🏞｜general-with-images

[winter siren]

Genie

/genie

[deft bison]

a man sitting in an armchair at the beach, in front of him a giant plasma screen which is planted on the sea. The screen shows human knowledge...

[tribal grove]

请生成一张“十二星座之射手座”的图片，风格：科幻风，注意规避版权问题。

[quasi cedar]

image to image

[shut sinew]

@sturdy wraith and read it out with the PNG info tab in A1111

[tawny lance]

Oh oh oh this is the chat I was looking for!

[weak stag]

good

[fleet goblet]

I am ready to adventure.

[deft bison]

the Sun and the Moon orbit above a completely flat Earth with no apparent limits for humans... HD

[deft bison]

Devils rise from the depths of the seas and invade the Flat Earth. Humans defend themselves...

[thorn prairie]

a gril with a cat

a girl with a cat on beach

[bleak wadi]

Hairstyles for men

[thorn prairie]

love

[marble spade]

Indian army with indian flag

[worthy nimbus]

A Chinese dragon 3D icon,Pink,Light yellow, Lavender color, A clean white background,Clay material,Equidistant, Smooth and shiny! Lovely, Girl style, Realistic use of light and color, A soft color gradient, Bee style, Nintendo fashion, Best detail,High resolution,3D rendering, HD--s 750--niji 5

[warm arrow]

Hello any one are freepik contributor here?

[blissful cairn]

😏

[earnest solar]

Darth Pesta, Dark Lady of Disease

[novel crow]

a set of 10 UI icons with consistent and cohesive styles, ensuring harmony in design elements. Maintain visual unity by drawing inspiration from well-established UI design principles found on platforms like Dribble and Behance. Pay attention to shape, color, and overall aesthetics for a professional and polished appearance.

[blissful bramble]

💀

[trail iris]

Hi

[lone stratus]

an open skull, many eyes grow out of the brain, decrepit, old man, withered, yellow robe, sitting on a stone throne

[ancient otter]

move

[bronze sky]

Not Stable Diffusion, but since Paint got updated I knew what had to be done

[viscid plover]

/

[buoyant trellis]

Industrial Design, Outdoor All-in-One Griddle, Rectangular Oven, Technical, Warm White, Smooth Curved and Rounded Shapes, Exquisite Details, Intuitive Operator Interface, Display Display, Lighting Decoration, Soft Lighting, Designed by Dietrams, High Detail, Luminous, Ward Background, Industrial Design, Studio Lighting, C4D, OC Rendering, Clean

[past glacier]

convert the kids into adults

https://media.discordapp.net/attachments/1004159122335354970/1200691610678198292/Labyrinth_6272_s.jpg?ex=65c71a33&is=65b4a533&hm=73344df026362bd65d9725daf27b1a183db6e0f76b1b55e9e8c991578dc45703&=&format=webp&width=900&height=613 convert the kids into adults

[twilit root]

Have you tried putting these kinds of prompts in the negative: kids, young, underage
Also try putting this in the positive prompt: Adult, grown man/woman, middle-aged man/woman, or specify the age such as: Man in his 30's, Woman in her 40's

[strange jungle]

Anyone good with loras....know how or suggest how magazine cover LORAs are trained to get the magazine cover style and even the text style?

Like this one: https://civitai.com/models/78559/mad-magazine-cover-1980s-style

[long beacon]

Hairstyles for men

[sterile schooner]

/arabic theme backgerounds

[thorny rose]

Same way any other lora is trained
They take a bunch of images of magazine covers, label what is in each image, and train with good parameters

[tired widget]

Turn the picture into an animation style

[blazing gate]

A new style rustic tomb door made of stone and amber.

[normal spindle]

what is this node called?

[scenic umbra]

hmmm a checkpoint loader with text encoders? haven't seen that one before but most of the "basic pipe" stuff I think it's on the Impact Pack

[deft bison]

the pharaoh in front of the pyramids, at his side a being with the head of a man with horns curved backwards and a hairy body and winged elephants guided by Satan the cursed in front of a people of frightened slaves... HD

[strange jungle]

When you say lable. I'd label the text, the images, all the elements of it?

[thorny rose]

Label the content of each image

[vestal delta]

style: Van Gogh; a little girl hugging a white fluffy kitten under a starry night looking to a beach

[wispy nest]

K

[wild sorrel]

what model is it?

[runic crystal]

Persian Legendary Beauty --v 5.0 --stylize 1000

[scenic umbra]

you can only generate at the bot channels but right now they're offline, check https://discord.com/channels/1002292111942635562/1047610792226340935 . Also you are copying a Midjourney prompt, those commands like version and stylize won't work with Stable Diffusion models

[stuck compass]

Hey guys, I need to ask another question....I am noticing that the images generated by SD in my local environment has this strange low-quality effect that is not found images generated by others. Here is a comparison:

They are generated using the exact same prompt, model, and setting

I noticed this low-quality effect on my other models too, so I am thinking there is an issue with my local setting.

Anybody have a clue on what is happening?

[scenic umbra]

exact settings down to seed? maybe embeddings/neg prompts?

[stuck compass]

Ok I just realized that maybe I am lacking a VAE

I am using the exact negative prompts, the seed wasn't needed in the higher quality image

[deft bison]

Satan curses him on his ark which is on the water, gives orders to his army of demons to sow trouble in the world, among human beings

[strange jungle]

So what would I label this image (and the other cover images) to get this as a LORA style, especially to get the text style.

[lime bough]

angry bird

[pallid ruin]

I used that prompt in sdxl 🤗 🐱

[pale steppe]

Him

[thorny rose]

Old school, angry person, dirty person, text, flying bullets, soldier

[fierce ferry]

My first trials with Stable Diffusion with different models. Feels kinda hard to control the image but I guess I get some progress ... Still having problems with prompts like lightning, what camera settings (I have no clue) and details ...

The anime short skirt was not intedendet 😉

[deft bison]

the Sun and the Moon orbit above a completely flat Earth with no apparent limits for humans... HD

[wild sorrel]

Well, you have found a way to contact them, that's a first step, why don't you tell THEM all that stuff?
We have nothing to do with it lol

nvm, he's gone lol

[lyric nymph]

Make this image with the Philadelphia eagles

Dreaming fo... /dream prompt:Vulcan, star Trek, a female Vulcan, tanned skin, elf-like vulcan ears, arched eyebrows, skinny, slender, femenine body, sensual pink sci-fi dress, in a vulcan white alien mountain, Vulcan robe, fisheye camera, fisheye effect, retro, selfie seed:2531543920

[hidden whale]

What model do you use by the way?

[stoic pivot]

AI made this, but I almost get a feeling there is a missed joke somewhere. I so wish it had been easier to do and save characters with AI and be able to just give them aothe expressions and so

[deft bison]

Baphomet is a dangerous demon, sitting on his throne. many humans worship him and prostrate themselves before him. Baphomet turned them away from God towards him... Photo HD, realistic

[grizzled jay]

Can someone rate this? I just used the basic premium chat-gpt package

[gaunt heron]

1/10 doesn't use Stable Diffusion

[grizzled jay]

lmao did I join the wrong server then? maybe it's dall-e as a backend

[gaunt heron]

ya, and sorry wasn't trying to be rude lol. ChatGPT uses Dall-E 3

[grizzled jay]

No worries lmao, I love rudeness
How out of date is dalle 3 then?

[gaunt heron]

It's not, it's one of the current leaders of generative imaging, super accurate

We all wish here that SD could do the same things in terms of prompting using language models

But also it's not available to use locally

[grizzled jay]

accurate in what sense? what if you wanted to make something original and creative?

[gaunt heron]

Accurate in its ability to understand what a prompt is asking for. I think SD excels in the creativity area because people can just create new models and extensions with everything open source, free and uncensored.

[compact lava]

How to use it ?

[craggy jay]

#🏞｜general-with-images

Help

[gaunt heron]

The bot is down so you'll have to install SD yourself or use an online version

Also this isn't where you use the bot anyway

[scenic umbra]

in red letters you have information about the errors, seems you are missing or have put the wrong information, also I wouldn't advise pasting images of your personal information publicly.

[mild tusk]

question do you guys know how to add text in image? i tried with control net but i fail

[fiery pond]

I erase all personal date besides my name. Also, i double checked and it was all correct

[fiery pond]

You didnt read the print?

[scenic umbra]

you have to read the part where it says "please check the following information and try again".

[crisp harbor]

I had a lot of hilarious fails with this prompt!

[crisp harbor]

Those look really good!

You can fix specific parts of your results using img2img inpaint mode. You highlight certain parts, like the face, then choose to inpaint 'only masked' and it will regenerate only that part of the picture again; if you're inpainting a small part of the image you can be very specific with the prompt like 'beautiful youthful female face, smirking with red lipstick and natural makeup'.

Regarding lighting, did you try just putting in the prompt? For example 'bright sunny weather' or 'dim gloomy forest'. You can also control how strongly the lighting effects occur by highlighting the prompt and pressing Ctrl+Up or putting brackets around the prompt, you'll end up with something like (bright sunny weather:1.3), which emphasises that part of the prompt.

With camera angles, I've used added things like like 'sideview', 'view from above/below'. But the effectiveness of that kind of prompting depends on if the model was trained with those camera angles.

[fiery pond]

@scenic umbra this only appeared cause i erased some information. When the infirmary was there it was all correct and tgat message did nor appears. Look the other message.

[scenic umbra]

I'd suggest contacting support or: [e-mail], you can also try opening a ticket maybe the staff can point you in the correct direction to solve this issue

[fiery pond]

Tx

[tender igloo]

[mint kestrel]

戴眼镜美女，蔑视的眼神，30 岁，中国女人

[sacred kayak]

the pharaoh in beijing street ... HD

[stoic pivot]

Prompts like "hyper realistic", "photo realism", "highly detailed", "super sharp" is very popular, images come out looking mechanical but with all the flaws AI do today, I started to find fascination in the opposite side, make it worse, make it bad, make it more human with the flaws of AI, and even it look... strange (to be nice) I like the result.

[pallid ruin]

Wow that images look interesting, the last one is so cute 🤗 , do you have the prompts?

[wild sorrel]

that's generally not good prompts.
"realistic", "realism" - tends to make 3d-like renders or close to game graphics, that's not what people usually mean by that.
If you want a photo - just prompt "photo of..."

Detailed, details - leads to well, literally mechanical details (not always, but happens)
Word super probably doesn't have alot of meaning for an ai, doubt it has visual representation for it.
Sharp okay, but again...can be interpreted in a different ways...

[delicate wharf]

how can I clear this filter in Civitai? I can't find the setting

[nocturne oak]

Have you looked at your settings in your account? There's a bunch of checkboxes on what to show/hide.

[stoic pivot]

Me go to CivitAI to look at a random image, int eh prompt "(best quality, high quality, photograph, hyperrealism, masterpiece, 8k:1.3)" and in the negative "(low quality, worst quality:1.4)
This is very common in AI, to make 1K images to 8K and strive for quality that is better than quality and more realistic than realism.

[stoic pivot]

I cheated and used a random part of a test doodle that was left after my friends had tested Krita and a drawing tablet. It was used as an IP Adapter. The written prompt was "silly looking furry anthro goat characters. the dark overlord Benny in his office, drinking coffee"

[pallid ruin]

I'll try something like that

[stoic pivot]

I added a extra image as a secondary IP adapter to get color and another character style

The result and the image used as IP adapter.

[delicate wharf]

what tool did you use?

[white pivot]

w8 I will post my worfkflow soon.

[white pivot]

https://openart.ai/workflows/elephant_present_36/vid2vid-txt2vid-for-animatediff-ipadapter-latent-upscaler-detailler-upscaler-mask-editor/LwT4AB8wEtKQilrjq2G8

my workflow, everything is explained

[leaden night]

how do i enable the logging of prompts and whatnot to log.csv?

[leaden night]

thats a hell of a workflow

[white pivot]

Made by the best

[leaden night]

so you prefer comfyui to the sd web ui?

[white pivot]

Definitely

Allow me to understand what I'm doing and skip what I don't need

[mild tusk]

do anyone know?

[spring mist]

there is a LoRA that might work for you maybe, Harrlogos XL. There are other ways i dont know, like i saw someone bake in a mask for text

[shrewd adder]

funny wizard challange 🙂

[hollow grove]

fiori gialli

[lime rapids]

Bowie's in spaaace 🚀

[tropic granite]

dream/

[junior chasm]

Im currently trying to use AI

[earnest solar]

.

[stray lintel]

hello

[grim pond]

https://tenor.com/view/waiting-spongebob-coffee-sad-gif-16525719

sombody make me a rolls royce with middle finger instead of spirit of ecstacy in photoshop

[lucid reef]

ComfyUI SD + magnific

[silent owl]

How to use the pic in the left to generate one like the pic on the right? Anybody help?

[stoic pivot]

I start to learn things like inpainting and so on, now is Stable Diffusion and AI no longer random, it start to be full control over the design, it just so that the AI can help me with getting the lines clear, after a stroke and several year of absent from painting, this is a fresh start. I still not found my old style from when I did paint and so, but it look I am going in the right direction. I only have trouble getting the right colors, there is tha AI and me on the same page.

[lyric storm]

(for conversation in general)

[fresh crater]

make AI version with "apply now"

[sharp rivet]

Computer repair, webpage, wrench tool, cartoon engineer, busy, fixing computer.

[pallid ruin]

You can use the bots channels like #1100170312106127410 for example, to generate the images using the "/dream "insert prompt" " command. I used that prompt on sdxl tho 😁

[slow egret]

#1047610792226340935

make AI version with "apply now"

[left cobalt]

Q version of Three Kingdoms characters Zhuge Liang

[novel ravine]

anyone want fries?

[west fog]

i am beatingm y head against the wall

i got close to what i wanted

saved the prompts and settings

Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 1167222159, Face restoration: CodeFormer, Size: 512x512, Model hash: cca17b08da, Model: darkSushiMixMix_225D, VAE hash: 735e4c3a44, VAE: vae-ft-mse-840000-ema-pruned.safetensors, Denoising strength: 0.5, Clip skip: 2, Hires upscale: 2, Hires upscaler: SwinIR_4x, Lora hashes: "Chinese_Wedding: 4c7d87a8c66d, ChineseDragon_v2: a544def7c131", Version: v1.7.0

and for some reason when i try to go back to it i get nothing but a garbled mess

[west fog]

i figured it out

for some reason it was duplicating the loras

[crisp harbor]

What does duplicating loras mean? How did you fix that?

[west fog]

@crisp harbor dont paste settings with the style saver

it seems to save the lora without needing it in the prompt

so when you put the lora in the prompt again it freaks out

this right here was the problem

[crisp harbor]

Oh I see, I guess applying the lora twice would be similar to just putting an insane prompt strength to it, since that also leads to strange results

[west fog]

@crisp harbor yep thats it. hopefully they will fix this

as it wasnt obvious

[woven helm]

#bots show me that i don't have permission to send message. How to generate image?

[chilly reef]

hello any good image creator here?

i want similar type of image i tried but mine is not good

and is blurry

[mild tusk]

try adding a negative prompts

blurry would be the one

negative prompts is what you dont want in your prompts

just dont forget to add comma

[wispy nest]

Does anyone have an idea why my hands get so bad? Any advice, ideas?

[feral sluice]

its kinda normal to get bad hands, especially if they arent in a very simple position

[wispy nest]

Can you give me an example of how to fix it? how to write it correctly?

@feral sluice because I don't fully understand simple position

[feral sluice]

with simple position i mean no overlapping fingers, weird viewing angles etc

you have hands in your positive prompt and then like 5 times hands in the negative prompt... maybe try to cut down on your negative prompts and see if it helps, you have alot of stuff there

[wispy nest]

okay i try

[subtle condor]

I liked these

[sterile mist]

whats the one thats like stablediffusion

[late solstice]

running into a weird issue with Style Aligned in comfyui,
when I generate 4 images(4 batch size), I basically get 2 groups. By that I mean that image 1+2 are the same, but 3+4 are a different style but the same as each other?
So when I run 8 images(8 prompts, 8 batch), I still get two groups. 1-4 the same 5-8 the same?
Really weird because everything is working and no errors appear in the command prompt.
Anyone see something similar?

[fierce meadow]

/how to create images

[pallid ruin]

you have to use "/dream "enter prompt here" " on one of the bots channels like #1103708504142925824 for example

[pearl loom]

my bros i need to understand how can i define this image to create more and more with different characters and environment .
how would u define this image to make the bot create a similar concept?

included the ultra raw bad quality of the colors\shadows\light\vintage
i like this image because it is "bad" in that way, i dont know how to explain how u guys got me

[toxic basin]

hello？

[subtle meadow]

Hi, I found 2 excel tables of Midjourney keywords and artists style prompts, and today I implemented to a Comfy node. As I see styled prompts working well in SD too. Images using Mj styles, the first the original without style, all other some style included, but same prompt and seed:

[dawn kite]

Hery there, i've been stugling for so long to generate a image like this one. ince its almost valentine i have a gift idea for my girlfriend and sadly i am horrible at drawing but used ai tools before. i want to create something like this(image) but then customised so it looks like me and my girlfriend but she's wearing a classic disney princess dress. I've been prompting and inpainting the entire time but it's just failing I would love for it to look like this art style since i am going to put it on a horizontal sheet of paper with a poem i wrote for her. I'm sorry that this might be easy and that i'm asking it but help would be appreciated! 😁

[gaunt heron]

sounds like a really difficult project dawg good luck

[misty prawn]

Hi, is there any model to generate flat illustration like this using stable diffusion?

[wide pelican]

请你帮我设计一个app图标，这个app叫“宠物友好地图”，能帮助养宠物的用户发现周围允许携带宠物的地点，另外还有社区、排名、遛狗足迹等功能

[serene sand]

Zhao Yun, a famous general from the Three Kingdoms period of China, in traditional armor, holding a long spear, on a horse, with a determined and brave expression.

[frigid ferry]

Has anyone got unclip working with automatic1111? I think ive got it setup right but the images it's creating look horrible and nothing like the blog posts im reading

[pallid ruin]

I had downloaded a model like that but I deleted and I cannot find it 😿, if I find it or something similar i´ll post it

[shut sinew]

I had decent success with this lora https://civitai.com/models/24833

[delicate wharf]

how can I make a batch img2img in auto1111?

I have like 100 images of a video and want to pass it through img2img automatically... (don't want to use animatediff or ani of that)

[shut sinew]

there is a batch tab

[pallid ruin]

There is a batch option, you have to put the input and the output folders

[shut sinew]

Some examples I made

[delicate wharf]

what's going on? This workflow worked in the past now it doesn't...

does anyone got a vid2vid workflow?

[delicate wharf]

man these workflows are full of nodes which don't load models or stuff and fail. please a simple vid2vid workflow that works....

[delicate wharf]

Can someone modify this workflow so I can add two loras? please 🙂

[drowsy tangle]

#🌠｜show-and-tell

[mystic ledge]

how do i make images like this?

[crisp harbor]

In Automatic1111 you can use the interrogate CLIP button in the img2img tab (just underneath the Generate button) to get a rough prompt to work with, this is what it gave for the prompt:
a group of dogs running in a line on a field of grass with their mouths open and their paws out, Elke Vogelsang, cheerful, a stock photo, kitsch movement

[quiet current]

Too vague of a question.
Prompt for multiple dogs, eventually use regional prompter for more control on each individual dog

[crisp harbor]

Using the SDXL model and 1400x700 resolution I got these:

I did change the 'group of dogs' to things like 'group of cute puppies' and the final ones I did 'four fluffy young dogs'. Also added 'photographic realistic' to the front of the prompt and added 'blurry cartoon anime' to the negative prompt. If you know the breed of the dog you want then you should specify that in your prompt.

[crisp harbor]

That sounds interesting, what is a regional prompter? I haven't used that before

[quiet current]

Basic idea, split your output in multiple zones / regions, each having their own prompt. eg : https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111?tab=readme-ov-file#regional-prompt-control (if you're at work, take care, examples are .... suggestive)

[crisp harbor]

Oh thanks, that's an extention to the webui? Sounds useful

[quiet current]

yes it is

[mystic ledge]

look closer

[quiet current]

oh yeah .... those ones .....

You use QRCodeMonster and controlnet

[crisp harbor]

If that was the goal then I agree the question was way way way way way too vague...

way way way way way way way way too vague*

[mystic ledge]

can you try and make one like that and show your settings? I've tried that and no results were as good as that

[quiet current]

Sorry I'm going to sleep, it's 1:30 am here, I'm done for tonight. Try asking in #📝｜prompting-help

[kindred pelican]

Wildhttps://civitai.com/models/293331

[thin dome]

https://tenor.com/baZRe.gif

[wispy nest]

Head up

[thick flume]

Retro Diffusion on Discord got hacked

Shortly after this the server was deleted

[cyan cedar]

Create a highly detailed and realistic image of a Zatar Chicken dish.

[wild sorrel]

probably another case of a server admin trusting 3rd party bots security more then they should've

[shut saffron]

hi

[solemn jackal]

hey? how to upload a reference sketch&

[tropic shell]

Hey there! The bot is not currently running. Please check the #1047610792226340935 for more information. Thanks!

[thorn chasm]

because I don't have permission for bots

[upper shard]

Hello all, I am a painter that uses AI as reference to paint from. I have really been enjoying it so far and it has been helping me paint when I don’t know what to paint as well as helping my painting skills.

I am hoping to get better at prompt writing and generating though as i have seen some AI images on Pinterest lately that have made my jaw drop.

I noticed if I up the resolution in stable diffusion the generator wants to add more detail even of the prompt tells it otherwise.

Anyway here are some paintings I have done recently with AI.

[abstract temple]

that looks awesome

[river jay]

Hi everyone - reaching out for a bit of help please. I'm looking to recreate these results ("result-1" and "result-2" generated using Photoleap mobile app's AI Scenes feature) using publicly available APIs.

The goal is to transform the original building but keep the features like windows, rooflines etc.

From my research, I think ControlNet M-LSD is the avenue to explore. I have done some experimentation using ModelsLab APIs but the results are nowhere near the quality of the attached examples.

Do you think results like the ones from Photoleap can be achieved using available APIs?

[wispy nest]

playing around with upscaling and discovered a comfyUI way to save as jpg for much less filesize

[shut sinew]

#🤝｜tech-support

[spring mist]

i skipped over reading replies, but ide guess thats QR code monster

[mystic ledge]

ye but i need a good prompt

[spring mist]

i havent used it yet, on the to do list

[pallid ruin]

The bots are still offline but here its 🤗 😋

[deft bison]

Sun and Moon gravitate above the Flat Earth, at undetectable limits...

[wispy nest]

Does stability gprc api has rate limit?I can only see rate limit on rest API in documentation.

[crystal harness]

So the main character is kinda fine. Like not blured but the eyes seem.... Kinda stretched by a glitch sort off?
The cute puff tho seems having blured-like eyes

[dry crow]

is that a 512x512 generated image?

[crystal harness]

I think it is

Bc i dont remember doing any other back then

[dry crow]

okay, because then the resolution is to small for the eyes to look good

[crystal harness]

Ah

[dry crow]

or better say, the subjects face is to small for the resolution

for example try close up portrait

and the eyes will look better

most peaople upscale images to get a better quality when the default resolution is smaller

[crystal harness]

Hm

[south vortex]

#🌠｜show-and-tell create a poster "Safety in a war zone" where the rules are expressed by a character, i.e. a comic book instruction.

[crystal harness]

Also close up portrait mode or something?
Sry im not good at all with this 😅
I goofied a bit with txt2img and img2img with a bit of guidance from a friend

[dry crow]

No as prompt for example

You can also try 512x768 as resolution

[paper folio]

Oh i have a issue too.

So my controlnet outputs are verry bad in quality

They eyes and faces are very distorted

[crystal harness]

Alr ill try upscaling and then prompt

Altho idk how far i can go with gpu of 1060

[paper folio]

So i just need to use an Lora to ddx these things

Fix.

And negative prompts

Right?

[crystal harness]

Oh btw do i use any prompts for upscaling? Or just img2img and throw a bigger size?

[paper folio]

There are upscaler models too if you want to try @crystal harness

They add details to specific type of images.

[dry crow]

Make sure you have --xformers --medvram --no-half-vae
In your webui-user.bat for best performance

Upscaling can be done with Hires fix In txt2img

[paper folio]

I saw on openmodeldb.

[dry crow]

Also use a good negative prompt

[paper folio]

Ty

[crystal harness]

Alright. Will try, thanks

[crystal night]

What model or LORA can I use to create 2D sprite keyframes like these?

[dry crow]

@crystal harness @paper folio
here is an example what hires fix can do, left is default, and right is upscaled

i used the negative prompt from the pinned message in prompting-help

[paper folio]

Wow thats stunning.
Where is high res fix?

[dry crow]

in txt2img

[paper folio]

Ohk

[crystal harness]

Oh damn

Wait, i should provide the picture for upscaling. How is it done in txt to img

[dry crow]

if you want to upscale images you already made, you can do that by recreating the image with hires fix enabled, (for that load the image into the PNG-Info tab, then click on send to txt2img and then set the hires fix settings and generate). or by upscaling it in img2img

[crystal harness]

Oh

[crystal harness]

just checked. i had the 3 flags except i had lowvram instead of med. so my 6 should be enough heh

also should the pytorch be on max split 512mb?

[warm vine]

Prompt : massive mechanical cat
Created with : https://fumesai.web.app/img

[crystal harness]

wow so cool

actually i wonder how you do the prompting

[warm vine]

I used the prompt : massive mechanical cat

The website has auto ai prompt enhancer with gpt4

[crystal harness]

oh?

ig i should update or sth?

i seem to have the july version

but i dont have openai :c

[warm vine]

It's just base sdxl model with gpt4 prompt enhancer

Mee too 😔

[crystal harness]

hmmm

wait what? but it just worked???

[warm vine]

sometimes it happens with gradio

[crystal harness]

huh?

also wheres hires fix

im blind?

this happens after loading to txt2img

yes, im blind af :c

[warm vine]

😂

[crystal harness]

do i leave this be?

considering previews

i think the ai has tricked me

or maybe thats what it meant with error, that it didnt like the pick

[dry crow]

remove the pytoch max split stuff comepletely

[crystal harness]

oh?

[dry crow]

yes medvram is enough

[crystal harness]

its like usage of my ram, right?

[dry crow]

you have 6gb vram right?

[crystal harness]

yh

[dry crow]

then all you need is --xformers --medvram --no-half-vae

nothing with cuda, pytroch sizes

[crystal harness]

oh so just delete entire line?

[dry crow]

yes

[crystal harness]

alr

[dry crow]

try Esrgan4x+ as upscaler and use 0.5 denois and set hires steps to 10

[crystal harness]

alr

also relaunching

btw whqats this error anyways

picture data seems to be loaded fine

except result is very different

its before you came by lul

the 1st preview is the same, i suspect bc of seed

[dry crow]

what was the denois?

[crystal harness]

idk if png info is the denoi

[dry crow]

make sure the webui is whitelisted in any browser adblocker

[crystal harness]

this btw

oh

[vernal garden]

Denoise at 0.7 is too high, means you are allowing it to make 70% changes. Do like a 0.2-0.3

[crystal harness]

oh

[crystal harness]

ok seemingly it doesnt even matter what i set denois to

only spits this out

it was the 1st preview and ye., denois 0.2 to 0.7 gives the same thing

now i cant even use the ai. just keeps saying socket disconnect on the console

or now only txt2img works

nvm

txt2img gets stuck on exactly 50%

[dry crow]

what settings are you using rn?

[crystal harness]

same as before

[dry crow]

can you show them?

[crystal harness]

with png info of the pic

dont mind prompts. the fren prepared em since i had no idea

[dry crow]

your webui is outdated

[crystal harness]

man

how do i update anyway

[dry crow]

can you scroll down to the bottom and check the versions?

[crystal harness]

the github got casted into the abyss of google

its from july if im correct

oh it has links, nice

[dry crow]

yea so half a year old

[crystal harness]

yh

[dry crow]

you can update by adding git pull to the webui-user.bat

[crystal harness]

im sure i mentioned it somewhere? ye

actually

i do have git pull

[dry crow]

what was your gpu agan?

[crystal harness]

gtx 1060

6gb

[dry crow]

remove --precision full --no-half

and it seems git pull doesnt work.
so if you restart check the first lines of the cmd

[crystal harness]

"not a git rep"

do i have to manually make a rep on the main folder?

[dry crow]

no i can give you the commands to fix this

[crystal harness]

ig sure

[dry crow]

To fix this go into the stable-diffusion-webui folder and click in the File Explorer bar (not searchbar) and type cmd and hit enter.

Then use the following commands:
git init git remote add origin https://github.com/AUTOMATIC1111/stable-diffusion-webui.git git fetch --all git reset --hard origin/master git branch --set-upstream-to=origin/master master git pull
Then launch the webui-user.bat

[crystal harness]

btw if i write the cmd in the folder

it sets to folder?

or takes some defualt on C

oh nvm i see

ok it now works

also i already added git pull onto the webui user bat too

[dry crow]

yes that will work now too

[crystal harness]

or nvm

now it cannot install packages bc out of nowhere it got access denied huh

[dry crow]

you can share a sreenshot of the error in #🤝｜tech-support

[crystal harness]

rechecking with the 3 commandline flags

if it happens again sure

after adding xformers medvram and no half vae it let me in

with style databse not found

[dry crow]

ah ok

yea thats normal

styles will get created when you save a style

[crystal harness]

also informed me about xformer

so a style isnt much important, right?

[dry crow]

yes

can be ignored

[crystal harness]

well i am trying to do the png info into txt fix again

ok so

the ai still does the preview instead of the pic i used in png info

but the general quality seems improved

settings are p much same

also whenever im trying to img2img it just dies

writes this seemingly random amount of times and just doesnt let me do anything but close the console

[crystal harness]

also theres no esgran4x+
aka theres the esgran4x and R-esgran4x+

example of img2img death

the progress on the webui comp-letly dissappeared and i can do nothing but close the terminal

yes, webui stops working as well. altho i can still traverse the tabs

[low sun]

images for whats happening:

[dry crow]

What's wrong?

[low sun]

right becomes green/grayish

[crystal harness]

what?

soon after

what is that 1006

its uh this settings. but you can remove everything and put the pic back in. same thing happens

1006

[dry crow]

can you try an other sampler?

[crystal harness]

i can try

euler?

[dry crow]

euler a

[crystal harness]

exact same problem

alr cant test stuff anymore for now

[crystal harness]

Actually should i do some weird reinstalling?

[dry crow]

You could reinstall or you delete the venv folder and test again

[crystal harness]

Whats the venv folder

Also ill try by deleting venv tmrw as today is late, and ill slep soon

[dry crow]

Its the folder that stores the core files for SD to work.
After you delete it, and launch the webui-user.bat again, it will be rebuild freshly

That can fix most of the bugs

[crystal harness]

Oh-

Alr. Thanks for tip

Ill start with this tmrw 👍

Also thanks for supporting

[dry crow]

No problem 🙂

[crystal harness]

Ill propably report back tmrw as it starts working or still breaks
So ye cya and have a nice day/night

[dry crow]

Okay, gn!

[wind wing]

can some one help me generate this image but from a different camera angle?

[topaz musk]

how does this look

[glacial pebble]

能用AI生成一个用虾摆盘的图片吗 摆成个龙的形

[gaunt portal]

Could you make a log for international conference AIRAPT2024 to be held in Ehime Prefecture.

hattori — 今日 16:17
Could you make a log for international conference AIRAPT2024 to be held in Ehime Prefecture?

[wind wing]

it looks pretty good but can you make it view of a person standing across the other side looking perpendicularly at this side of the river in the original image, thanks for trying to help me out @topaz musk 🙏

this is kind of the the original prompt i used: ancient sumeria, standing on a path next to a blue green canal, palm trees and plants, beautiful

[wispy nest]

Infiltrating Anti AI discord

[heady anchor]

abolf changed a lot since he was rejected from art school

[dense current]

any1 knows a model that can replicate this style

[wispy nest]

well curiously i went and asked google gemini for models of that style by uploading the image, but it replied it can't help with image of people, the censorship nonsense is crippling the technology ... but you can try something like https://civitai.com/models/113447/helloflatcolorful2d

[dense current]

style isnt matching it. I wana replace char in the game so style is important

[crystal harness]

can i somehow fix my browser if it gives me 1006? i assume its an error code but i dont really wanna keep swapping default one for it just to reswap soon

[fleet goblet]

I just got access to Gemini in Canada, and it's image creation is rather interesting

Painted Pixel Art looks really cool.

And normal generations as well.

Definitely capable.

[split needle]

how did you do that?

img2img?

[hidden shoal]

gm frens

im looking for design frens that can help on a DAO thats being released in summer

[frigid sky]

Is DreamStudio down?

[dry timber]

What's a good way to get rid of the inpaint smudge on images like this?
Latent upscaling always ended up horridly smudgy everywhere or artifacted to hell and back. Despite only using .2-.3 denoise strength

[wind wing]

my mom

[split needle]

Wow those are nice

@wispy nest which model is that

[jovial moss]

Modern and minimal view of a 5-story apartment

[hoary siren]

what style do u call this

[shut sinew]

Its kind of Minimalism but also not

[sage ferry]

Generate 2024 Year of the Dragon Spring Festival greeting cards

[modest niche]

forgot i had this merger of dreamlike + my sci-fi diffusion

[runic venture]

apple

[tired basin]

Banana?

[tired basin]

Bots are down chief #1047610792226340935

[quiet current]

still no bots

[tired basin]

Inpainting might be the option and only mask the bit of the bottle where the liquid goes

[loud sun]

Chinese dragon, festive and red-hot

[tropic shell]

Please see #1047610792226340935 for more information on the bot. Thanks!

[tawdry magnet]

hello guys ,

can anyone tell how to get this results in img2img? #🏞｜general-with-images #📝｜prompting-help #😊｜co-creators

possible?

#💬｜general-chat

[fierce ferry]

Hello, I tried upscaling a picture but it removes all the dirt while upscaling. Any tipps how to mitigate that?

[paper folio]

I dont know much but maybe use different upscaler which does not denoise but sharpen the image

[fiery delta]

怎么玩的

[spark finch]

4xNomos-HAT-L_otf is such an amazing upscaler... it's just... sooooo ridiculously slow, and goofs the eyes, but a little touch up, or lazy like me run through secondary tile no upscale, fixes that right up. But jesus... soooo slow....

with tiled no upscale second pass

just nomos-HAT-L by itself

[heady anchor]

hat upscalers are the best out there they have more params than any outscaler out there,shame that auto1111 doesnt supports HAT

[spark finch]

😦

did not know., this bastard is a 160mb model upscaler, ridiculous!, and takes like 10m to do its work, but holy hell... it is amazing

[heady anchor]

yea its great 😲

[spark finch]

if I really wanted to play right, I'd adetail or hand paint the eye regen, maybe keep the skin detail cleaner than the noscale tile upscale*

that nomos one is nuts, even better than some of the other's I'd tried, so far it is my fave... I'm sure better exist, but thus far, this is the best I have come across

[heady anchor]

theres also nomos in DAT format,not as good as HAT but still way better than ESRGAN

[spark finch]

I wish they had a 2x

that wouldn't take 10 minutes from an sdxl upscale

I need to change the name of the output, but it really is just...amazing. I am quite blown away by the output, even bare with no detailing, it is wholly acceptable, doesn't have the topaz smearing like you see in a lot, too badly.

[north anvil]

Textgen + StableDiffusion client in development: https://www.youtube.com/watch?v=QcZWAN3xXXs

[hoary siren]

Dreamy Jungle Oasis

Alien moss --- not sure how to make it that ultra crisp hyper realistic look tho lol

[tepid whale]

graphic design

[junior lava]

Been embracing the chaos

used bing image creator, prompt was "combine all art styles"

this one was "dream any style"

this prompt "inspired flow"

Exploring image space is neat

[still rose]

Finally enjoying making AI now I found out the draw things applications for macbook as now the image generations are quick

[thorny frost]

First my test cyberpank world

[still rose]

Hey I like that 👍

[thorny frost]

Next Test

Also try mixed Superman with venom symbiosis, shadowmancer, and few prompt lights like dream lights etc, oh and background from bioshock

[still rose]

Hey can I ask a question?

[thorny frost]

yeh if i can answare 😄

[still rose]

Ok fren could u make that superman image again or sum similar and see how photorealistic u can get it (if that isn't to much to ask I would wanna see how that would look)

[pallid ruin]

It looks so cool

[thorny frost]

Better idea is if you try do yourself and learn, get knowleg. I dont have prompts anymore, when i do others arts i delete old

[still rose]

Ah ok thank you for the reply yh I'm currently learning the realism part

As I come from MJ where prompts are way easier than SD but I don't want to keep relying on a subscription based thing and being restricted

[thorny frost]

yeh MJ is for me to easy i mean SD have more controll and more models, plugins u can controll all, in MJ i think u can only use prompts and u get few styles, in SD u can experiment with styles and i use this for superman

[still rose]

Ah i see ATM I'm using m2 pro macbook pro until I get a RTX 4070 super or maybe a 7800 xt and I'll use Linux with ROCm

[pallid ruin]

The 4070 super will be much faster

[still rose]

So I'm sort of limited ish kinda with some stuff I can do as I have 16gb of ram in the macbook

[still rose]

Yh I prolly won't need to do dual boot Linux like I'm doing now with 6600 xt I'll just use windows

[pallid ruin]

Would you mind sharing the prompt? 🤗 I´ll try to make it more realistic, also which model were you using?

[still rose]

Actually would a used 3090 for £600 be better I can get a refurb one with 2yr warranty

[pallid ruin]

That would be awesome

I got a used 3060 for like 150 😁

[thorny frost]

I have 4070 not super,

1024x1024 10sec

[still rose]

Alr now I just need to know if gaming performance is sim to 4070 super without RT and any upscaling

[gaunt heron]

3090 is a powerhouse card, with 24gb of vram you can play any game you want at 4k

I do

And renders super fast

[still rose]

That's a steel tf

[still rose]

Alr I'm gonna be going from 1080p 240hz to 1440p 165hz with the GPU upgrade I don't play many comp games anymore

[gaunt heron]

You'll be good for years

[still rose]

I have to use Xformers correct to get even faster results or?

[gaunt heron]

yes

[still rose]

Understood ty I'm surprised the 16gb m2 pro macbook pro renders XL models as fast as it does

[pallid ruin]

here where I live a new one is like $700 thanks to taxes U.U so it was really cheap

[still rose]

I think here in the UK we get good deals on GPUs

Well that's what I like to think

[thorny frost]

if u want u can use stable matrix for one click install, and u get there icon settings where u have all options --xformers --medvi something and others to just click

[still rose]

Alr last quick question if I'm making original characters in like the style u would fine in a AAA graphical style like CoD, Tekken and Mortal kombat as a example what model or what would I need? I've been struggling to find answers online

[still rose]

Ty for the info I'll keep that in mind once I get a new GPU

[thorny frost]

you need experiment with models, with one models will be nice with not

for my superman i use few chekpoints models i try realistic is kinda cool like human but i want anime so i use anime models

[pallid ruin]

I accidentally deleted my txt with like 30 prompts 😿, i´ll have to search again

[still rose]

Can u name a few models for me to check out and try, once I find how to do it consistently enough I want to pop it into photoshop and make a game cover or poster style artwork

[thorny frost]

better is to find yourself look i wrote tekken https://civitai.com/search/models?sortBy=models_v5&query=tekken

[still rose]

Ah ty

[brazen sand]

I just noticed how bad are hands in some real life photos, if I had make some hand like that in SD I would have try to correct it

No wonders the AI gets "bad" hand information

[winged token]

hi

image generation is no longer supported ?

[nocturne oak]

#1047610792226340935

[paper folio]

does antone kow what is name of this dress?

what she wearing at bottom

It is like a skirt but not a skirt

its usually worn in office

another image

[wispy spindle]

pencil skirt

[paper folio]

thanksss

[primal spindle]

yummy

[frosty granite]

how to generate images

[wispy nest]

Nothing like ai sushi

[quiet current]

nothing prevents you from using local instance or some online SD services

[tawdry magnet]

instant id not showing in control net control types

i already uploaded control net files in control net foloders #💬｜general-chat #🏞｜general-with-images #🤝｜tech-support

@dry crow

[quiet current]

go in #🤝｜tech-support ? Also make sure everything is updated and reboot your sdwebui

[dry crow]

your controlnet extension isnt updated

[tawdry magnet]

how can we update?

@dry crow

[dry crow]

click on the extensions tab, click, check for updates, then apply and restart the webui

[tawdry magnet]

i tried already.

@dry crow can we remove extension and install agian?

[dry crow]

yes

[tawdry magnet]

may know how?

[dry crow]

you can delete the mod from the extensions folder

then restart and install it again

make sure your controlnet models are in models/controlnet

[tawdry magnet]

@dry crow ok

[thorny frost]

Can someone tell me what i do wrong that hair are not the same qualty? most of the time SD generate me hairs like that half blured halv sharp

[subtle oxide]

hello rif (╯°□°)╯︵ ┻━┻

[thorny frost]

My next test: a cool rodent wearing a suit, cyberpank, white mask, (style one piece:1.2)

[deft bison]

some forms of gigantism from the past...

[ancient hound]

Wow 🤯

[still rose]

Hey people I would like to ask what is the most photo real image u have seen made by AI that u genuinely thought was real?

[azure nexus]

héroes, Vinieron a la tierra para seducir a los hombres de la tierra., ((La ropa interior también se puede ver en la falda.)))) La hermosa mujer en cuclillas frente expuesta., pecho grande y enorme, (((La ropa interior también se puede ver en la falda.)))

[steel apex]

Taylor as a Skyrim companion. I don't care for Taylor but this is nice.

Skyrim guards.

[fallen blade]

any good model to generate simple characters with this 'anime toon look' like this one?

[steel apex]

When you are a level 40 Mage in your first playthrough

[hoary siren]

Thank you!

[hoary siren]

Ngl its not that simple haha...
I used my own refrence i had then with many variety images merged and painted in my own image and sent it back to AI
then i repeated like 3-4 times...

Im more photoshop proficient so i just use the AI to clean up the rough/unnatural edges so i dont have to do as much work lol

[dusk matrix]

Rosa na parede

[limpid lichen]

@pale yoke i clicked that first one about a cat, which was someone faking it by using sd1.5 and then sdxl for img2img. lame. then i looked at the date of the reddit post from 2024. That one linked to qinglong's

[pale yoke]

Oh cool yeah I guess I didn't see that result from 2 weeks ago thanks again, I'll try it out

[limpid lichen]

it also helped that i had found it earlier and remembered it existed

i'm not some amazing googler. if that second click didn't have it, i woudl've given up lol

[kindred creek]

seems to harder and harder to find stuff on google nowa days

[pallid ruin]

Aaaaa nice!

[tawdry magnet]

#💬｜general-chat #🏞｜general-with-images #💬｜general-chat #🤝｜tech-support I would like to pay if anyone can help me to train my Sd for getting these results.

kindly reach me

[tribal mountain]

Im trying to build an image-to-image flow in comfy.ui but I keep getting an error... its something to do with SDXL and i and g. Does anyone now of a place I can get support troubleshooting errors like this?

[static breach]

Helloooo

[spring mist]

StableCascade

[full mortar]

how's cascade at generating text?

[spring mist]

testing rn

[spring mist]

Pass

"holding a sigh that says "love"

[full mortar]

thanks, well it looks fine, if it can do results similar to dall-E3 it would be really impressive

[spring mist]

long text

[full mortar]

mmmh, it got very good with the letters, it'll probably distort text passed through controlnet less

[pallid ruin]

Wow nice, do you know how much vram does it use? I dont know if I should download it or not with a 3060 (12gb)

[spring mist]

unclear to me yet, thoes were online gen

[inner oracle]

I'm wondering if anyone knows a website or service that can for example, remove the text from an image and intelligently fill it back in, to create a textless version of the family feud logo.

[spring mist]

not particularly but the method of inpainting is what you could use to do it

[fleet goblet]

Cascade's a definite improvement.

[inner oracle]

forgot about that since it doesn't work on phones. good thinking!

[hidden whale]

Do we have any stable diffusion models that can generate images in this style ?

[sterile kiln]

Nice, its SDXL?

[tired basin]

any sd1.5 realism model with a painting prompt should be able to turn out something like that

[tired basin]

shakker.ai maybe?

[junior jungle]

Chatrapathi Shivaji Maharaj, clad in regal attire, sits atop his powerful steed, his demeanor exuding strength and valor. With one hand firmly gripping the reins and the other holding his sword, he gazes ahead with a determined expression. The horse, adorned with traditional embellishments, moves gracefully, its mane flowing in the wind as they traverse the rugged terrain of Maharashtra. In the background, the setting sun casts a golden hue upon the landscape, adding an aura of grandeur to the scene. The image captures the essence of Shivaji Maharaj's legendary leadership and unwavering spirit, symbolizing his dedication to protecting his people and upholding justice.

[echo flax]

/"Design a charming anime-style illustration featuring a lovable dog adventurer on a white background, set against a backdrop of serene landscapes or bustling cityscapes, capturing the excitement and wonder of exploration in every pawstep."

"Design a charming anime-style illustration featuring a lovable dog adventurer on a white background, set against a backdrop of serene landscapes or bustling cityscapes, capturing the excitement and wonder of exploration in every pawstep."

[thorny frost]

New experiment, Opal Loras

[river isle]

https://www.instagram.com/p/C3Us87pOkAM/
https://www.vxthreads.net/@fairyracingofficial/post/C3UwQ0IrlT-

fairyracingofficial

Today is Valentine's Day! The cutest day of all~!! ✨💖

#retro #fairies #n64 #🫂 #💋 #💖💖💖

Instagram

Not SD but made with AI

My Valentine's Day gift to the AI community

😚

[hidden shoal]

comfy ui error

[thorny frost]

u dont have this chekpoint change another or download

[pallid ruin]

you have to move or download your checkpoint to your ComfyUI\models\checkpoints folder

you can click on top of the 1.5-emmapruned to select the other models

[hidden shoal]

i cant!

i get this when i click on it

[thorny frost]

man, u dont have any models checkpoints, u need download some and put in folder comfyUI/models/StableDiffusion or checkpoints

[wide radish]

young Harpy woman, with red skin, white eyes, long white hair, and red wings

[mild tiger]

how would I go about getting the "preprocessor generated detectmap" in comfy from a controlnet? specifically with MLSD

[thorny frost]

YT https://youtu.be/gS62nkcml0k?si=gTSFK6qhMCtqGiR8&t=373

[thorny frost]

Comeone know how i can create something like that, library with prompts in comfyUI? Its cool to have ready prompts for everything like seperate lights, arts styls etc

[ember girder]

what mix of asian is this?

because i try to generate a similar face, the asian faces i get generated are a lot more "asian", you know what i mean?

[honest tendon]

bruh it stuck like this what should I do 💀

[smoky oak]

Anybody happen to know how some ody could go about removing the shadows from a generated image? Such as a body, removing any of the shadows or lighting off of it to use as a flat texture for a model? I'm not too sure how I would do it, was curious if anybody else knew

[quasi patrol]

Hello all please can anyone give me this type of prompt please just one i will practice

i tried all but getting bad errors so help me out guyz

[dry crow]

You should install the webui following my guide.
Its in the Pinned Messages of #🤝｜tech-support

[sterile kiln]

Hey guys, Stable Cascade is working on RTX 2xxx?

[signal reef]

Is there a way to use openpose controlnet or anything else to set the pose of a non human character (eg animated apple)?

[tiny pasture]

Can this server generate the required images for me? May I ask who knows?

[tardy tinsel]

I made this in a tool I'm building called pettypaint.com . Using images drawn in the browser + images from different image platforms collaged together to generate a new image. It's meant to be used in VR or desktop computer. It has a few hundred different XL and 1.5 models available with a few hundred loras that can be used in combination and text prompts.

Long story short: I can't draw, but i like to try, and I use XL and 1.5 models to enhance my creations.

[fleet goblet]

So Sd.next inpainting?

You can already do masking in most SD repositories, whether by native or extension.

[quiet current]

#1047610792226340935

[narrow osprey]

#💬｜general-chat

[limpid lichen]

so it's software as a service instead of running natively?

bummer

[spring mist]

Maybe this would handle it better, i havend used it https://github.com/abehonest/ControlNet_AnimalPose

[green elbow]

hi can we use stability.ai img2img inpainting API with Controlnet?

[limpid lichen]

depth map might be better to create keyframes with for that style

[delicate wasp]

Hello everyone, please I have a question, still a newbie. With the stable diffusion image to image, please could anyone explain why the generate image does not match the body size of the in put.

[ebon steeple]

Try resizing the head to fit ur models body

[delicate wasp]

ok thanks very much

[gentle eagle]

great

[wispy nest]

https://cdn.discordapp.com/attachments/640978348830490625/1208337192003444737/689.mp4?ex=65e2eab4&is=65d075b4&hm=32d866e9a39aea76e672de39fb743fc716818f990edc7c363f6d9de58757fbd6&

[subtle depot]

I made something similar today, hehe

While testing out my new GPU

[keen copper]

noob alert Hi guys, how are you making these images? Are you using SD XL 1.0? Are all of these made using a subscription of clipdrop.co ?

[radiant saffron]

/game_statsone back view,one Front view, kawayi, look like Golden Toad, The three-legged golden toad, with a tail of one foot,cute, wearing red Men's cheongsam,like wealthy,smile on his face, whole body, 100% white background, simple 3d mode,3D rendering Natural lighting, 16K , best quality --niji 6

[digital spruce]

How are you guys creating free images?
Im a noob

[prisma surge]

@digital spruce we can't since more than 3 weeks, because the bot is down. Nobody know when it will be repaired.

[keen copper]

What about the above images shared

[prisma surge]

From what I understood. You can download Difusion or other AI locally to create images with your own computer.

But you need a fairly powerful PC to successfully run this type of software. And not having to wait 1000 years to have usable images (and hoping that the software does not crash the PC). For example for me I think it's dead.

[dry crow]

whats your GPU?

[loud willow]

check the game i'm playing

[prisma surge]

@dry crow here are the specifics of my PC (and my GPU). Unfortunately, I'm not quite a noob in the IT field.

I have no idea if it's really ok or not to run good AI creation software locally on my PC. I had tried tutorials on the internet to have stable local distribution quite a few months ago, but it hardly worked (10m of waiting to produce a single image of ridiculous size = unusable).

I don't know if it's much improved or not at this level over the months and if my PC could better support the news.

[edgy rover]

I need to buy credits "as I go". What can I do?

[dry crow]

hey, thx for the info, unfortunate this is not enough to run SD on that GPU

only on CPU moed which is very slow

[prisma surge]

@dry crow Unfortunately you are right. I'm afraid the only solution is for me to buy a really good computer, knowing that I want it to also be able to run the most demanding games.

Please, do you know how long it takes a good recent 1500 euro laptop to be able to create 4 images at a time of the same standard size like midjourney ? Many youtubers use desktop computers, or cut the time it takes for AI software to create images locally. So I couldn't find this information.

[subtle depot]

I’m using SD 1.5, SDXL, Turbo and the rest

Locally on my computer

[prisma surge]

@subtle depot please, once you have validated your prompt, how long does it take your computer to create this type of image locally ? 🙂

Do you create the images one by one, 4x4, or more?

[subtle depot]

Depending of your settings, your computer, video card etc., anything from a few seconds to some minutes

The image I posted earlier took about 4 second, to create the base image I then tweaked more in with inpainting

[subtle depot]

This image, @prisma surge

Yes, I can sometimes do batches of images but mostly go one by one while adjusting the settings @prisma surge

[prisma surge]

@subtle depot thanks a lot for your answer. Could I know the specifics of your PC and the price range it cost you ? History that I can compare to mine, and to the PCs currently available on the market?

[crisp stream]

everything looks a bit too clean for my taste and I´m missing details/complexity. It also tends to portrait everything and it at least seems like cropped/zoomed in images have been used for training. Over the release of versions, while the general quality has improved the complexity has suffered a lot.

This is SDXL in comparison:

[crisp stream]

SD 1.5

[pallid ruin]

I hope community fine-tunned models improve this @crisp stream or at least stability with the next sd model

[crisp stream]

hope so as well, just wondering why it has that trend in general

[pallid ruin]

Same prompts

leosamsHelloworldXL_helloworldXL50GPT4V

sd cascade

🤣

Maybe its just because they removed famous people or something, in other examples sd cascade wins

sd cascade

sdxl

"a cute mouse typing on a keyboard, playing the game minecraft"

Mmm most of my old prompts don´t work correctly on sd cascade, I gotta try new ones

Well the same happened from 1.5 to sdxl so its not the worst thing ever

sdxl

sd cascade

😍

Here sd cascade wins

sdxl

[crisp stream]

@pallid ruin like said, the overall quality has improved, yet it got less creative so to speak over time. Your example of the mouse shows it pretty well, while Cascade looks llike higher quality (though it´s got that typical more stereotypical Cascade touch I exoerienced so far, almost like a filter) the SDXL one is way more interesting. I feel like Stability AI seems to go for a generic approach where everything looks cleaner, more accurate, more real in close up. And while I can se the use for it in terms of mimicking real stuff, it simply gets more boring from an artistic point of view. After all I feel the quality and also strength of AI is its weirdness, its surprisingly out of this world composing and combination of things, reason I endorse following a creative part, at least as well. I know many want to produce another even cleaner lewd image and cuuuuute images of women and cats (sry for sarcasm 😄 ), yet please let us be versatile, complex and also going beyond state of the art 🙂

[crisp stream]

I don´t really mind adjusting, not the least for adding versatility, then if it makes complex weird things hardly possible anymore I feel like "come on, where´s the fun in that?". It feels similar to when 3D arrived, when people were like "Aaaaaw, that lion in the staircase in Jumanji looked so good", I felt like "boooooring". If I wanted to see a lion I can go to a Zoo, to Africa or simply look at a photo. It´s about things here we usually don´t see or haven´t even thought of to see what makes anything in that regard interesting from an individual perspective 🙂

[crisp stream]

appearance/look quality-wise yes, composing/interpretation-wise SDXL wins (here). If you look at that chef´s hat, it´s more realistic, sure, yet in SDXL it´s hilariously caricatural, while SDXL as well adds more (creative) details to begin with (like water splashing). Like said, I can see the use for a more classical photo-realistic clean look (like stock photography), yet I perceive AI image generation a creative tool at first and that´s what I´m playing and would like AI staying able to produce

[pallid ruin]

Yeah maybe that´s right, it also took into account the background, I forgot posting the prompt earlier "a cute pigeon cooking dinner on the strawberry forest, wearing chef clothes and hat"

I hope stability uses a better captioning for the next model, that´s what dall-e did with dalle3 (the only problem is that is a censored mess)

[faint rose]

I need help

[crisp stream]

and at least by default Dall-E applies the same style over and over again, similar shapes, similar contrast/colourscheme

[crisp stream]

Who doesn´t? 😄

[faint rose]

This ruff thing keeps failing

[pallid ruin]

Yes, I also noticed that it´s IMPOSSIBLE to get photorealistic images, it was possible back then

[faint rose]

@crisp stream can you help ?

[pallid ruin]

What is that? is it for sd? 👀

[crisp stream]

sry, not expertised, at least in that regard 🙂

[faint rose]

Yes , I have it on a sd card

[crisp stream]

Didn´t know there were Stable Diffusion cards to begin with 😉 😄

[faint rose]

I mean I download the stable defusion on a sd card for more space and not my computer drive

[pallid ruin]

What repo did you use?

[crisp stream]

I´m pretty sure I´ve seen quite a few photorealistic images. Can ask someone who´s dabbling around with Dall-E about it

[faint rose]

What is a repo

[crisp stream]

was just joking, joker 😉

[faint rose]

I'm using github desktop

[crisp stream]

like said, not experienced in that regard

[heady anchor]

what does ruff have to do with SD

[faint rose]

@crisp stream sorry I can't take a joke because I'm awake at 1:28 am with limited brain cells

[pallid ruin]

I mean if you were using automatic1111 , comfyui or other repositories @faint rose because I´ve never seen something like you sent

[faint rose]

@heady anchor it's not working because it's stuck at ruff

[crisp stream]

the joker is supposed to take jokes anytime regardless ;-):-D

[pallid ruin]

You can try asking for help on #🤝｜tech-support , maybe someone there knows about the error @faint rose

[heady anchor]

sorry but idk why you using ruff to install SD

[faint rose]

@crisp stream just lost a other Brain cell

[crisp stream]

naah, it will add back up once you´ve risen from the doom of time

[faint rose]

I just might go kukukuku, kukuk

[crisp stream]

42? 🙂

[faint rose]

Crazy

Well anyways thanks for the help guys

[crisp stream]

pleasure 🙂

@pallid ruin just one example from Dall-E done by to.ok (also asked for more realistic realistic stuff)

[pallid ruin]

It looks nice!

[crisp stream]

yes 😄

[pallid ruin]

I downloaded this image but I could never find the prompt 😁

Also I have 0 imagination, I tried generating something similar but it didn´t work

[crisp stream]

isn´t there anything like Lexica for Dall-E? Ask ChatGPT (if it knows about Dall-E 3 that is)

[pallid ruin]

I´ll try with llava 1.6

[crisp stream]

Here another one from Dall-E 3 by to.ok

[subtle depot]

Sure thing. I have a self built system including: Ryzen 5700X, Asus B550 Plus, 16gigs of system memory and Geforce RTX 3090. The computer also has other components like hard drive, power etc. Total cost is at around 1500 euros, the 3090 is second hand and easily the most expensive part there. My second computer is Mac Studio M1, cost about 2500 euros. The Mac is quite slow as it’s not optimized for stable diffusion yet but still very usable and ok. I have it mainly for my music stuff but is also a killer for video editing.

Depending on your budget there’s a lot of possibilities and combinations you could go. For around 1000€ you could get a decent computer for stable diffusion and perhaps for gaming, if you build one from parts your self.

[crisp stream]

Mind the most important component is the videocard, the rest doest mattert mauch

[blissful cairn]

Raspberry pi with 4090

👀

[subtle depot]

Sure the GPU is the most iportant single piece for Stable Diffsion.

[crisp stream]

yes, just pointing this specifically out because you wouldn´t need to spend that big of an amount on the usual peripherie 🙂

[crisp stream]

😄

[subtle depot]

I think around 1000 euros is something you can get a decent machinery with, in case you're getting everything brand new. You can easily go under that with second hand parts, and picking something from sales etc.

CPU actually matters a little how fast your GPU operates. I've seens some comparison about this. So with my computer for example, if I'd have a diffeent CPU I might be able to squeeze a bit more out my gpu.

But sure, it's fair to say the gpu is the most important and the rest does not matter that much. I'd just add maybe to that line, as long the rest isn't something too slow and old.

[crisp stream]

yes, indeed, thanks for pointing out

[crisp stream]

Btw the match just started (landing hot) so I quickly had to alt-tab over and couldn´t write "that" anymore, then you had already replied anyway 😄

[subtle depot]

haha, cool. What are you wathing, which match is it?

[crisp stream]

Apex

[subtle depot]

Nice nice!

[crisp stream]

currently diving again 😄

[subtle depot]

Have fun! 🙂

[crisp stream]

already back to lobby 😄

already connected again 😄

[crisp stream]

ty

[steep sleet]

Using IPAdaper-FaceID (PINOKIO) I have swiftly tailored some pictures of a young woman eating cotton candy. Barbie-ish color scheme!!!

[wild sorrel]

I'm more impressed ESlint has no problems with it

[rocky echo]

prompt high quality, 8K Ultra HD, vivid colorful, Seamless patterns, fabric art, Art station, stary night many colorful detailed Design combining flower and leaf, splash arts, aesthetic for Tshirt design, white tone, photorealistic, ultra realistic, by yukisakura, awesome full color,

[keen copper]

and @prisma surge thanks for explaining! The hardware required is quite robust for making images. I wonder for all the people invested, if it's really really worth the investment. I mean the amount of GPU prowess (24gb of vram) required is insane lol. I think I'm going to be dependent on online services for now. What are my best options? I only know of MJ as one of the best ones and recently started playing with SD XL 1.0 on together.ai. The latter is what led me down here 😄

[subtle depot]

You can get along and do images with much less than 24GB of VRAM hehe. Something like 3060 12GB/8GB is just fine, or even less than that

[keen copper]

Thanks! I think it would make sense to invest if I can turn this hobby into something commercial. Which is what I'm trying to see via online services..

[subtle depot]

Things are changing and developing fast in the AI field, investing something massively might not be the best thing

I'm really for the day when AI learns actually making good looking stuff without having to put in extra negative promts etc unless you actually want something. But normal stuff like anatomy would work out of box

[keen copper]

Some fav pieces I could scratch out with SD XL 1.0

[zealous wolf]

translate this picture to english

[faint rose]

@crisp stream

I fix it

I'll send you a picture

Joker getting his brain cells back

[faint rose]

I'm doing this on a Nvidia GeForce GTX 1650

[crisp stream]

🐶

[blazing field]

Hello to everyone!! 🙂
I need to be useful to my wife. She is an aesthetic doctor and needs an image (before and after) for a gummy smile treatment. You know. A smile with a very thin upper lip and when smiling, the gum and frenulum are prominently visible. This is a very simple and beautiful medical treatment.

I'm trying to make the first image, the before. I'm using "Fooocus", because I saw that it is an AI specialized in faces.
It should be noted that I don't have much of an idea and I have managed to run it because there is a .bat on github. Otherwise I wouldn't know how to do it.

The fact is that I can't get it to display the image correctly. No matter how detailed the prompt is regarding the topic of gums, it only makes normal smiles. Normal lips + pretty teeth. And it is precisely the opposite what I am looking for. A variant of that. But I don't know how to do it anymore. Can anyone help or guide me?

infinite thanks in advance community!

I used this: https://github.com/lllyasviel/Fooocus?tab=readme-ov-file
veeeeeeeery grateful in advanceeee

I used this prompt: "
A close-up view of a woman in her 40s, capturing her wide smile with her thin upper lip, clearly exposing the gum tissue and the frenulum that connects the lip to the gum. The texture of the gum tissue is pronounced due to her wide smile and the tension of her upper lip, which draws the viewer's attention directly to the space between her lip and teeth. "

What i have said wrong????

[vapid sigil]

doctor with a beard doing an ultrasound

[bleak sable]

something i made with stable diffusion like a week ago

[desert spire]

Здрасььььтеее

[pearl ginkgo]

Free myself

[desert spire]

RIP DREAM

OK

[desert spire]

OKOKOKOOK

[desert spire]

you are an excellent psychologist

[clever oar]

доброго вечера

[desert spire]

добрейшего:)

[limpid lichen]

what is with model creators marketing their modesl with sales speak and hype language?

[clever oar]

был бы добрее если б я мог запустить любую модель на своей 4 гиг видеокарте)

[desert spire]

этот вечер не заслуживает говорения о проблемах, все равно я не смогу купить тебе более лучшую видео карту. П.С. и себе:))))))))

[clever oar]

судя по твоим картинкам у тебя уже есть что нужно)

[desert spire]

rip dream

[edgy rover]

The AI wrote the ad.

[limpid lichen]

dont think so. it's too tailored to each merged model and uses too current ai language. someone crafted it.

[limpid lichen]

https://civitai.com/models/301974/falkor-fuchur-neverending-story-sdxl fuck yeah. sunday civit combing is reaping great returns today

[spring mist]

sooo many things

[clever oar]

why

[desert spire]

becouse

[clever oar]

oh this bad(

[desert spire]

good good

[clever oar]

😭

[desert spire]

[clever oar]

good night 😴

[edgy rover]

Some stuff I've generated

[mild jay]

How is this lamp for sd 1.5, is there anything i could improve upon by chance?

[arctic laurel]

https://youtu.be/gK7pMKkj4kE Fooocus + SVD

[frank onyx]

is there any way to fix this error

[quiet current]

#🤝｜tech-support

[formal kraken]

just discovered the meme gold of SD 1.1

[thorny frost]

Guys what upscale you use for real photo faces? I have problem with uspcale settings

1. raw first photo

2. after upscale

after uspcale is there is worst eyes and teeth

3. after ADetailer

you need guys open in window full zoom

for upscale i use

[glacial quail]

/一只丁香色的边界牧羊犬，走在草原上

[quiet current]

That s not the bots channels, also bot channels are down (cf #1047610792226340935 ) and please stick to english.

[thorny frost]

ehh

[jovial jungle]

Personally I avoid upscalers like the plague and just use inpainting. More work, but much higher quality.

[thorny frost]

but in inpainting u can only manipulation photo, change something, not upscale image to bigger resolution, i use resolution upscale because if i generate 1024x1024 in base i wanto to 2048x2048, how u can upscale image in inpainting? with hires? hires also have upscale esragon 4x

[jovial jungle]

I'm not sure what the default upscaler is in your software, but in the webui's extra tab you can blow up your image without changing any details, you can inpaint afterwards for the quality.

you can get stuff like this doing that method.

though I did enchance it in photoshop a bit

[thorny frost]

do you have a moment for look what i mean? i will share u screen on dc?

[sharp crater]

Which are the top-3 or 5 general open source image generation models right now?

[limpid lichen]

i was hoping someone would diffuse this. Quality composition!! thank you o much for sharing. i been waiting for it

[limpid lichen]

OH LOBABOBLOBLAW IS YOU. Thanks even more

[arctic laurel]

you’re quite welcome 🙂

[lethal minnow]

anyone know what extension this is?

[wild sorrel]

I had something similar, search for aspect ratio extentions, it should be there

oh wait...if whole thing on the screen, including "performance" - is part of the same ext - idk, I had only aspect ratio part

[heady anchor]

thats not a extension,thats foocus another webui like a1111

[boreal brook]

画一只会飞的蝴蝶

butterfly stay rock

[quiet current]

bots are still offline and that wouldn t be where nor how to post request for it, cf #1047610792226340935

[umbral apex]

prompt: a pink dog

[faint rose]

Can any one help me understand this

Problem solve

[wispy nest]

Heya

Which companies offer private image generation based on stableai?

[feral compass]

Hello teachers
Where do I download ControlNet in this picture?
For reference, it's at the bottom of the img2img tab.

[lethal minnow]

they said they were using disco too
are they compatible

is disco a safetensors or what

[wispy nest]

just dont use safe anything

image generation is freee

[analog basin]

my generated images looks nothing like what i uploaded on controlnet what am i missing here?

[analog basin]

also, installing stable diffusion made my youtube videos messed up lol wtf

[limpid lichen]

https://cdn.arstechnica.net/wp-content/uploads/2024/02/Will_Smith_fake_AI.mp4 legitimately thought was a new ai version

@wispy nest

[wispy nest]

oh wtf

why does it say ai videos now

[limpid lichen]

year ago people made "ws eating spaghetti" and he released his own spaghetti eating video now lol

thats the joke

[wispy nest]

oh phew

will smith is funny man

[limpid lichen]

even if that chris rock moment wasn't a bit, it was funny as f

a moment that sparked an amazing year of jokes

[wispy nest]

it prob was. he was laughing at first and then just changed mood like bam

[limpid lichen]

it was surely not a bit. in the moment we were all unsure, but since, a lot has come out. Rock's comedy tour touches on it in a few funny ways.

[wispy nest]

oh fair lol

[covert raptor]

does anyone know how to setup a good face swap? willling to pay $

[lofty ivy]

Hey Diffusers. I’ve noticed the eyes on my final renders are abit weird, with artefacts and the eyelashes bleeding into the skin. I’ve tried using prompts and negative prompts but it doesn’t help much… any thghts?

[heady anchor]

thats normal you need to use adetailer to fix the faces otherwise they come out deformed,specially at low resolutions

[oblique raptor]

Hello! I'm trying to use this https://platform.stability.ai/docs/api-reference#tag/v1generation/operation/masking, but it generates a completely different image everytime and it seems to be ignoring the black mask. Any ideas or suggestions?

[sweet harness]

Hey Diffusers. I’ve noticed the eyes on my final renders are abit weird, with artefacts and the eyelashes bleeding into the skin. I’ve tried using prompts and negative prompts but it doesn’t help much… any thghts?

[silk panther]

#dreambot

[keen stone]

these AI are basically unable to create good masks from what ive seen.

[fierce ferry]

Hello, a friend asked me if I can make him a picture for his tabletop tournament. He asked for a medow with a river in the middle and two knights duelling, one in a white and blue checkered tabard, one in a black and yellow representing the two clubs which organize this. It is no commercial event though ...
I created that with comfyUI and quiete like it. But it has flaws (the right knight should have more heraldry and also the weapons look kinda odd. Since I am new to SD and see it at a good way to practice (I need practical examples to learn) I search a way to fix those problems but lack of a good starting point. How would you fix the weapons and expand the knight? Do you have good starting tutorials for that? Regards

[fierce snow]

A silhouette of a warship, abstract blue water ripples, with a blue background.

[quasi patrol]

Hello 👋 I have one doubt in face swap please can anyone help me it will great please guyz because very important to learn that so help me out 😌

[oblique raptor]

That's quite sad. It feels like image-to-image/masking is just the same as image-to-image. Any suggestions on how can we can improve masking? I'm trying to implement in our app Picasso. You can check it out here: practal.ai

[quiet current]

You could inpaint the little details, use controlnet to correct the poses and maybe use some loras to correct the look of some weapons (I always struggle with guns on SD)

[honest tendon]

how to get models for controlnet? im following a youtube tutorial but cant do exact follow because I dont have any available model

[quiet current]

you have to download them (from huggingface, civitai or whatever) seperately . Make sure to use controlnet models that match the current SD model architecture you're using (don't use a sd15 controlnet model if you're currently using an SDXL model)

cf https://stable-diffusion-art.com/controlnet/

[honest tendon]

I still dont fully understand the classification of many programs/extensions/models/etc.

From what I understand now 1111 is a UI for using models, SDXL/SDXL turbo is two official model from stable diffusion but 'just' a model like what I can found in civitai, and Controlnet is 1111 extension that somehow need a model to run but I dont really understand is it the same "main" model I use in 1111 or comepletly different model specifically only for controlnet extension

[keen stone]

no i have no clue, i myself am looking for good AI for generating stuff... there's surely improvements still to be made

[quiet current]

It can be confusing at start. It's gonna be a long post. But hopefuly everything will make more sense after that.

Model is a very generic term for "file holding some sort of data to be used with a neural graph network". There are different algorithm/programs using neural graph networks, stable-diffusion is one of them. But 9 times out of 10 when people use the word "model" they mean "stable diffusion model". Meaning "models" that are to be used with "stable diffusion programs".
There are also many different Stable diffusion models, the main official ones are : stable-diffusion 1.5 (or sd1.5, sd15), stable-diffusion 2.1 (sd2.1, sd21), stable-diffusion-xl (sdxl).... Just to mention them there are also stuff such as stable-cascade, stable-diffusion 1.6, stable-diffusion-turbo, StableLM, etc.... 99% of the community made models are using one of the official model as their base and performed additional traininng on them to give it some "extra flavor".

A1111 also known as; automatic1111, automatic1111's stable-diffusion-webui, sdwebui... is a UI / a program made to use all the image generating stable-diffusion models, including : sd15, sd21, sdxl, community made ones, etc... A1111 version has nothing to do with SD model version.
Controlnet is an extension of a1111 so an additional plugin for it. And yes it does require its own model to work as it's not working alongside the stable-diffusion pipeline. The goal of controlnet is to take some input image, extrapolate some information from it using depth recognition or line recognition, segmentation, etc.... and use that data to """"constraint"""" the output of stable-diffusion. So controlnet models are made to analyze the input image not to generate stuff unlike SD models. But because it still communicate with SD, each controlnet model are made for a specific SD model (eg: "control_sd15_depth.pth" is made to communicate with other sd15 models and works on extrapolating depth informations from an input image) .

[honest tendon]

So all image generation actually processed inside or by the models. A1111 is just the user interface so these models can be used

SDXL is one of most popular model and official, and community made models like we can found in civitai is actually modified official models, and still fall under same classification as "model"

ControlNet is one of many A1111 extension, the use of it is to extract some information from existing image and add it as extra input/control for image generation. Controlnet Models is completely different thing than actual image generation "Model", and the difference of each Controlnet Models is on how they extract information from an image.
Plus ControlNet only works on some models, so I cant use it on every community made model. please correct me if im wrong

[quiet current]

Pretty much bang on except for the last part.
"ControlNet only works on some models, so I cant use it on every community made model." => You can use controlnet on community made models as they're using the same architecture as their base model. However the same limitations apply, if a community model is made using sdxl as a base, you will not be able to use a controlnet model made for sd1.5

[honest tendon]

okay if the controlnet model only works with sd1.5 architecture then all community made model that actually modified version of sd1.5 will works

[honest tendon]

thanks alotfor the explanation it really helps

[quiet current]

Keep in mind that I gave you an overly simplified explanation of everything.
So while you can keep that diagram of yours to better understand "how to use it". If you show it to actual AI specialists / students they'll probably be foaming at their mouth seeing it. Mainly because controlnet is actually altering SD's neural graph network.

[honest tendon]

Okay gotcha

so people use ControlNet to create something like this

[quiet current]

Could be. I don't see any clear usage of controlnet in this one. Seems like a pretty "generic" landscape to me. I don't see why someone would need to ue controlnet to get to that kind of result.

[honest tendon]

Its Rick Astley, I see people in Instagram posting something like that as a meme

[quiet current]

oh yeah I see it now.

Yup that would be controlnet QRCodeMonster thingy

[fierce ferry]

Thank you, is there a good tutorial on doing that? I haven't used inpaint and controlnet yet

[quiet current]

https://learn.rundiffusion.com/hidden-text-and-logos-with-stable-diffusion-controlnet-and-qrcodemonster/

that's gonna teach you how to use but probably won't teach you how to install it, google where to get the model and it should be fine

[fierce sphinx]

a girl

[dawn seal]

Rick Rolled yet again!

[timber frigate]

...

[deft bison]

All the satellites are hanging on balloons above a Flat Earth ! ... people wave their hands and shout !!! ...

[queen scarab]

yo not sure if thats the right channel but I have a question regarding stable, diffusion and controlnet. I currently (for testing purposes) try to change the hair of a girl (shorten, colorize it). The color works and the shortening not really works while preserving the correct image. I know stable diffusion has its limits, due to lack of imaginating whats exactly under the hair (if you shorten them), however my attempts are like this in the image above. Im currently experimating with controlnet (canny) and "a person mask generator". Any tips on how you would approach this problem?

[native yarrow]

Hey man! First of all, the image looks super nice and I love the art style, I can't help but admire the type of art that you used and I would really like to replicate the art style, I can't really understand which promt you used for it (just the painting type and art style), could you tell me what prompt, Lora and maybe model you used to make it? I'd like to maybe create a character of myself using it or just scenaries, thanksss!!!

[lofty ivy]

Thank you!

[tired basin]

@oblique raptor - you gots issues chief

[oblique raptor]

Woah! Thanks for the info!

[tired basin]

Let us know when its sorted 🙂

[oblique raptor]

Yes Sir!

[lofty ivy]

hey so i got adetailer to help me fix my character eyes and hands and honestly its making no changes, or making things worse. check it, this is the before:

and the after

like atleast with the before, i can manually draw and make changes on photoshop... am i missing something?

[lofty ivy]

halp 😛

[sharp crater]

Can someone suggest me some high quality "general" dataset that contain prompt and images generated from top models maybe SDXL etc..

[tired basin]

You could just head to civitai and collate to your hearts content

[spring gulch]

Why does this situation occur on the arm of the Lora I refined, what is the reason for it, and how can I avoid this error

[dry crow]

Your on AMD?

[subtle depot]

hmmm, clearly making things worse imo, before and after. You can see the processing area and colours etc are altered in a bad way. I sometimes get those when I use a LoRA or Checkpoint that isn't fully compatible with adetailer. Denoising strenght being too high, too many steps if using turbo or distilled model, possible weights in LoRAs etc might also cause this.