

Will overcook

I usually keep everything below 1.4

[topaz prairie]

I read through #1072220168534642768, #1072229020520947753, and #1080946152318443610, but it seems that Stability.ai never created documentation for DreamStudio, and there's no glossary even for the terms I see like cfg and checkpoint. So I'm looking at YouTube for explainers on those, automatic1111, comfyUI, theory, etc. to try to understand this.

But I'd like any recommendations people have

[sterile kiln]

@dry crow and you all guys, do you know a good SDXL model for Skribble/Sketch?

[topaz prairie]

I don't really get how people understood any of this when it first came out

[wild sorrel]

well...read few things, started testing things out, then some new guides on youtube, few there...
https://stable-diffusion-art.com/

it's a mess anyway, alot of stuff outdated, some tutorials are straight up bad, but eh, you'll figure it out, focus on something simple first

[topaz prairie]

yeah a lot of these are a year old so I'm sure very outdated, but it's crazy there's so little out there to actually explain the theory, terminology, etc

[pallid ruin]

I was using realisticvisionv6 that´s based on sd1.5 so im not sure, Im running it on a 3gb gtx 1060 but it can even run on cpu (but it will be very slow)

[hearty lynx]

I'm trying to use the imag2img mode to modify an image but it's not working. The image with the transparent background is the input, the one with the white background is the output, and the prompt is "A golden retriever sitting on the moon". How do I make it follow the prompt?

I tried both 1.5 and XL 1.0 and my hardware is a 4070 GPU and 14900KF CPU with 64GB of ram

[pallid ruin]

I added something to the image and it kinda worked, sometimes if the background is completely blank it doesn´t add anything

[hearty lynx]

thanks, using an actual background fixed it

[half glacier]

Would anybody happen to know why none of my controlnet models show up in the adetailer controlnet models dropdown list? I have controlnet itself working fine and can use the models in the controlnet tab. But in adetailer, there are no controlnet models listed in the dropdown.

[sage ibex]

What's everyone's favorite LoRA?

[white pivot]

I try some lora with planes. I would say I would remove all bombs, gas tank, weapon under the wing. It will give you better result. Once you have a good model, you can try to add more systems under the wing.

[lucid bane]

A golden retriever walking across the woods in Costa Rica.

[regal radish]

I am trying to install Reactor but as you can see i am getting errors stating that the I am having trouble installing Reactor for Stable Diffusion.
I followed all the installation steps and as you can see in the CMD window it gives mainly the error that the insightface==0.7.3 is not found.
But the strange thing is that i do have it installed as you can see in the screenshot...yet still it says that it can not be found...????...
As you can also see that i have Visual Studio installed and also the community build tools
And you can also see in the SD extensions tab that Reactor is installed.

What could be the problem?...

Thanks in advance for the help.

[quasi parcel]

Can someone tell me how can i get the sub tab "Preprocess Images"?

[dry crow]

Have you selected the build tools for c++ in the community build tools?

Also post technical questions in #🤝｜tech-support
Then more people will see it that can help

[junior yoke]

Create 3D illustration featuring a realistic young Albanian boy with a blue backpack containing the blue and yellow flag of Kosovo walks down a narrow cobblestone street lined with Spanish colonial buildings. Many locals and tourists fill the street. The boy smiles excitedly as he holds up a camera to take a photo. He has black hair and wears green shorts and a gray t-shirt. Behind him, a woman in a red flamenco dress crosses the street. The surrounding buildings have clay roof tiles, iron balconies, and arched windows in classic Spanish style. People talk outside cafes and restaurants under green awnings. The scene is full of life on a sunny day.

[dry crow]

thats for the bot channels #1080261341362786384

[mystic marlin]

Supergirl photo of the day. I think it came out alright for what it's worth.

[half glacier]

I was having the same issue earlier tonight. This ended up fixing it for me. https://www.reddit.com/r/comfyui/comments/18ou0ly/installing_insightface/

After that the reactor node successfully imported without error.

[half glacier]

Though its actually pretty cool despite how ugly the workflow layout is. Here you can see what is going on. I set up nodes so that you can load any video, and it will automatically generate keyframes with animation skeletons matching the movement of human shaped figures in the video I am using. So in this case, I loaded a tiktok dancer video, and am stealing the animation data to replicate the dance she is performing.

The detection is working much better than I thought it would, and its even picking up frames for more complex fast moving motions that I thought it would have a problem with.

And its pretty quick too since Im having it use latent consistency models

[grave reef]

Why are you talking like you set up that workflow? I know exactly which .json file that is lmao

[tawdry magnet]

any img2img specialist?? #💬｜general-chat #🏞｜general-with-images #🤝｜tech-support

[half glacier]

You misunderstood me. The workflow is from a tutorial. I was just excited that I got the worldflow to work. Not sure what I was doing wrong.

Took me days lol.

I dont know why I have never had much luck with comfy. But now I am finally starting to like it. But maybe the alcohol made it work. That was the only thing missing before tonight.

One thing I am still confused about though is how the LORA stacker works. Does the order in which you load the LORA's in the stacker make a difference? Like do the ones loaded earlier have more priority?

[brave gale]

Nobody? 😔

[wispy nest]

What u need

@prisma iron

[light bear]

does anyone know fooocus

[wild ice]

Hello!
Do you know any models or prompts to create images with this art style?

[peak spear]

Anyone know of any models/LORAs that could approach this sort of style?

[grave reef]

You can use ipadapter with that image to copy it's likeness.

[peak spear]

That doesn't really seem to work XD

[grave reef]

Hmm give the image to chatgpt to describe it then use those as prompts plus look for lora with those keywords chatgpt gives

[deft walrus]

Yay Firefly

[pallid ruin]

Sadly the bot doesn´t have img2img option (I think)

[nocturne creek]

Generate a cartoon avatar in an anime style based on an uploaded image

Generate a cartoon avatar in an anime style

Generate a cartoon avatar in an anime style

[pallid ruin]

@nocturne creek there´s no option to edit/use an uploaded picture in this server

you can use the bots like #1100170312106127410 for example but to generate images from 0 with a prompt

[nocturne creek]

OK

@pallid ruin Does this bot have the ability to generate images?

[pallid ruin]

Yes, you can use the command "/dream" on the bot channels and enter the prompt after that

[nocturne creek]

"After uploading the image, do I complete the image generation function using the '/dream' command?"

[pallid ruin]

There´s no option to upload images, to do that you have to run a local installation of sd, you can use the command like this for example

Then the bot will generate the image that you prompt

[steel stag]

just made this. Thought it was cool. Not perfect but came out pretty closely to how I envisioned it

[turbid root]

A painting of a cute goldendoodle in the sky, wearing a suit, natural light, with bright colors, by Studio Ghibli

[austere grotto]

plus

[bronze marsh]

Tommy Egan en dessin animé

[limber raptor]

Hey Guys, I am looking for some online tutorials on how to take line artwork that I created and apply any style to it and then output some sort of vector or layered file. ANy help is appreciated.

[pallid ruin]

You can generate images with the bots channels like #1100170312106127410 , you have to use the command "/dream"

[subtle meadow]

I working on the node of hand refiner, and after one week settings most of time the result is really good in photoreal images:

[pallid ruin]

Nice job!

[faint sorrel]

Who is interested in stablestudio?

[dry crow]

Hey for that you should use the controlnet lineart model

Or the lineart anime

[median vortex]

Hey Guys, does anyone know what the issue is? in the first pic, that is what it looks like when generating, the second pic is when its finnished, why do my faces become so ugly? 😦

[deft bison]

Satan speeds towards heaven with a group of devils. They collide with burning meteorites

[rigid mauve]

Refer to this picture. The picture is clean and the overall style is a meditative state.

[late goblet]

/bee

Satan speeds towards heaven with a group of devils. They collide with burning meteorites

Satan speeds towards heaven with a group of devils. They collide with burning meteorites

[steel stag]

made a little motorcycle-themed profile picture for my brother :) he liked it

this is how it came out of SD

[inland hull]

i am only new into SD (auto1111) .. but get this into session:

and more (^_^) -> http://maitag.de/semmi/stable-diffusion/ (my first tryes)

totally amazing (and stableRunningSURE!!!)*hugs!

I am tryed also to make some walkcycle into variations by (but failed there into at now ... also controlnet+lora) ... -> the .blend and png here: https://github.com/maitag/peote-samples/tree/master/anim/walkcycle/assets)

https://github.com/maitag/peote-samples/blob/master/anim/walkcycle/assets/walk.png?raw=true

https://www.youtube.com/watch?v=QIe02hyR-ak

[inland hull]

have only this at now: http://maitag.de/semmi/haxelime/peote-view-remaster/anim-walkcycle/ 💧

(wanna make it walk into SD-style ;:) [my "goal" for this year ;:)]

someone tryed to generate UV-maps here ?

[pallid ruin]

Try adding some realism prompts in the negative prompt

Also this vae is for anime (in case you dont have it) https://civitai.com/models/23906/kl-f8-anime2-vae

it goes on your models/vae folder if you are using automatic1111

[sick pagoda]

Hello everyone

[sudden cipher]

Hello everyone,
I have a hard time finding a workflow who would help me transforming mediocre 3D characters into photo realistic humans (see picture attached).
I would appreciate any help achieving those results (A1111 or ComfyUI).
Thanks !

[pallid ruin]

In a1111 it can be done with img2img and a cfg of about 0.5-0.65

You can also use controlnet

(example stolen from reddit)

[gaunt path]

Black background, dark Q version, IP image, blind box hand office, pink purple fashion skirt, medium long hair, headphones, publicity sexy, big eyes, open mouth smile, legs standing, arms folded, rich in details, blind box, bubble mat, 3D, HD, C4D, fashion, with impact

[quiet current]

you're using codeformer or some sort of "face restore" option

[covert pagoda]

automation build, wip

8kish photo

[shut sinew]

Isnt that a render?

[covert pagoda]

I guess if you wanna define render

I just used the automation photoscene

[covert pagoda]

im changing up the headlights a little. Just a little more modern touch

[covert pagoda]

what a perfectly confusing thing to happen

Ai says AI is bad

[heady anchor]

https://tenor.com/view/fact-of-the-day-facts-fax-fax-bro-fax-bruh-gif-26309542

[quartz ravine]

No, he says it's good

so good, that artists who make stuff that AI can do should stop complaining and quit their jobs if they can't adapt

Ai Weiwei said: “I’m sure if Picasso or Matisse were still alive they will quit their job.

he criticised art teaching that focuses on creating “realistic” images. “It takes AI a second to do it. So that only means what they have learned very often is meaningless.”

[covert pagoda]

Oh wow I completely missed a word in that headline lol

Or all of them…

[wispy nest]

tested both bing image creator and sdxl discord bot, say less microsoft

[fast swan]

Hey guys I have a question please help me how can I generate a photo using AI but using an existed photo that I have

I have photo for a girl she passed away long time ago I want to make her mother happy to make this girl look like as if she's sitting at a table talking to angels in heaven do you have a way to make this visualization or imagination happen please

[median vortex]

INterestng, would you know where i can find that?

[quiet current]

it would be somewhere in the settings tab. Something like Settings->User Interface->Face restoration model
I don t have access to an auto1111 instance at the moment

[median vortex]

I see, thank you!!

[median vortex]

This is what it looks like, cant find it 😦

wait i might have found it lol

[median vortex]

Thank youuuuu it did the trick!!!

[quiet current]

for the future, keep in mind that those are features for realistic human faces. So it will mess up anime stuff

[daring sable]

Looking for help. Did i get my settings right? Its not generating the right pose

[shut sinew]

please continue in #🌶｜off-topic this doesnt really belong here

[indigo pumice]

This looks pretty amazing. can you share with us, how did you actually made it happen ?

[dry crow]

your input image is already a pose, so you need to set the preprocessor to "None"

[royal tangle]

Someone just fixed a problem that was bottlenecking my pipeline for over a year - PROMPT 2 MOVEMENT 🌈

https://meyourhint-momask.hf.space/file/cached/12138/sample_repeat0_759737885.mp4

near real-time prompt 2 movement-openpose to create a de facto 3D model by running it through stable diffusion every frame. SD just killed blender & co for artistic stuff and small anime productions

testing pipeline:

agent _> storybook and movement prompts (sb)

sb _> memo_gen _>memo_gen gets sliced into frames (sf)
sf _> openpose + depth map _> stable diffusion
frames _> video

=
aka autonomous prompt and scene; aka fully ai generated visuals for comics, anime etc.

this is insane

this is 2D - no fucking blender - no fucking polygons - fuck nvidia

[cobalt oak]

Any tips on how to improve the photo realism, I use sdxl V1 the prompt is here: https://app.bench-ai.com/workbench?id=4c417077-b0c7-11ee-8ba3-6aef6c450620

I upscale using esrgan after

[alpine lantern]

It is the first time I use it, I downloaded a trained Lora but when I put it in collab it generates black images. What am I doing wrong? help pls

[sterile temple]

A lora is like an add-on you combine with the base model checkpoint. So you need to load a base model checkpoint, look at which one the lora creator used for a start. Then put the lora files in the lora folder and you add lora:SugarOP to the prompt to load it

[alpine lantern]

https://aieasypic.com/inspire/models/detail/sugar-one-piece-v10-62457 He used SD 1.5 and this lora, it doesn't say anything

[sterile temple]

looks like the use something called cetusmix as the base checkpoint

[alpine lantern]

where did you find that information?

[sterile temple]

click on an image and it shows the metadata on the right

[alpine lantern]

It doesn't come out, but it doesn't matter, I'm downloading it from https://civitai.com/models/6755?modelVersionId=20229

So, I put the cetuxmix in: path to model. right?

And where do I put the SugarOP.safetensor lora then?

I create a folder called lora on the drive and put it there without selecting it? Would just writing lora:SugarOP be enough?

[sterile temple]

it looks like you are using A1111, which should have a models folder and inside that folder there should be a lora folder

the cetus mix file goes into the stable-diffusion folder

[alpine lantern]

I have these sites.

[sterile temple]

I guess use model download for cetus mix and then the download lora for the sugarop

[alpine lantern]

but in download lora asks me for a link not a path:(

[sterile temple]

maybe click the /content/drive link and have a look in the folders?

[alpine lantern]

Well, I'll try it another time, thank you very much for the help!

[wispy nest]

turned ai art into wallpaper

[livid tapir]

@dry crow hey im in here now

[dry crow]

hey, here my input image from the internet

and here the output after using reactor

you can load any image in there and it will apply the face to the output image.
you can also add a batch of images in a folder and it will create one image per face

[livid tapir]

that looks like exactly what i need

[dry crow]

yes thats what i thought too ^^

@livid tapir if you want to install the Reactor extension, please follow this install guide carefully:
https://github.com/Gourieff/sd-webui-reactor#installation

[livid tapir]

can you take that same face and say create 10 images of that same face perfomring electrical related tasks or something liek that and it generate them

[dry crow]

yes

for that you only need to change the prompt and set batch count to 10

then you get 10 images

[livid tapir]

i need to develope a workflow around that process. that is perfect

[dry crow]

you placed the lora in the wrong folder

it needs to be in models/lora

then select in in the lora tab

[knotty pollen]

the heck is this?

[lethal burrow]

Is this happening in other bots?

[dawn seal]

Yes it is.

[lament steeple]

https://media.discordapp.net/attachments/1100170312106127410/1195772124552515706/Three-dimensional_space_very_black_background_the_corner_of_the_black_space_is_visible_the_silhouette_of_a_man_is_standing_seed-0ts-1705164688_idx-0.png?ex=65b53492&is=65a2bf92&hm=44db68c9952f3101b799505cfb9ca6f75182cbca86fb40156f33be44434ff3e1&=&format=webp&quality=lossless&width=582&height=582 https://media.discordapp.net/attachments/1100170312106127410/1195772124216967168/Three-dimensional_space_very_black_background_the_corner_of_the_black_space_is_visible_the_silhouette_of_a_man_is_standing_seed-0ts-1705164688_idx-0.png?ex=65b53492&is=65a2bf92&hm=00e52839ea869d64ee0e74e0879bc90c8f04be20bdfc7f2ee882a3f677adf163&=&format=webp&quality=lossless&width=582&height=582

[tawdry magnet]

img2img is not working, just showing 0% loading.

#🤝｜tech-support

#💬｜general-chat

[pale blaze]

#😊｜co-creators como puedo crear imágenes

[gloomy rapids]

hi, i want to change my character hands inpaint mode, as u can see but it does not change for 6 pictures, why?

[dry crow]

hey, are you sing an AMD gpu ?

[gloomy rapids]

yes 😦

[dry crow]

okay, which one and whats in your webui-user.bat?

[gloomy rapids]

rx 7900 gre and set COMMANDLINE_ARGS=--use-directml --opt-sub-quad-attention

[dry crow]

AMD cards need these Args:
--use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling
to make inpainting work you have to change --no-half-vae to --no-half
but using --no-half will result in a much higher vram usage so be aware of that.
Also for inpainting you have to set the Inpaint Area to "only masked", then also set the resolution to 512x512 (thats not the output resolution):
Then it will inpaint really fast

[gloomy rapids]

oh thank you so much i'll try, I searched for a long time on the internet and could not find a solution, thank you very much

[dry crow]

no problem, i have an AMD card too, so for any questions feel free to ask

[gloomy rapids]

❤️

i am getting error when i use this args: --use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling

i changed from --no-half-vae to --no-half, it fixed but this time, inpaint mode doesnt work

i dont know 😦

[dry crow]

what error did you get?

[dry crow]

in the cmd pls

[dry crow]

ohhh, you installed the webui the wrong way (by downloading it as zip), so it never got any updates. so your using an older version of torch that get the error, also the newer version is faster

[gloomy rapids]

i installed so much another version and i got some error, so yes

[dry crow]

best is to checkout my AMD Install guide in the Pinned Messages of the #🤝｜tech-support message channel
and install the webui into a new folder. you can then copy your models over

[gloomy rapids]

oh thanks, ill look

[gloomy rapids]

@dry crow hi again me sorry 😄 i installed this time it works with these args, but inpaint mode does not work, still same 6 pics,i have selected only masked mode

just a second, ill change it to --no-half

[dry crow]

yea with --no-half it should work

[gloomy rapids]

should i add --lowvram because of getting RuntimeError: Could not allocate tensor with 9994240 bytes. There is not enough GPU video memory available!

10gb vram is not enough

[dry crow]

that should not happen if you used 512x512 and inpaint only masked

maybe you forgot to set these settings again after restart

also every time you get an out of memory error you need to restart the webui to clear the vram

[gloomy rapids]

setting is like this

i can create 1 pic

but if i change to 6 pic it gives error

[dry crow]

restart and try with 5 or lower

but like i said --no-half uses more vram. so you should only add it if you want to inpaint or make 2 different webui-user.bat shortcuts

[gloomy rapids]

i changed like that set COMMANDLINE_ARGS=--use-directml --opt-sub-quad-attention --opt-split-attention-v1 --no-half --upcast-sampling, it works perfectly now, thank you so much, I know I'm tiring you out. thanks

[dry crow]

so you removed --medvram? thats trange cause that normaly helps for AMD gpus

but good that it works now

[gloomy rapids]

yes i removed otherwise i am getting error, i dont know why but it works

[glossy vapor]

can anyone simplify how extensions work?

I downloaded an extension for better hands and feet because I am absolutely lost on how to make them better and not mutated

https://github.com/jexom/sd-webui-depth-lib

I downloaded the zip file, after I unzip it do I just throw in all the folders? or every individual file?

maybe there is a better alternative

but I'm trying to get rid of this "shadow" effect I keep getting on hands and feet

[crimson flower]

So I tried to generate an image for "Software Engineering" in Fooocus.

[pallid ruin]

extremely detailed and sharp background, massive sparks and fire particles effects
8k, hi res, shallow depth of field, dynamic angle movie grainy still, 35mm, kodachrome, 1990s broadcast quality. (extremely detailed textures)
a knight from dark souls with a machete made of fire, standing in the night

[slender spruce]

monkey

[tribal meteor]

d

[gloomy rapids]

which model did u use?

[dry loom]

So I'm messing around with trying to get images generated quickly. This set is from the Dreamshaper 8 LCM, generated in about 25 seconds on a laptop RTX 3070.

With these parameters used in mind:

parameters: high quality photo, one car, sports car on asphalt, facing camera, fully visible, in daylight
Negative prompt: noise, letters, numbers, low resolution, compression artifacts, race livery, off-center, warped, deformed, driver, people, blurry, oversaturated, multiple cars, watermark, logo, asymmetry, license plate
Steps: 10, Sampler: LCM, CFG scale: 2, Seed: 3839933948, Size: 512x512, Model hash: a4f3e1526c, Model link: https://civitai.com/models/4384, Model: dreamshaper_8LCM, Version: v1.7.0
Is there anything you reckon should either improve speed and/or image quality without sacrificing the other?

[pallid ruin]

I used the sdxl model nightvisionXL

[merry coyote]

Try many cars and not multiple cars for negative

[quartz ravine]

this is great

[pallid ruin]

Lmao

[grave reef]

#🌶｜off-topic

[fresh canopy]

Hi, im trying to find models more suited to graphics illustration preferably SDXL based, found this one but most images seem to sugest its used as Lora? Also its about 70Mb, but no further info is provided: https://tensor.art/models/624412542015405514

Anyone knows of a Lora/Model that could output things like this?:

[dry crow]

Then its a lora. Models are always bigger than 1.98gb

Sdxl models are always bigger than 6gb

Hey, What's the model size and what's your GPU?

[delicate wharf]

https://github.com/TencentARC/PhotoMaker
Can this be used in comfyUI or auto1111?

[fathom quiver]

Any recommendations to take old facebook photos and remove the noise in them

[crisp harbor]

What do you mean by noise? Do you mean there's too much stuff in the background? Or the resolution is too low?

[fathom quiver]

you know what, its not really noise. Just a picture on facebook that is poor quality. Bad resolution older camera

[crisp harbor]

You could try upscale the images. Theres so many different upscaling methods. I used this guide, it's simple and uses already built in options in the Automatic1111 webui that this Discord has instructions for installing in the tech support pins:
https://www.howtogeek.com/894661/how-to-ai-upscale-any-image-with-stable-diffusion/

I don't think any single upscaler produces ideal results in every situation though. You could Google or maybe YouTube guides on how to upscale blurry photos if you don't get the desired results.

The guide does provide multiple options though (like using two upscaling models at the same time)

[covert pagoda]

Yeah ideal methods of just photos and just text are different

[fresh canopy]

I imagined so, but for some reason A1111 doesnt pick it up, np, searching for models more suited to graphic/vector art

[polar shore]

hey guys, i keep seeing these images where they add detail or completely change the style of an image, is this possible with sd?

[dry crow]

hey, yes thats possible

but not easy

i made this

[polar shore]

do I need a specific extension or how exactly do you do this?

[polar shore]

actually looks amazing haha

[dry crow]

yes thats done with controlnet and IP-Adapter

[fathom quiver]

This is an example: but a photo with bad quality like this:

Is there a way to make it crisp. Upscaling in extras just makes the same thing bigger really

I can take an IP adapter of an old facebook photo and even txt2img and it retains that bad quality of the original

[crisp harbor]

I found that using the upscaling in the extras tab does tend to remove the lossy compression artefacts (e.g. those rainbow coloured distortions) to make some things more crisp, mainly things that were initially monotone. But it doesn't create new details that were not in the photo to begin with.

For faces specifically, the guide I linked to does describe GPFGAN and CodeFormer which are also options in the same extras tab. They will add new details to faces to make them more realistic at larger resolutions (the guide says CodeFormer is better at skin details and GPFGAN is better at eyes, but you can tick both at the same time and adjust strength of each method).

[pallid ruin]

bacon trees 😁

[old marsh]

What i need to do, what is Terminal&? i use A1111

[wild sorrel]

It shoul give you gradio link to follow (above this message), just paste it in your browser.
You can ignore this message

[old marsh]

I heed to host it forever

How to gradio deploy?

[sterile kiln]

YOu want a local install?

[old marsh]

I have a1111 on my server

I need to share it

And keep running

I put --share into webuibat

But it says that link will expire in 72 hours

[sterile kiln]

share?
You mean just install on your desktop computer?

[fresh canopy]

Every free "forward" service (gradio,ngrok,etc..) will have some sort of somewhat short relay time , if you want permanent you need need to install something like DDNS service or static IP and mess with port forward to router or have a paid solution

I use AnyDesk because of straight IP TCP Tunelling and 2FA , but if I wanna share it use Gradio or Ngrok

Using Gradio (--share) means disabling managing extensions remotely

[sterile kiln]

If it's a host server, there's no need for DDNS, it's a static ip.
I think he's talking about the --listen launch argument, which enables external traffic.
@old marsh
You can also change the default port with --port xxxx
Remember that all ports below 1024 need root/admin rights, for this reason it is advised to use a port above 1024. Defaults to port 7860 if available.

[fresh canopy]

Maybe its a "home server" no idea what passes by server this days 😛 , Also , pretty sure "listen" is for local only , "share"generates a public (Gradio) link

[sterile kiln]

Use --listen to make the server listen to network connections. This will allow computers on the local network to access the UI, and if you configure port forwarding, also computers on the internet.

they are talking about OTHERS computer locally, by example, from 192.168.1.2 to 192.168.1.3

and, internet obviously

[fresh canopy]

Of course you can TCP tunnel with listen , heck you can do it without any if running in "server"

[sterile kiln]

I don't see the point, especially if you have no control over the link.

by opening tcp traffic, you access the same server, without bridging to the gradio site.

[fresh canopy]

Share with a friend or colleague? If your workplace computer is behind some firewall or you don't have control over port forward? No idea, but I guess the thought behind it is for short term share

[hushed crane]

Good evening.
Tell me how you can generate large size images. When I install it manually I get an error.

[quartz ravine]

mfw a server says AI art is banned, so they hyper analyze every image

[wild sorrel]

Install what manually and what kind of error?

[pale copper]

steak

[wild sorrel]

Not sure what interface that is, invoke?

[lyric sapphire]

all these i generated

[thorn nebula]

despite being here since the start, i havent been catching up with SD for half a year, so now im a redarded user that has fell off. this is some of the last crap ive done before i ended it all
Have any big and popular extensions come out? Is 1.5 still relevant? Is auto1111 still relevant? Are there any YTers other than MattWolfe and MattVidPro AI?

[old marsh]

I just want to know what is Graio deply and how to make link not to expire so fast

[ebon cradle]

Can someone please tell me what these art styles are called

Thanks :))

[fresh canopy]

Left looks Picasso but cant remember the phase name, top right definitely Cubism and so is bottom right, but probably one of the many variations like sintetic Cubism or whatever

Surrealism probably for the first one left

If using A1111 you could probably use Clip interrogator extension:

[ebon cradle]

Thanks alot mcurto ❤️❤️

Ill search more for it

[fresh canopy]

Try this one : https://huggingface.co/spaces/pharmapsychotic/CLIP-Interrogator

[ebon cradle]

Thanks alot ❤️

[fresh canopy]

Tha page looks busy, extension works really well on fast mode (be sure to read the about tab)

https://github.com/pharmapsychotic/clip-interrogator-ext

[wispy nest]

we findin aliens today

[finite hound]

i just downloaded the program and it makes me crappy pictures like this, anyone know whats going on?

[dry crow]

Hey, what's the model you tried to use?

Also the resolution needs to be at least 512x512 for good images

[finite hound]

when i put it in 512x512 it says i dony have enough space

[dry crow]

What's your GPU?

Also what install guide did you followed?

[finite hound]

Intel(R) UHD graphics family total avaible graphics memory 4107 MB dedicated video memory 128 mb shared system memory 3979 mb

[dry crow]

Only that? No nvidia or amd GPU listet in the task manager under performance tab?

[finite hound]

where is that information?

[dry crow]

Open the taskmanager and click on the performance tab

On the left there should be listet
CPU
RAM
Drives
Ethernet
GPU 0
GPU 1

[finite hound]

yes

[dry crow]

If there is only GPU 0 and thats the Intel and no GPU 1
Then you can only use the CPU for image generation

[finite hound]

there is gpu1

nvidia geforce mx230

[dry crow]

Ahh okay

[finite hound]

https://stable-diffusion-art.com/install-windows/ heres the installation guide

thanks for helping me btw!

[dry crow]

Np, can you check what's inside your webui-user.bat?

Right click to edit it

You need some settings in there for better performance. These are not Included in that guide from above

[finite hound]

k gotcha

im on the .txt file

it says:

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half

call webui.bat

[dry crow]

Okay, remove the --no-half
And add --lowvram --no-half-vae
Then save and relaunch

Then try again a 512x512

[finite hound]

lets see...

it worked!!

thank you so much : D

[dry crow]

Perfect np 🙂
For better speed stay with models (Checkpoints) that are 2gb
Like Dreamshaper v8
You can find many models on Civitai

[finite hound]

my goal installing stable diffussion was to use control net to color my lineart drawings using my colored drawings as reference

dont know if it will work tho

[cloud sorrel]

you better get some better graphics first

[dry crow]

Oh that could be hard, and if it works it would be slow.
Idk if controlnet even works with 2gb vram.
But when you want to try it let us know in #🤝｜tech-support

[finite hound]

how?

[cloud sorrel]

i meant, you need better graphics card for that

something like 30xx with at least 8 GB of VRAM

just do not get nvidia 16xx, since that works too slow with stabile diffusion
my son's 1660 SUPER is super slow with stabile diffusion, even it is decent for gaming

Radeon on windows is not fully optimized, so it better to get nvidia for now (CS1o know more about that)

[finite hound]

ok so i opened an sketch on controlnet and marked it as lineart, then on unite one i opened a diferent picture with colors in shuffle mode, thinking it was going to take picture 1 and give it the colors of picture 2

but it ended up giving me that abstract picture

i also tried to write "male face" on the text to image but it gave me a hiperealistic black and withe picture that also had nothing to do with what i wanted

[cloud sorrel]

you need to have a proper contronet model selected

did you donwload them?

i have a meeting in 5 minutes, so i need to go, but someone else will help you

1.5 controlnet models

sdxl controlnet models

ip adapter models

see you later guys

[dry crow]

@finite hound
You need the controlnet models from here:
https://civitai.com/models/38784?modelVersionId=44877

Select the model at the top bar, for example Lineart, and download it + it's config file below the download button.
Then move the two files into the models/controlnet folder

Then restart the webui

[finite hound]

thank you both"

[gusty ridge]

red cat, flying in blue skay

red cat

[cloud sorrel]

are those smaller than those on hugging face?

[dry crow]

Yes (700mb)

[lofty ivy]

hey diffusers. what prompts would you suggest to get something like this- specifically that lighting setup?

[azure plume]

@dry crow

[dry crow]

What webui is that?

[azure plume]

its fooocus

[dry crow]

Ahh okay

[azure plume]

the only program that isnt giving me trouble with this lol

i could probably do it with comfyui but its daunting

[shell plover]

hi guys

somebody know install the stable difusion on collab?

[dry crow]

Then I would suggest set the VAE file to None

As that can cause the most Problems when used together

[azure plume]

i did sadly :/ i will try to force it in the settings rather than per checkpoint

[azure plume]

ohhhh this could be it

that checkbox

ill find out in a few mins

[dry crow]

best would be to get this setting on the front page. to do that:
go into Settings -> User Interface -> Quicksettings, there add sd_vae and then hit Apply and reload. (onmly if not generating)

[azure plume]

so its not turning yellow anymore except for 1 frame, and the shirt takes that information and turns yellow lol

what in the world is going on

[lofty ivy]

you sure? Im getting more animerealism or atleast cyberealistic 2.5D from this. the ones you suggested are photorealistic models...

[dry crow]

I think that wasn't meant for you ^^

[azure plume]

yeah i dont think so hahah my images are decent besides that color swapping

[dry crow]

Cyberpunk, neon city, and an anime model

[dry crow]

That swap is interesting xD

[azure plume]

its blowing my mind, like am i just gonna have yellow stuff in all my images now or what LOL

[azure plume]

this is the reverse, with the 1.5 checkpoint going first and SDXL as the refiner. its doing the same thing

this one doesnt fix itself lol

i cant understand why this is happening

[sinful urchin]

I reinstalled the webui with git recently, so I could update with git pull in the future. When I use safetensors as checkpoints, it produces stuff like this for any prompt

wondering why and how to fix it

[lusty kestrel]

what the fuck

what do i do

im just trying to replace the background

[lilac heart]

trimmed some details in ps lol

[cobalt oak]

Some of my favorite images, that I made using SDXL on bench AI

Please let me know your thoughts

[azure plume]

i would try something like "rim lighting" cause thats what that is

[limpid lichen]

scenic long shot, a river through a valley,view through the trees, a bear in the distance next to the stream
@valid pivot

the idea is to make something else the main subject of the photo

scenic long shot of a gigantic ancient cedar tree, bear scratching his back in the distance

[finite hound]

@dry crow So I downloaded the lineart and shuffle models for controlnet and i tried to merge my lineart picture with my color picture but once again it says that i dont have enough cpu memory

[dry crow]

Can you post a screenshot of the error in #🤝｜tech-support ?

[valid pivot]

that works, thank you

[valid slate]

Hello, guys when installing SD on pc, i am getting this error, can anyone help?

[dry crow]

Hey, it seems like your provider or DNS is blocking the connection to github.
If your from Asia you may need to use a VPN for the installation.
For more help come to #🤝｜tech-support

[valid slate]

thanks. i'll write you there!

[dry crow]

Check also if you have a dedicated firewall or proxy that may blocks

[rich prism]

Can anyone help me? I've attempted a simple prompt with all of my models, and the result is consistently like this. Sometimes, it even becomes low-resolution images, although I specify an HD prompt. I'm using an RTX 2060 Super. Can anyone assist me in resolving this issue? thankyou for your help

[rain gazelle]

Just someone's basement retro gaming collection...

[dry crow]

the thinkdiffusionxl is an sdxl based model.
so the resolution has to be 1024x1024, for it to look better

[restive shuttle]

hello

[dawn seal]

I LOVE the images of Herne. They are fantastic!

[silk tendon]

Anyone has access to this discord server?

I'd like an invite, thanks

Links to it are expired

[cobalt oak]

Appreciate it

[dark robin]

I think I put this in the wrong chat.

Have a question. I have a very low resolution image from a photoshoot that I'd like to try upscaling with A1111.
https://i.imgur.com/QvD0rQk.jpg

There is an image from the same photoshoot that has a relatively larger resolution;
https://i.imgur.com/UKWNhlK.jpg

Is there a way I can use the higher resolution photo somehow to assist in the enlargement of the low res photo? Can ControlNet do something like this?
Thank you.

I can't bring myself to pay $40 for Magnific and this is just an example. I'd like to do with other photos as well.

[brave merlin]

The image is not low resolution, it's 3000x3000px...
It's very grainy

[stoic sage]

Anyone know where the preprocess images tab went in A1111

when i look it up i dont see anyone talking about it or any reason why it should be missing

[brave merlin]

Extra ?

[stoic sage]

that's all just upscaling

is there some setting i need to turn on to enable it?

[brave merlin]

I don't think so. There probably was an update that removed it because it does it automatically ? Can't guarantee, but that's how I've seen things working, these days

[stoic sage]

does it do it automatically? I haven't tried yet so

[brave merlin]

Try it

[stoic sage]

yeah one sec

[brave merlin]

There is a way, just img2img it, then faceswap with her face.
I'm creating her model right now, just to check how it will work, but I got this already with img2img

[brave merlin]

By the way, I'm creating Lora's now, because I never got how embeddings work. What's the difference ? How easier/harder is it to make ? Is it usable for creating a specific person ?

[stoic sage]

I've never done this before so idunno

[brave merlin]

Gotcha

[stoic sage]

i just slapped some pictures of my dog in there and wanted to see what would happen

[brave merlin]

Please tell me the results

[stoic sage]

dunno if this is normal or not

[brave merlin]

Auto1111 ? Yeah

[stoic sage]

been like 400 seconds and not a single image has been made

ill just leave it i guess, see you in 4 hours

[brave merlin]

Not sure this is what you wanted, but here we are

[proper night]

I keep seeing images like this posted online but can never find what model was used

I've found things that were close but not quite there

Was wondering if any of you know...
Cuz I keep seeing this style but the posts never have any details (pretty sure it's ai most of the one with hands are often fused)

[stoic sage]

@fresh canopy This is what I get after 10 seconds for 5 steps 💀

i just saw ur cfg comment

[fresh canopy]

Use 512px, CFG 1 and Euleur a or SDE Karas

[stoic sage]

should i be doing medvram arg

or any special args

[fresh canopy]

Too many steps and it starts "tripping"

[stoic sage]

real

[fresh canopy]

#1178805966335119472 message

[stoic sage]

pretty good

mine dont look good though

best i get

and it takes 10 seconds each

do people still use --xformers

[fresh canopy]

10 is ok , sampling method has a great influence

[azure plume]

does anyone know why when i put my clothing into brackets like this {jeans|shorts} for multiple options, the quality nosedives, look this is with brackets and without

im on automatic1111

same goes for the background or anything else i put into these brackets {}

[azure plume]

i fixed this. turns out you have to enable the combinatorial gen in the dynamic prompts plugin. i dont know what the syntax was doing previously to drop so drastically in quality lol

[azure plume]

i cant tell with just one image. but i want the output to be random based on variables, instead of it weighing them all

i think its fixed tho im not sure

[shell plover]

a do you want random background?

[azure plume]

lets say i do {at the beach|in an apartment} i want it to randomly choose between those two each time

not both

one would be at the beach, one would be in an apartment

but in the same prompt

[shell plover]

aa if you put [at the beach] ,decreases the force, doesn't it?

[azure plume]

yes, but i want them all to have the same weight, so it chooses randomly, not one over the other

[shell plover]

but then you put them all in []

but i dont know if that work in SD, in PixArt that works

[azure plume]

no i put them in {}

oh i see what you mean, i dont think that works

ahhh i found out what was wrong. dynamic prompts was disabled when the quality was bad. i need to just have it enabled but not check the combinatorial box, then its random and not in batches. i need to check the quality when its finished rendering now lol

[azure plume]

yep if you try to do random combinations in brackets without this enabled the quality nosedives

its fixed with dynamic prompts enabled

weird how it would work in the first place without it turned on though

[quartz ravine]

brackets are just letters like any other letter. Automatic1111 adds additional filtering to remove them and process them before sending to the image generation.

Also, A1111's bracket processing rules are a nightmare built ontop of spaghetti

I dare you to try to understand the difference between
[cat|dog:1.0]
and
[cat|dog:1]

Hint: you'll get completely different images

[willow quartz]

The scene is full of tranquility, warmth, and soothing atmosphere. It can be a study room or a living room, with a hostess or family members.

[azure plume]

thats so messed up lmao

[broken linden]

hi! what woudl be the best method to recreate that symbol?

[fresh canopy]

Use Google image search, find some vector/image but at better Res, upscale with anime model, could probably get close to 8k sharp Image

[broken linden]

unfortunatelly i cant find this symbol anywhere

[wild sorrel]

idk, I found few
https://cdn5.vectorstock.com/i/1000x1000/42/24/throwing-trash-icon-on-white-background-vector-27794224.jpg

[fresh canopy]

I recently did for fun a screenshot of around 1k from a vector imagebank (I bought the model after because it was work) cleaned in PS watermark and got to 16k probably as sharp as real vector

It was 5 EUR so bought it anyway

Didn't need to...

[broken linden]

yeah, ive found similars, but not the one on my picture

[fresh canopy]

@broken linden If you have that Screenshot with more resolution I'm sure would help in search, as it is: https://lens.google.com/search?ep=gisbubb&hl=en-PT&re=df&p=AbrfA8orTiSV3ySHBbnQT5x78zha31-3u9oWu5RhlEbCN9fU4v7bhLgav3BjSrEwe3qFuMtWFCnGo_BP4EuYz49imb22AKhVSBObSMHVG8t5KztZWotdSNN2qlFfoRc0lRiCiF9IjS5iZjEPUnkFVFf3I-t7Wk21itRRKfoa5F-DKgiZgVVQC87IeR6TE2b8ucbW4MeTGvwl4FPmbA%3D%3D#lns=W251bGwsbnVsbCxudWxsLG51bGwsbnVsbCxudWxsLG51bGwsIkVrY0tKR0k0TW1Zek1XWTVMVE5qTXpNdE5EVmpPQzA1TW1GakxUaGhNalUwT0dFek9EaGlOQklmTUROeU1WRTFZMHBYVXpSalRVbG5SM0JuU1V4dVVETkpiMmxJWmpCb1p3PT0iXQ==

[sterile kiln]

https://www.youtube.com/watch?v=t1RTgznup5c

[dense radish]

gatos bailando

[limpid lichen]

super scarmario

[tribal laurel]

Pitch of a fictionary dystopic movie about the Roman Empire, in the style of Warhammer 40k, for AI lovers (Morgan Freeman cloned voice)
https://youtu.be/N9itVxjBE7U?si=YB6awmwNRTeDddv4

[void scroll]

That’s awesome

[feral sluice]

this is the image with the pasted in hand from gimp i was trying my last few attemps to make look better

is it just too small? the image is 1024x1024 and i tried inpainting at 2048x2048 res

controlnet with lineart, controlnet with depth, canny, ip-adapter, and various combinations of those

[wild sorrel]

idk, I inpainted smaller parts, just need to make really high pixel padding thing, whatever it's called

just takes some attempts

[feral sluice]

Only masked padding, pixels?

[wild sorrel]

of reinpainting over and over...

lemme check

[feral sluice]

im doing this for more than 4 hours now so yeah i already have like 1000 pictures of that hand

[wild sorrel]

I usually inpaint bit by bit , like
get image slightly better - reinpaint it, repeat like 50 times lol

[feral sluice]

i also tried various hands from the depth library

i did that for like a few hundred attempts but the hand only got worse i got no progress whatsoever

i did that succesfully on other pictures too, but here it just doesnt want to work

[wild sorrel]

yea, this will need really high value, so it can "see" more things around, make it somewhat similar

[feral sluice]

well i have whole picture activated, so it sees the whole picture...

maybe it sees too much

it thinks its a horrible looking supergirl so she needs horrible hands

[wild sorrel]

😄

[limpid lichen]

higher denoise, less controlnet power, end controlnet earlier , and umm. maybe throw some artist names at it? thats a more complicated fist pose for sure. knuckles out is a lot of lines. plus it looks like its small too so that makes it tougher

[feral sluice]

which controlnet should i use even?

the hand is already perfect actually, only the seems arent and the texture isnt either, and maybe the color could be improved aswell

if i use lineart on 2kx2k, the thumb isnt seperated enough and it just garbles it together with the other fingers....

maybe i should draw a fine black line around the thumb in gimp?

[wild sorrel]

I had pretty weird way to adjust color, making it as I want, but that can make your image only worse...
Just copying part of the image from somewhere else and put it on hand , let it inpaint and figure out what color it should be (again, lots reinpaint iterations)

there should be better way, idk

[wild sorrel]

that's an option too

[limpid lichen]

i'm getting decent prompt understanding by using "front of fist" , "palm side of fist", "arms out palm fist" . Those get me pics with the right side of it more consistently

making sure the prompt has an idea of what you're inpainting helps blend pixels really well

1.5 understands fist positions so much less. i'd inpaint with sdxl instead. juggernaut has a very decent inpainting version

[feral sluice]

i just got disconnected...i will try that, thanks

[feral sluice]

you mean a specific inpainting checkpoint from juggernaut, or just the regular juggernaut checkpoint?

[limpid lichen]

specific inpainting of it

https://civitai.com/models/245423?modelVersionId=288402

[feral sluice]

i had problems with inpainting checkpoints aswell to the point i gave up on them -.- they wouldnt draw me anything...

but i will give it a try, thanks for the link

[limpid lichen]

sdxl on it's own has an inpainting layer, but something about these models extend it? i'm not so sure. the base sdxl inpainting model was put out by the diffusers team https://huggingface.co/diffusers/stable-diffusion-xl-1.0-inpainting-0.1

[sleek elm]

general-with-images

a dog is running on the water surface

[dry crow]

You only can generate images in the bot channels for example #1100170365604483202
With the / dream command

[sterile kiln]

Dude, just use Stable.Art.
you'll save an incredible amount of time, and it's so much more practical and precise... Etc

[feral sluice]

whats that?

[sterile kiln]

An inpaint/img2img/txt2img tool for Photoshop

Stable.Art load in PS, and connect to A1111

[feral sluice]

i dont have photoshop :/

[sterile kiln]

Depending on your "values" about having something you don't, I can help you x)

(i don't know if its a good translation)

check your pm @feral sluice

[feral sluice]

ty but I'd rather use some open source software 😅

[sterile kiln]

Oh yeah i understand, that's almost all I use too. But sometimes you just have to admit defeat, the work of certain large companies, and the federation around them, can't be overtaken like that.

[feral sluice]

what even is the upside of using photoshop? you mean for the whole process? or just for inpainting? i dont get it

[sterile kiln]

The inpainting on the A1111 web page is disgusting. imprecise. etc.
With photoshop, you can use the lasso: boom inpainting. You can even use blend masks to erase the outline of a layer of inpainting that's too intrusive, or only half satisfactory. And do it very, very fast

It's Firefly only much better, with Civitai models 😄

[feral sluice]

you can do the same thing with automatic1111, just need the right extensions

[sterile kiln]

Trust me, not like this :
https://www.youtube.com/watch?v=utjxgMQbshk

[sterile kiln]

with a simple stamp to use the smoke on the background to cover that ugly piece of wood right in the middle then: inpaint with a prompt like this:
a man's shoulder, with blue fog, smoke and shadows in the background, trending on CGSociety, Intricate, High Detail, dramatic, psychedelic painting
The piece of wood has completely disappeared. It took me 2 minutes

The image before all inpainting :

[pallid ruin]

So cool

[stuck jolt]

a 4 floor modern house on a sloping land with under garage and a pool on the second floor

[wraith knoll]

is there a better faceswapper then faceswap lab or reactor?

[dry crow]

Depends. For faceswap in other styles. IP-Adapter on controlnet is better

[wraith knoll]

I mean photorealistic faceswap

[shell plover]

Anything XL VS MidJourney

which is the better?

who wins?

[pallid ruin]

Midjourney has more details, but I like the lighting in the anything model

Tbh its hard to compete with midjourney unless you use a lot of loras

[shell plover]

i dont know if this is the better model XL for background

[pallid ruin]

Do you have the prompt? I´ll try it in the sdxl model that I have @shell plover

[limpid lichen]

entirely a subjective choice. technically, both seem very competent at what they do. Stylistically they've gone different ways.

Anything XL, never heard of that? Is it another situation where other people made a model and named it after the anything v3 model that was really popular ?

anything v4 was so bad but everyone started using it because they thought it was the next version. it was just some jealous kid who hated that his models weren't as popular as Av3

official civit anything page, you'll note how they skip over v4, its because of that jealous kid. https://civitai.com/models/9409?modelVersionId=90854

[limpid lichen]

thinking about how he'll eat it

[shell plover]

yes , "a photo of a very beautiful white castle with clouds all around, close up, beautiful windows made of shiny precious stones, red tiles and pointed roof, with beautiful angels flying around the castle"

[weak stag]

i love that

[shell plover]

which model do you use?

[pallid ruin]

This is with nightvisionXL photorealistic

I wish there were more loras for SDLX :p 🥺

[sterile kiln]

RealVisXL

[shell plover]

same prompt?

[pallid ruin]

Yes, 20 steps, 7cfg

[wraith knoll]

Does anyone here

know inpainting only changes a face

and not add what i want to implement

[sterile kiln]

If you scroll up, i was talking about inpainting, i posted a video, just take a look

[wraith knoll]

how do you connect it to photoshop

[shell plover]

the euler anscestral is better for XL models?

[wraith knoll]

?

Cause thats epic

[sterile kiln]

? huh

[wraith knoll]

How do i see it on my photoshop

I got it from the GIT

stable.art

[sterile kiln]

I installed it a long time ago, I couldn't tell you, look at the Github description

https://github.com/isekaidev/stable.art?tab=readme-ov-file#how-to-install

[sterile kiln]

And yeah, thats epic

[pallid ruin]

DPM++ 2M Karras is better in my opinion, it has more details, the same with DPM++ 2S a Karras

Its a bit slower to process tho

[shell plover]

i gonna try with both

[wispy nest]

anyone experimenting with face swapping ; )

[shell plover]

the first is euler anscestral the second is DPMKanrras

its the same

on XLmodel

[pallid ruin]

🤔 maybe it depends on the model

[swift spear]

B

[pallid ruin]

I tried with roop on sd but it lagged so I just sticked with facefusion and gpen 512 model, Its the best in my opinion

[wispy nest]

i use react-webui, facefusion looks good, will try it out

[dry crow]

yes

[subtle condor]

mmmmmmm

[wispy nest]

damn ipadapter is sooo good, but i need realistic style

[azure plume]

why is this happening? left is txt2img adetailer, right is img2img adetailer, same settings

i want the results on the right to be in my automated workflow in txt2img

i dont know what the deal is

[dry crow]

What are your settings in both?

[azure plume]

same exact settings in both

[dry crow]

I dont have any of these settings in txt2img or img2img

[azure plume]

im using automatic1111

[dry crow]

Same

[azure plume]

sorry i forgot to mention this is adetailer

[dry crow]

Ah well

Now it makes sense xD

[azure plume]

lol, do you know what could be going wrong?

[dry crow]

Nope didn't used Adetailer with these settings

[azure plume]

its weird, its literally perfect on the right. and when it zooms in to show it rendering its a lot clearer

[dry crow]

What are your other img2img settings?

Like denois ?

[azure plume]

theres a checkbox in adetailer to completely bypass img2img

so it just activates the adetailer

so i know its not my img2img settings

[dry crow]

Ah okay, and you use both 1st, 2nd settings

In adetailer

[azure plume]

yeah exactly. it should give the same result

i use 1st tab for face and 2nd for eyes. but in img2img i just do one tab for the eyes

cause the face was already done in txt2img

[dry crow]

So its not 100% the same settings

[azure plume]

not EXACTLY but ive also tried doing eyes in the 1st tab so they run first, and same result

im going to run it again with face off completely

[azure plume]

but if this works that would kind of defeat the purpose, cause then id have to run the face in img2img

no its the same result

[dry crow]

hmm strange

[azure plume]

could you grab the adetailer extension and test it for me? just make a random chick and try to improve her eyes just by turning it on

i want to see if it really is my settings or if its just like that

no worries if you cant i would appreciate it tho

[dry crow]

only enable it with default settings?

in txt2img?

[azure plume]

yeah even with default settings i cant get it to work so if you can thats a huge improvement

besides the face model

[azure plume]

for the adetailer model pick this one

thats all i changed, just so it would pick up the eyes

[dry crow]

ok

[azure plume]

you should have it by default

[azure plume]

oh fml i have hiresfix on also with these settings

im thinking its not picking up the hiresfix, cause when it zooms in img2img the quality is crazy compared to txt2img

[azure plume]

thank you btw ❤️

[dry crow]

okay now ive mde 2 images in txt2img with adetailer. first is with yolo8s and the second with the face mesh eyes only:

[marble nimbus]

Wow, I like the details. Can you share the info if possible? I'm a noobie

[azure plume]

were the eyes messed up a bit before generating?

thats kinda crazy though they look pretty good

i noticed i get the same results in img2img when i dont resize the image to the size it was upscaled at. if i could figure out how to turn that on in txt2img i think that would fix it

thank you for trying that for me

[dry crow]

you mean in the preview?

[azure plume]

yeah

like after the hiresfix but before the eye fix

[dry crow]

idk, havent looked at it

[languid pagoda]

Haven't used ai in a long time tried checking out stable 1.5 i still had vs bing's new ai thing and oof it did the prompt so much better lol

how's stable diffusion 2 these days? It was controversial when it was new

[marble nimbus]

Everyone tells me to "stay the hell away from it" so I don't think it's bright

But maybe I'm just prejudiced

[shut sinew]

If you mean 2.0 id rather use SDXL these days

[languid pagoda]

yeah 2.0

ah

so still controversial I guess

[shut sinew]

Everything is either made for 1.5 or SDXL now

[languid pagoda]

maybe sdxl can keep up better, midjourney and bing have gotten crazy good

[shut sinew]

I really like SDXL, but some 1.5 models also stand out

[languid pagoda]

would you mind running the prompt I used to compare with bing's ai on your SDXL? i'm curious how it handle it out of the box

[marble nimbus]

SDXL takes a lot longer for me to generate, so I stick with 1.5 these days

Because I'm not on my main device, it's only 6gb gpu in this

[languid pagoda]

I don't mind longer times, I just want quality

[marble nimbus]

do you want me to test your prompt on sdxl?

[languid pagoda]

yeah that'd be nice thank you i'm curious

A highly detailed drawing of an knight with a fur cloak with nautical suit elements and a lantern, drawn, inktober, sketch this was the one i used to test

[marble nimbus]

any loras or models?

[languid pagoda]

no just base

[marble nimbus]

sure

[languid pagoda]

🫡

[marble nimbus]

Oh wait, I don't have the base sdxl on mine, where can I import it fastest?

I'm using sdxl models time to time, but never used the base tbh

[languid pagoda]

ah tbh I have no clue, if you don't have it's fine, thanks for trying

[shut sinew]

Dont have the base model, but tried a few others. You probably have to adjust the prompt for SD and use loras and other models maybe

[languid pagoda]

ahh thank you for trying, that seems similar but crisper

yeah you'd probably have to to get similar results

tbh for now i'm just going to use.. bing's free tokens when i want to mess with ai prompts

lol

[shut sinew]

I prefer having a local install with full control

[shut sinew]

Added a Lora and negative embed. Kinda tends to make wolf heads or similar, dunno why

[languid pagoda]

me too, also img to img and other features. But I just don't have the energy to fiddle a ton, just for inspiration this'll work fine if I don't use it a lot anyway

[languid pagoda]

oh that's a different result but pretty neat lol

one day we will go back to the glory days of crotch lanterns

one day

[shut sinew]

MS Paint Lora

[languid pagoda]

lol

that one is a closet furry

i do like the idea of a stick with lantern

[shut sinew]

And fried egg lora

[languid pagoda]

a what now lora

lmao nice

[shut sinew]

https://civitai.com/models/255828?modelVersionId=288399

[languid pagoda]

has science gone too far

[shut sinew]

Not yet

[languid pagoda]

you're right, we need a whole breakfast assortment lora

[shut sinew]

Ral has some really nice Loras

And some really stupid like this and the beer one

[marble nimbus]

omg they're so cool!!

And they're both for 1.5 and xl

I'm in love

[marble nimbus]

With revAnimated model + extra details 1.5 lora

[languid pagoda]

ohh interesting

very smooth

[pallid arrow]

I now have an enourmous failed bicycle collection

[marble nimbus]

canon event

[pallid arrow]

the thing is I tried installing kohya ss for lora and failed on that either

instead, i'm making forest pictures now.. simple

needs some impaint, i guess

[sterile schooner]

||mountai top view , moutains surrounded by fog style : realastic||

[subtle condor]

The lora lora:drow_armor-000003:1 i made. Its my attempt at getting Minthara's armor from Baldurs Gate 3. Since she is the primary model in the lora the figures resemble her in some form or fashion. Yes, my prompt is crazy i know. I ran it through several models with the lora settins at 0.6, 1, and 1.5.

cowboy shot, 1woman drow, elf, pointy ears, (light purple skin), (vivid red pupils), detailed eyes, black lips,white hair,black circlet, priestess, elegant, ((drow armor)),breasts,buckles, leather,blood, sword, candles:2, white marble columns and braziers, (hyper realism, soft light, dramatic light, HDR), ink scenery, (perfect body), jewelry, (seductive, sexy, alluring), perfect lighting, perfect shading, volumetric lighting, subsurface scattering, (photorealistic:1.6), (fantasy art), (beautiful, stunning, gorgeous),dark backdroplora:GoodHands-beta2:1 lora:drow_armor-000003:1
Negative prompt: (bad_prompt:0.8), (artist name, signature, watermark:1.4), (ugly:1.2), (poor details:1.4), bad-hands-5, badhandv4, blurry,young,teen, child, loli, kids,FastNegativeV2, (worst quality, low quality), EasyNegative, (BadDream, UnrealisticDream),[([mutated hands | mutated fingers]:0.65):([mutated hands | mutated fingers]:1.3):0.40], [([extra limbs | missing limbs | floating limbs]:0.4):([extra limbs | missing limbs]:0.7):0.5]:0.8)]:bad anatomy:0.3]:[[([mutated body : ugly head:0.25], [([mutated hands | mutated fingers]:0.5):([mutated hands | mutated fingers]:1.3):0.40], [([extra limbs | missing limbs | floating limbs]:0.4):([extra limbs | missing limbs]:0.7):0.5]:0.8)]:0.8], horns
Steps: 80, Sampler: DPM++ 3M SDE Karras, CFG scale: 7, Seed: 874930416, Size: 512x768, Model hash: 3699d7fa76, Model: luminaverse_v10, VAE hash: c6a580b13a, VAE: vae-ft-mse-840000-ema-pruned.ckpt, Denoising strength: 0.5, Clip skip: 2, Hires upscale: 2, Hires steps: 40, Hires upscaler: 4x-UltraSharp, Lora hashes: "GoodHands-beta2: c05ed279295e, drow_armor-000003: 704320b78683", TI hashes: "bad-hands-5: aa7651be154c, badhandv4: 5e40d722fc3d, easynegative: c74b4e810b03, BadDream: 758aac443515, bad-hands-5: aa7651be154c, badhandv4: 5e40d722fc3d, easynegative: c74b4e810b03, BadDream: 758aac443515", Version: v1.7.0

I have a 1024 x 1024 lora of her armor i built yesterday but havent tested it yet.

[marble nimbus]

Thanks! I will definitely play with this

[subtle condor]

i wonder if i should post that lora to civitai....

[marble nimbus]

I mean... why not?

[subtle condor]

Im testing the 1024x1024 ones now. ill show some results and get yalls opinion first

ive only made a few lora's so they may not be great. 😦

[marble nimbus]

You made "a few" more than me, which is 0

so be proud

[cold harness]

Hello all, new to discord

Um

Wo, someone at least reacted

[subtle condor]

https://tenor.com/view/cat-gif-6997856904020830598

[pseudo rapids]

https://tenor.com/view/cat-cat-meme-funnt-cat-black-cat-sleepy-cat-gif-16409614852005252877

[pallid ruin]

this is with nightvisionXL

[languid pagoda]

why does it keep adding animal helmets lol

that's looking nice though

[pallid ruin]

Maybe the AI gets confused because of the "fur cloak" , i´ll try changing it to see if it remove the animal helmet 😁

Yaay got it

"A highly detailed drawing of an knight wearing a fur cloak nautical suit, holding a lantern, drawn, sketch"