🏞｜general-with-images

[hazy warren]

Does she even lift?

[deft bison]

a hermit observes from his window the hubbub of a bustling civilization...

[woeful hare]

you gotta check your local vendors, Ive been ontop of the prices since the 30 series since it was hell to get anything at msrp back then

[nimble mason]

Yup still the same situation with local vendors

I already have a 4090 so I've stopped looking seriously but yeah

[ebon steeple]

I know some of them are real enhancers

[ebon steeple]

integrating advanced digital techniques such as ADK (Artificial Detail Kernel), hyper-diffusion, LDSR (Local Detail Super Resolution), and DAT (Dual Attention Transformer) for depth and complexity. SwinIR_4x, cutting-edge image resolution enhancement.

[hazy warren]

😮 gotta know what’s best for the job and will give the most detail

[lost valve]

most of those are trash

there are only a handful of good ones

maybe an anime upscaler too if ur into that

[hazy warren]

I didn’t see 4x sharp on there

[lost valve]

its because u have to download those

[hazy warren]

Does it add more details? Or just shapen it?

[lost valve]

both

the NMKD give superb face details

[hazy warren]

In SD i tend to get bad skin details, would it fix those?

[lost valve]

it is one step towards remedying it yeh

another step is loras

[wispy nest]

as a time traveling scientest in futurisum earth in 1 million years advanced philosphers world nebula star city a diffrent universe timeline a paradox of life and ai back to the future mad science action knolage scene town scape arora borealis final fantasy star ocean day dream ultra hyper futura hologram futura black light nova alien tech vision if the meaning of life wisdom and peace rembrant Michaelangelo master acid trip shroom trip quanta quest 3d meta verse scape scene movie scene

Thats like a 3rd of the promt

The other 2 3rds will be a mystery

[ebon steeple]

Defenitly listen to @lost valve

[lost valve]

inpaint that joint g

[wispy nest]

Bro has peg leg

[wispy nest]

Its interesting asking gpt something and then asking it to say less woke is as if it has a woke filter that is actually masking something deeeper

Most people interpret it as being a baby ai but I think diffrent

[wispy nest]

Like how a camera has multiple filters but few lenses

[quasi patrol]

Hello why always get this results guyz can someone help to fix this please

[wispy nest]

I'm guessing your refering to the hands add more anatomy key words and 5 fingers natural form higher step img

[carmine quail]

https://preview.redd.it/qry4ravodymc1.png?width=1024&auto=webp&s=4dcce11d808c9c9ab3f7d78ce6535b555eb7564a

https://tenor.com/view/spongebob-squarepants-begging-pretty-please-beg-on-your-knees-pray-for-mercy-gif-11592581965281036637

[quick notch]

"Create a 3D anime-style watercolor cartoon scene illustration. Tom, an 8-year-old lively boy with slightly chubby cheeks, trendy light blonde hair, and vibrant blue eyes, seen laying comfortably on his bed. He is chatting with his unique green wrist watch that has a face on it. A mixture of enjoyable expressions on his face adds vitality to the scene. He is wearing his fun cartoon patterned yellow T-shirt and relaxing pants."
请生成图片

[nimble mason]

Here are the images you requested.

[wheat heath]

怎么使用

[nimble mason]

Here is the image you requested.

[ocean turret]

guangzhou

create guangzhou images

[noble zephyr]

i think i improved the consistency quite a bit. still not quite as consistent as i'd like it to be though.

[ocean turret]

How to create images through text?

I want drawing a dragon.

@ocean turret...

[noble zephyr]

if you want that kind of style, you can try this lora. i haven't tried it myself, so it may not be what you expect. https://civitai.com/models/269440

[ocean turret]

thank you!@noble zephyr

[noble zephyr]

no problem! :D

[ocean turret]

👍 @noble zephyr

[scenic wasp]

@nimble mason creat A European Cup football studio ，with 1000 square meters, featuring three areas. One area has a large screen, with a SUV Volvo parked on the left side. In front of the screen, there are seats for 7 people. On the right side of the screen is a cyberpunk style street scene. The overall theme is a summer outdoor drinking scene. On the right side of the street is a stage for performances. The large screen has XR capabilities, allowing for the expansion of the sky and distant forest scenes.

A European Cup football studio ，with 1000 square meters, featuring three areas. One area has a large screen, with a SUV Volvo parked on the left side. In front of the screen, there are seats for 7 people. On the right side of the screen is a cyberpunk style street scene. The overall theme is a summer outdoor drinking scene. On the right side of the street is a stage for performances. The large screen has XR capabilities, allowing for the expansion of the sky and distant forest scenes.

how to use ?

[noble zephyr]

Clownshark Batwing isn't the bot, they're a human being, generating images for people as they please. you can't just demand them to generate an image for you. the bot is currently down. keep an eye out on #1047610792226340935 for updates

[scenic wasp]

i see 。。。

[nimble mason]

Here is the image you requested.

[nimble mason]

Beep boop. Your feedback has been registered in my data banks for processing.

[noble zephyr]

lmao

[ocean turret]

create dragon images.Thank you.@Clownshark Batwing.

@nimble mason create dragon images.Thank you.

[noble zephyr]

i have some right here #🏞｜general-with-images message

[ocean turret]

@nimble mason please create guang zhou city images,thank you.

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[ocean turret]

@nimble mason Thank you very much. are you human or robot?

[nimble mason]

Beep boop. I do not understand your question.

[ocean turret]

@nimble mason Sorry. I am curious.

[quiet current]

@ everyone new bot channel is now named Clownshark :p

[nimble mason]

Beep boop. Correct. Author's warning: my prompt comprehension is known to be poor. Some deviation from desired results may be observed.

[agile hatch]

Mahadev

Need image of shiva in kailash and snow

[nocturne oak]

So make one

[agile hatch]

@nimble mason give image of shiva with snow

[nimble mason]

Here is the image you requested.

[quiet current]

As a reminder, please keep in mind that Clownshark is providing images for free, they re not a bot. And yes the prompts are tweaked a little bit :p, so no they re not representative of actual SD output.

(shiva being a divinity I wouldn't want people to get the wrong idea and start picking pitchforks)

[deft bison]

approach through the imagination some aspects of ancient pharaonic architecture...

[quiet current]

I have a couple of egyptian stuff I did last year (probably even late 2022) that I never shared here I think. It s old stuff so there some flaws.

[copper bridge]

The packaging box for toy cars

[quiet current]

nope, this is still not the bot channels, and the bot itself is still down, cf #1047610792226340935

[copper bridge]

Hi, may I ask for text to generate images? What should I do

[quiet current]

install one of the UI for stable diffusion locally, or use stable diffusion online services such as https://dreamstudio.ai/

[copper bridge]

Thank you. I'll try how to use it.

[deft bison]

the nighttime nightmares that feed on our negative behaviors...

[modest vector]

Seal astrounaut cute

[woven swallow]

||wide angle，the giant space station of the future ，cloudy,galaxy,stars，light effect，colorful ，by Greg Rutkowski, unreal engine, James Gurney, artstation，highly detailed matte painting，bright color scheme||

[quiet current]

nope, this is still not the bot channels, and the bot itself is still down, cf #1047610792226340935

[steady skiff]

Are we still weeks out?

[errant totem]

@bot planta eating cat

[quiet current]

....

I m not gonna copy pasting the same line every 5 posts, am I.

[nocturne oak]

Yes.

[quiet current]

https://tenor.com/view/stitch-sad-sad-stitch-gif-14364046974961747120

[pallid ruin]

xd it made it the other way

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[errant totem]

Oh thanks!!

[slender fern]

interstellar halo

[wispy nest]

Something like that

[deft bison]

the hypocrites who put themselves at the front of the stage...

[shut sinew]

When its done

[jovial bobcat]

hi.

I am new to discord.

[hearty olive]

Link to this?

Thanks btw

[cyan shoal]

https://www.reddit.com/r/StableDiffusion/comments/1b91fly/comment/ktxuamu/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

[hearty olive]

Thank you very much!

[cyan shoal]

You're welcome

im lucky I spotted this even

[noble zephyr]

ooooh!!

[deft bison]

enchanted garden ...

[nocturne oak]

Waht's with the creature in that third pic? lol

[half prairie]

so I've managed to reproduce a similar enough artstyle between different character LoRAs

but it's pretty consistent that trying to stack any more LoRAs on top of this ends up in significantly less cohesive images, regardless of how much I change the strength

what do people do to maintain consistency like this while also adding additional features?

(like for example I wanted more revealing clothing so I tried a lora for lingerie, but it was both ineffective and screwed up the quality of the whole image)

I'm also sure I'd have to use more LoRAs to actually get remotely interesting poses whereas this is the most basic portrait I can get just for testing

[main dew]

is he implying the weights will drop when the discord finetuning starts? im confused

or just a list of things to happen

[cyan shoal]

probably in order

first access to discord, then the API thing then he will drop the model weights

that's what I assume

[amber arrow]

Efficient Nodes is drastically changing the outcome of my generations, even though every value of every config is the same as generating with vanilla ConfyUI nodes

[spring mist]

you have one lora at strength_clip 0, hit advanced on LoRA stacker to see what thats set at, im not saying you are wrong just wondering if you are at apples to apples

[nocturne oak]

Also, as a side note, the positive prompt has "no makeup". This isn't going to work the way you intend. It's not going to magically associate the word "no" with only "makeup", and because the token interpreted from the word "makeup" exists in positive, it's likely to have the person wearing makeup in the final image...as it does in your images. If you don't want makeup on the person, just put the word "makeup" in the negative.

[amber arrow]

thanks for the tip! but unfortunately nothing really changed 🥲

[amber arrow]

oh, that makes sense! thanks for pointing that out

[nocturne oak]

Change your CLIP_STR_1 to 1.00 and see what happens on the workflow

Basically...set both clip and model to 1.00

[amber arrow]

I'll try it! But for me, no matter if its set to 1 or 0, clip strength never made any difference 👀

[nocturne oak]

Well, the other thing is that you might want to see if the weight_interpretation is closer at a1111

[amber arrow]

I tried all weight interpretations and they changed almost nothing on the final image :/

[nocturne oak]

Have you attempted to try a different clip_skip value to see if it gets closer?

Without seeing under the hood of the efficient nodes to see how they're wired together internally with the code, I can't tell you why exactly it's different...could be an order of operations thing somewhere that we can't resolve. Typically, I now stay away from the efficient nodes just because invariably I'll want to muck with things anyways. But ymmv.

[amber arrow]

Yep, after noticing this disparity I tend to prefer the results that Vanilla comfyui generates

I was just messing around trying to make things ~efficient, but noticed that do make noticeable changes on the final result

[nocturne oak]

If you want to make the nodes more efficient to work with, don't use those...just use the node grouping function. Select multiple nodes, right click, add to group.

Once they're grouped, right click on them and manage the group node. You can make a bunch of adjustments like what order they're in, what widgets show, what inputs & outpus show, and what names they have.

[deft bison]

@nocturne oak look at the fountain suspended in the air and held by invisible creatures...

[nocturne oak]

lol...definitely some odd shit going on in those pics.

[deft bison]

@nocturne oak

[arctic laurel]

Jared Leto is: t h e b a t, m a n

[grizzled sage]

his "i'm batman" would be whiney sounding

[arctic laurel]

“…HEY…I’m…Bat-Man.”

[grizzled sage]

ai is like, madame bat. its base model died while studying bats in the congo

farleyverse required more exploring

[wispy nest]

Hoping someone may be able to give me some advice on this. I started with text to image, then moved it over to inpainting, wanted to go for a jekyll and hyde type of thing. Zoomed way in and did a pretty precise mask on one side of the face and put my prompt in. Got it pretty close to what i wanted and moved that over to inpaint. been trying to mask and inpaint the mouth and not having much luck haha will paste in the images and generation info

using forge, albedobase, and kohya. nothing else. this is the original text to image

this is as close as i could get it to what i wanted, and i havent had luck inpainting after: Steps: 110, Sampler: DPM++ 2M Karras, CFG scale: 4, Seed: 2997833310, Size: 1024x768, Model hash: 1718b5bb2d, Model: albedobaseXL_v21, Denoising strength: 0.8, Mask blur: 35, Masked content: fill,

[sturdy fulcrum]

Girl with long blonde hair, big brown eyes, blue sky background, holding a cat in her arms, Japanese Anime Style, High Resolution, Best Quality, 8K

[grizzled sage]

whats your goal? consistency? i tried a shot. turned on soft inpainting and used controlnet depth.

[wispy nest]

you used the parameters from this png on the second image? im aiming to keep the style of the original image, just withought the sharp contrast between what was and wasnt masked. I havent messed with soft inpaint much, so ill give your parameters a shot with my prompt. You did fix the main thing i didnt like, the transition at the mouth

[grizzled sage]

you'll want soft inpainting on. i just threw the defaults on

[wispy nest]

im thinking 110 steps MIGHT have been a little bit over kill too haha

[grizzled sage]

maybe haha. i get the experimentation though

[wispy nest]

will try that with and without controlnet. havent really messed around with control much yet. thanks!

[wispy nest]

hello how do i generate an image?

[nocturne oak]

You don't. #1047610792226340935

[wispy nest]

sorry i guess i joined the wrong server

i'm looking how to generate AI images

i'm all new to this

[nocturne oak]

Start with #1080946152318443610.

Look up guides for Automatic1111.

Read about SD 1.5 and SDXL.

[wispy nest]

Thank you! i'll look further

[nocturne oak]

Then eventually, you can make this kind of thing 😉 :

[wispy nest]

no one gonna tell him about 1 click huh? XD

accidentally just generated a batch of two images at 4096 x 3072. bout 15 minutes eta... i am kind of impressed.

[Memory Management] Estimated Remaining GPU Memory (MB) = 15875.858192443848
Moving model(s) has taken 0.76 seconds

[deft bison]

an active souk

[deft bison]

children who use smartphones excessively and without intelligent parental control...

[quiet current]

nope..... this is still not a bot channel

bot channels are still down

cf #1047610792226340935

[arctic laurel]

just a fooocus img2img

[dull flame]

@slow patio

it tries hard, but just cant

[gray folio]

redraw the picture

[slow patio]

Is that any better than what you were getting?

[dull flame]

turns out it seems like its something wrong withthe workflow and not so much prompting

[manic zealot]

yo what do i download for stable cascade?

[noble finch]

Stable Diffusion is not so Stable now, lmao

[quiet current]

an average day for chome users

[golden pebble]

Hey guys, any advice how i can get this with better quality while prompting?

[smoky oak]

@glossy herald I saw this on the ideogram feed, and I instantly thought of you lmao

[knotty orbit]

In 2024, many Chinese families are playing hand in hand, with parents and boys and girls beaming with smiles. Against the backdrop of spring flowers blooming in Tiananmen Square in northern China, boys and girls are the main subject of the picture, with panoramic shots featuring rich and vibrant colors

[wispy nest]

turning my images into videos

turning images into video test

[late rune]

MONKEY FEEDING A PIG AND A DOG

MONKEY FEEDING A PIG AND A DOG

https://tenor.com/n46m.gif

[nimble mason]

Here is the image you requested.

[wispy condor]

你们用过pipeline这个网站吗 AIGC+社交功能的产品 可以在网站里生图和聊天 类似AI版的discord

[dull flame]

i would appriciate if someone could help me compare image quality

this is with realesgan_x4_anime:6B (2048x3072) 35sec generation time

this is with ultimatesdupscale (1024x1536) generationtime 85sec

witch one is superior?

[junior sky]

ultimatesdupscale versions have less upscaling artifacts. I don't like anime/manga in general. But they seem superior in my opinion.

[ebon pelican]

Here is the image you requested.

[robust path]

Thank you

[vapid moat]

Photo real fox

#🏞｜general-with-images fox

[quiet current]

mfw I'm explaining for the 10th time in a day that this is not a bot channel and that bots are down, cf #1047610792226340935

[junior sky]

Here is the image you requested.

[thorny frost]

Korean male with neckles shape black crystal heart with blood in white robe, after generate edit in photoshop

[radiant grail]

cute grey cats

[junior sky]

Here is the image you requested.

[jolly basin]

Hello, does anyone recognize the model used in these pics please ? thank you ^^

[dull flame]

hmmm, it might be a fiamix

[lucid lion]

#💬｜general-chat blue

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

Here is the image you requested.

[lunar bramble]

logo

Here is the image you requested.

[dawn seal]

Maybe I'm confused. Isn't Stable Cascade supposed to take an image and change everything but the face? I can't seem to copy an image into the UI as the base image. I have no idea what I'm doing wrong.

[nimble mason]

Here is the image you requested.

[main dew]

@arctic laurel the hand holding the sword is malformed and the sword itself has a weird dual blade

the broch in her hair has no form and her eyes have artifacts / look bad

yeah im kinda nitpicking just pointing it out

[dull flame]

what model is that? pretty awesome

has lineage feeling to it

[golden pebble]

does someone know how i can achive this quality? what base model could be used etc?

[main dew]

sd3...

[deft bison]

Hell is full of aggressive giant pythons and scorpions. the damned with bulky bodies scream, cry, thirsty drink boiling liquid. the fire devours everything in its path...

[limpid lichen]

we're all going to hell last i took measure. jesus never freed anyone from the commandments and we're all worshiping mickey and 1000 other idols

[gaunt heron]

haha these are great, especially the puking lava bois

[limpid lichen]

yeah tonites literally a golden idol worship night

love the enderman for best, um, enderman

[deft bison]

@gaunt heron hahaha 🔥

[hazy warren]

I kinda wanna make Jesus

hard to get a realistic image

[timid stream]

Korean male with neckles shape black crystal heart with blood in white robe, after generate edit in photoshop

[noble zephyr]

here's my most consistent pixel art animation so far! the quality took a bit of a hit when i converted it to a gif, but i'm still very happy with the result!

[junior sky]

That's some nice work.

[noble zephyr]

thank you kindly! i spent quite a bit of time on the workflow, so it's nice to get some recognition :)

[junior sky]

I think i saw your progress. You posted earlier versions of this, right?

[noble zephyr]

i did! i'm also going to be sharing my full workflow in https://discord.com/channels/1002292111942635562/1212330984192614440 soon

[junior sky]

Very nice. I have no use case for this at the moment, but i will definitely look into this. But this looks already very usable. If you run a rembg, then you have a pretty cool game asset.

[noble zephyr]

that's my end goal! i'm no good at art, but i would like to make games. so since no one has actually done this using SD, i figured i'd do it myself

[junior sky]

Well, good luck. I know the struggles of not having game assets. What engine are you planning to use?

[noble zephyr]

eiter Godot or Gamemaker, since I'm not really good at coding and both have pretty fleshed out visual scripting tools

[junior sky]

Godot is a great choice in my opinion. Python like code syntax, free and fast.

[split blade]

my attempt

[hazy warren]

Not bad

[split blade]

ty

I didn't think Designer was gonna let me do it

[wispy nest]

nice

[late ridge]

Why I cannot install Layer Diffusion Plugin?

[golden pebble]

what can i do to make it more detailed

[quiet current]

looks like you have too many loras fighting each other. try to change their strength and/or activation period.

[hazy warren]

Gotta find better prompts.
Also set to 1024, and hires fix enabled

[golden pebble]

what means "activitaion period?

[golden pebble]

I'll try, thank you

masterpiece, best quality, 2 girls holding hands, sitting, looking at viwer ADDCOMM

river, cherry blossom ADDBASE

(masterpiece, best quality:1.2) (masterpiece, best quality , lora:db_bulma-10:1.0 solo, 1girl, bulma, turkise hair color, smile, beautiful eyes, looking at viewer, braided ponytail, red hair bow, pink shirt, short sleeves, pink skirt, belt pouch, purple kerchief lora:db_bulma:0.8 ADDCOL

masterpiece, best quality, solo, 1girl, karinuzumaki, smirk, looking at viewer, glasses, purple jacket, long sleeves, shorts, thighhighs lora:naruto_karinuzumaki-11new:0.8

these are my prompts

[quiet current]

you can tell a lora to activate at step 5 and stop at 15 for instance

[golden pebble]

That sounds interessting, how do i do this? With refiner?

[quiet current]

doing that can give SD "lora free steps" at the end giving it some room to have a better global coherency

[golden pebble]

how can I do it?

[quiet current]

we talked about it last time, you can try using composable lora https://github.com/a2569875/stable-diffusion-webui-composable-lora?tab=readme-ov-file I haven't used it in a long time tho. Not sure if it's still functional.

prompt editing might work too, never tried to combine it with lora https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#prompt-editing

there is also that extension (just discovered it) https://github.com/cheald/sd-webui-loractl

[golden pebble]

thank you, I'll try them out

[quiet current]

basically the idea is the same for all three. Only activate lora for xxx of the generation steps.

Simply editing your prompt, removing adding trigger words can go a long way too.

But yes the more Lora you use, the more difficult it will be to get a proper image in one pass.

using img2img and doing it in multiple pass is also a valid approach

[golden pebble]

I'll try, both of these loras were made by the same creator, so i thought they harmony with each other much better

[amber arrow]

true, considering that Jesus wasn't a white guy with blue eyes and long hair

[magic nymph]

Hell is full of aggressive giant pythons and scorpions. the damned with bulky bodies scream, cry, thirsty drink boiling liquid. the fire devours everything in its path...

[junior sky]

Thank you for the info. I will be more careful with my life decisions. If that was supposed to be a prompt: #1047610792226340935

[nimble mason]

Here is the image you requested.

[deft bison]

@limpid lichen are you sure we will all go to hell?

[nimble mason]

Id ask in the comfyui channel

[nimble mason]

whether they come from the same person won't really matter - it's all about the training set and to a lesser extent the base model it was trained on

if you dabble in comfyui workflows at all, you can also do all this via inpainting via chaining advanced ksamplers with latent composite masked nodes

[cyan shoal]

aww it's not CERVEZA CRISTAL!

[quiet current]

If you have a Lora for slit eyes and a lora for a nicolas cage face, assuming the two are weill trained, they interfere with each other. Because nicolas cage doesn t have slit eyes.

they ll try to edit the same part of the images

[static breach]

hellooo

[quiet current]

#1047610792226340935

[golden pebble]

is there a gallery in SD where everything i gernated is there?

[golden pebble]

this is awesome

[jolly light]

hey guys, have you seen the new AI that creates thumbnails just with your video title?

this is crazy, me as an editor im loosing my job 😞

[noble zephyr]

Just curious, how big are y'alls AI folders?

[quiet current]

without the venv and outputs

I might have a hoarding problem

[nimble mason]

@noble zephyr 1.95TB, something like 160k folders and just over a million files... About 60,000 images in the last four months

All SD

Something like 235 checkpoints

[frank frigate]

This kinda reminds me of gojo using hollow purple 😭

[exotic path]

guys why does the lora on the bottom wont load on my webui (while everything else loads fine)?

[grizzled sage]

is it for sd15 and you're using sdxl? or visavi

[exotic path]

ill try checking

[exotic path]

how to check which sd version ur using?

for the lora that didnt work its using sdxl as base model

this is what it says tho

[grizzled sage]

thats the ui version

the lora is built for 1.5 if it's not working on sdxl

loras are only compatible with their trained architecture

models have various architectures. the two major ones on civit are sdxl and sd15. those are seperate from the software version you're using.

[exotic path]

so the lora will only show if i use a compatible model?

im using anylora which is built on sd 1.5 and the guweiz lora is built on sdxl, i guess that explains it

[grizzled sage]

yup you got it. not entirely an intuitive thing for beginners admitingly. the confusion is understandable

[gentle geode]

Hey guys
I have a question.
So my prompt for this 'stable-diffusion-v1-6' image generation was: "Hello:
in return, I got this very creepy image, that I send below

is there a way to prevent getting CREEPY results?

[nimble mason]

nope

the solution is to learn to like creepy

[gentle geode]

No way to moderate?

[quiet current]

what were your settings for that ?

[gentle geode]

stable-diffusion-v1-6
cfg_scale: 7,
height: 1024,
width: 1024,
steps: 30,
samples: 1,
Prompt: "Hello"

[quiet current]

sd1.6, try with resolutions around 512x512

Also try with a proper prompt

"hello" doesn't mean much

[gentle geode]

thanks for the tip, as for the prompt I was casually testing my api wrapper code, excpecting to see a person waving their hand if my code worked, got a half heart attack instead 😂

[quiet current]

I mean surely there's a person in there... if you look closely... not that I would recommend doing that.

[junior sky]

Whats SD1.6 did i missed something?

[gaunt heron]

someone's confused

[fringe kernel]

Hey I hope this is the right channel for this, but I'm looking for something a bit specific I'd truly love if anyone could help. <3
Been following this guy that makes custom pokémon for a while new and he uses some kind of AI to create his images (mixed with a bit of elbow grease to polish the images off), but I truly don't know what AI image generation he is using, or what to do to get similar results.(I attached some of the things he has made like this, I'd truly love if anyone knew what or how I could get similar results possibly ^^) Sorry if it is the wrong place to ask this I am not sure.

[quiet current]

https://platform.stability.ai/docs/release-notes#stable-image-v1-release

This model is designed to be a higher quality, more cost-effective alternative to stable-diffusion-v1-5 and is ideal for users who are looking to replace it in their workflows. stable-diffusion-v1-6 has been optimized to provide higher quality 512px generations when compared to stable-diffusion-v1-5.

[junior sky]

You sure this model is available for the public? Can't find any downloads. And even if it is, i must be very odd to use SD 1.6.

[nimble mason]

yeah i remember hearing about it and not seeing anything about a public release of the weights

[quiet current]

weights are not public to my knowledge, it's only available through the API / dreamstudio

[nimble mason]

that's too bad

[delicate heart]

only real way to know is to ask him, but hey if its AI + elbow grease that could be anything

[quiet current]

@delicate heart
#💬｜general-chat message following up.

[delicate heart]

I definitely get "not enough memory" errors beyond a certain resolution close to 1000

not sure where the limit is but if length+width are both 1000+ it definitely happens for me

[ocean turret]

drawing a rabbit,please.thank you.

[delicate heart]

A1111 amd gpu problemz

[nimble mason]

Here is the image you requested.

[ocean turret]

drawing a sweet rabbit,please.thank you.

drawing a school bus,please.thank you.

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[ocean turret]

drawing a Chinese ink painting，please. Thank you.

[nimble mason]

Here is the image you requested.

[jovial tiger]

cascade softens things out

[ocean turret]

@jovial tiger Thank you.

Drawing a blue picture about information technology.

[jovial tiger]

information technology rabbit

[ocean turret]

Drawing a blue picture about information technology and factory.

thank you.

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[jovial tiger]

cascade does well with rabbit

[ocean turret]

Drawing a deep blue color picture about information technology and dragon. Thank you.

[rose wasp]

drawing a policemen using a rugget tablet in the street

[nimble mason]

Here is the image you requested.

[woeful parcel]

Like a live

Convert 3d

[nimble mason]

Here is the image you requested.

[jovial tiger]

Here is the image you requested

[woeful parcel]

Make metal

[jovial tiger]

what kind of metal?

[heady anchor]

this one https://www.youtube.com/watch?v=HDw4OivO45c

[nimble mason]

Here is the image you requested.

[jovial tiger]

white paper man has many talents

fire you say?

huh, just figured something out about comfy. you can copy an image from another app, then past onto comfy and it'll auto bring it up in a new load image node.

[ocean turret]

Use deep blue color drawing a tesla car,thank you.

Draw a Tesla car in dark blue, the background of the picture is an electronics factory, thank you。

[nimble mason]

Here is the image you requested.

[nocturne oak]

No no...I got this.

[ocean turret]

@nocturne oak very good！thank you.

[nimble mason]

Here is the image you requested.

[nocturne oak]

The new 6-wheel Tesla is very popular.

[ocean turret]

But this car more two tyres.

[nimble mason]

Here is the image you requested.

[wispy nest]

Does this utilize sd3?

[nimble mason]

Beep boop. Requests are being granted during this early access trial period.

[wispy nest]

Draw an image of a man behind bars that is on a video chat call with their lawyer

[nimble mason]

Here is the image you requested.

[unique flame]

Create a high detail photorealistic intricate octane render image of an owl with skin the texture of pineapple.

👀

[nimble mason]

Here is the image you requested.

[ocean turret]

Drawing Chinese rural new year.thank you.

[nimble mason]

Here is the image you requested.

[hazy warren]

hey, this one is cool

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

Here is the image you requested.

[nimble mason]

Here is the image you requested.

Here is the image you requested.

[junior sky]

Ah, the @nimble mason image gallery. Always nice to guess what the initial prompt was.

[icy orchid]

Let's play a game, can anyone guess what kind of architecture of a model was this image generated with? (SDXL, 1.5, Cascade, 2.1...)

[sour gale]

generate FB Logo

[timid dock]

Anyone has a clue what kind of model is this or where it was generated ?

[lilac pollen]

帮我生成一个 现代化中式美女在打扫卫生的真人图片

[pallid ruin]

在这里，我尝试生成图像，但我使用了翻译器哈哈

我不知道翻译是否正确 😁

[nimble mason]

Here is the image you requested.

[pallid ruin]

WOW WHAT? this got accidentally generated by the wrong prompt "\192.168.1.10\Server"

[junior sky]

Please take a look at this #1047610792226340935

[inland summit]

Someone can just delete the square base of my statue generated with Stable Diffusion? I can send the file.

[quiet current]

Nope, bots are still down, please read #1047610792226340935

[north haven]

can somebody help me out with this masking issue with invoke-ai? the actual problem is, no matter how precise i paint the mask around this guys face, the complete right side of the face gets distorted a bit, completely ruining the facial expression, i also tried to use an image without the background, the results werent any better. i even outlined the guys body with a black 2px outline, but even then at least the eye/brow section gets distorted

i should provide an example image, thats what gets generated all the time

is this even a supportchannel or should i move it to tech-support?

[quiet current]

#📝｜prompting-help is probably the best place for those questions

[north haven]

thx

[cyan shoal]

Got my tiny prompt list ready for SD3

[fringe slate]

https://youtu.be/Vc2hvp3P6ZA?feature=shared

[grizzled sage]

https://tenor.com/view/pablo-escobar-alone-gif-18837103

[main dew]

been almost ready for a week

and the preview

[wild sorrel]

let em cook

[viscid glacier]

@grizzled sage system isnt too shaby, should do for a couple of years till i buy a 96 core threadripper

[grizzled sage]

the 1%ers

[viscid glacier]

haha, yeah i know, who'd have thunnk it

i didnt think it was all that special till i ran the same benchmark on my i7, 16GB laptop and got a score of about 2200

[grizzled sage]

@versed cobalt cyberpunk isaac. instant-id. ezpz

[versed cobalt]

@grizzled sage that is what i need! but i am unsure of how to properly download it and where to get the Download. I posted in general chat this question of mine -> ( I see there is ip-adapter instant id sdx ) and (control instant id sdxl on civitai) - {not sure if this is the one i need or not} but if so no tutorial on the process to Downlaod it properly. I really want to get consistent faces sorted and hopnig this will make my current process a bit more effective..ANY help would be great

do you care to be that guy that helps me get on my feet with ip-adapter 🥺

lol

(for fooocus) that is

[nimble mason]

Here is the image you requested.

[wispy nest]

meh

I wrote this cgi promt almost 1 year ago and still going bananas

[carmine quail]

https://tenor.com/view/its-happening-gif-23353691

[pallid ruin]

so cute 😊

[wispy nest]

idea for this was based on kena bridge of spirts with a ai robot friend I usualy make anime a photorealistic works but the cgi is nostalgic and helps me learn more engenering with promts and what words to put first is a bit more complex to do cgi

[wispy nest]

i have come to the conclusion by adding with friend with various modifications you can get a second charachter that isn't a clone to achieve this you need to have various clone debugging negitve promt key words to reach most potential

the question is how Many can you get in one frame not being a clone

[grizzled sage]

prompting is king

[jovial tiger]

adding a friend is always good

[wispy nest]

Yes I would agree

[wispy nest]

I'm thankful you have shared this image the hue of blue here is amazing contrast to this image works perfectly

Wonderful

[jovial tiger]

A black panther leaping, in the style of John Howe ADDCOL
A blue elf with white flowing hair, in the style of John Howe ADDCOL
Yellow eyes and a blue scarf , in the style of John Howe ADDROW

A black panther leaping, in the style of John Howe ADDCOL
Blue armor and a wielded sword, in the style of John Howe ADDCOL
A stone valley, in the style of John Howe

here's the prompt for it, just need to use that with the regional prompter extention if you want to play around with it

[wispy nest]

I will try it but just for fun ok

Close up studio shot on a cgi boy cute with ai robot friend :;"cgi hd 3D cartoon gfx:70 render :%100 "composition cinematography pixar studios design 3D render ultra high quality 8k resolution concept art kena bridge of spirits gameplay ps5 forest futuristic scifi quanta quest 3d deep dream simulation vr video game digital new world ready player 1 type of nostalgia core jolt shock blue light energy ambient supervillain cityscape rare videogame kingdom hearts style astetic matrix hyper glitch no escape enchanted wisdom clean realistic images

[wispy nest]

That's really cool backround well done

Done

I'm cooking sum now

[jovial tiger]

different model

[wispy nest]

lovley color

[wispy nest]

Like how these came out

I have a hand promt too

[wispy nest]

so much detail

your sign in bottom left?

or stamp

[jovial tiger]

nah that's just from sdxl

[wispy nest]

1.5 anime got so many stamps

was bad for me

[grizzled sage]

thats a beautiful jaguar

[jovial tiger]

even better one coming. i just put it through the claude 3 regional prompter writer.

[wispy nest]

woww

gn

[hazy warren]

@simple frigate couldn't get him riding a T-Rex

[simple frigate]

After 8 images I got the t-rex-mobile

[wispy nest]

lol

[hazy warren]

😮

[wispy nest]

🤭 thought i'd try a t-rex mobile. am bored

[last talon]

A black panther leaping, in the style of John Howe ADDCOL

[wispy nest]

🤔

[simple frigate]

obligatory question:

🤔model???

oh wait I'm dumb

SD3

prob

[simple frigate]

at least confirm or deny 😭

[hazy warren]

looks like a footy logo

[wispy nest]

lol

[simple frigate]

josh, literally what model are u using

[wispy nest]

👀 bing

[simple frigate]

explains the prompt understanding-

[wispy nest]

🤤

[hazy warren]

you hungry?

[wispy nest]

a lil bit

[hazy warren]

have this eat a burger'

[wispy nest]

💀

the lil teef

[wet pilot]

https://cdn.discordapp.com/ephemeral-attachments/1092492867185950852/1217344448032280646/IMG_0568.jpg?ex=6603af5d&is=65f13a5d&hm=51f504bbf84ffdb0d23449a4602bd684f4885f88dc43690f0c54a7744e5ad2e5&

[wispy nest]

@hazy warren lol

[hazy warren]

XD

[hazy warren]

Zelda!

[gilded birch]

Within the palace walls, a young infant sat on soft silk cushions, enjoying a sweet treat. Plump little fingers gripped a colorful lollipop as tiny rosebud lips suckled happily at the sugary candy.

[frosty sky]

young Chinese girl in glasses typing on a computer in 3d digital illustration 2, in the style of quirky character designs, warm color palette, unicorncore, vray, study, sharp & vivid colors, studyblr --ar 4:5

[junior sky]

@shut sinew Sorry to ping you, but someone here is posting gif-walls. (solved)

[stoic shale]

a black child

a black child is swimming

[noble zephyr]

did you generate this? it looks awesome! i'm going to try to get something similar, with my own twist of course ;)

[quiet current]

very unlikely

[noble zephyr]

yeah, that's why i asked. it almost looks too good to be generated

[quiet current]

it got the typical jpg artifacts

yup it's not AI stuff https://www.dadasou.com/ny/qijie/23964.html

it's just another lost soul trying to summon the dead bots

[noble zephyr]

i was talking about the above image, not the one of the child

[quiet current]

ooooh

maybe then yes

[noble zephyr]

i mean, the resolution checks out. reverse image search doesn't show anything either. PNG info doesn't recognise it tho, but it could be made in comfy

[quiet current]

Looking at stuff like the blurry/bubbly texture at the top of the head + the horn being cut for seemingly no reasons + etc... I'm enclined to say it's ai stuff

[nimble mason]

Here is the image you requested.

[celest wraith]

Кактус

black man in white uniform

[nimble mason]

Here is the image you requested.

[bronze mantle]

superman

[noble zephyr]

can you guess what my inspiration was?

[wispy nest]

Spiderman

[noble zephyr]

not quite

or if you wanted an image of spiderman: check #1047610792226340935

[crisp stream]

I don´t know 😄 Btw thank you

[noble zephyr]

it's hollow knight! i love that game

[crisp stream]

and yes, generated here with the pure base model

locally

[noble zephyr]

dang, color me impressed. it looks really good!

[crisp stream]

Thank you, glad you like it 🙂

[crisp stream]

❤️

[celest wraith]

Нарисуй пингвина

[crisp stream]

if it wasn´t clear by now, yes it is 🙂

[reef kelp]

What is the name of this blue tank ? hm

[floral flicker]

Giving a quick plug for a relatively simple fork I did of the --sd-webui-ar-plus extension.
https://github.com/altoiddealer/--sd-webui-ar-plusplus

[clever oar]

predator close

https://wh40k.lexicanum.com/mediawiki/images/thumb/2/26/Gladiator_Reaper.jpg/150px-Gladiator_Reaper.jpg

dont see in game

[crisp stream]

@noble zephyr

[noble zephyr]

looks dope!

[crisp stream]

Thank you, how about a successor of your´s? 😄

[noble zephyr]

sounds like a good idea! i'll get to work right away

[noble junco]

https://cdn.discordapp.com/attachments/1194704493418389504/1213082098059972608/H53pW5SldAAAAAElFTkSuQmCC.png?ex=65fd683e&is=65eaf33e&hm=67956a9e9b163391643930be9d6c83aaf129bb771525ee4350d0678ae223b526&

[simple frigate]

do y'all think it's possible to get this specific style in current SD, maybe even fine-tuned SD3. Because SD struggles with being detailed and creative outside of the focal point (the character)

[round quiver]

Imagem de uma noite estralada de um casa aconchegante e iluminada por dentro

[noble zephyr]

@crisp stream i feel like i could've done better, but i ran out of patience lmao

[crisp stream]

Hihi, funny anyway 😄

[wind egret]

my image to image works fine, but for some reason same prompt under mov2mov, keeps generating pin ups

this is what img2img is outputting with the same settings

[robust path]

Me reaching out to SD3, but there's only silence

[crisp stream]

yet a heart 🙂

[robust path]

😊

[crisp stream]

a little dark, yet a heart as well... somewhat 😄

[shell marlin]

how to use this bot, how to generate images??

[jovial tiger]

it's down

[calm trout]

hi, why is the background on my printouts so bad

[calm canopy]

whats your prompt?

[wispy nest]

AAH!

[calm trout]

its comfyui workflow

may i share it?

[wispy nest]

oh no bing thats just wrong, look at the guy... wtf is that data from? xD

[grizzled sage]

its based on your prompt.

dall-e surely has billions of parameters, so it's going to be your prompt that determines the output primarily

[cold meadow]

Looks good. I imagine a test SD3 prompt could look like that.

[jovial tiger]

same prompt put through a regional prompter thing.

[cold meadow]

That looks very 1.5 / Nai-based.

[grizzled sage]

wonder why @wispy nest won't share their prompt

[jovial tiger]

putting it through LLaVa image to text, Animated, comical public transit scene featuring a white van in motion down a city street with buildings lining the backdrop, and a clear blue sky overhead. The family of diverse individuals on top are animatedly expressive, while the open-hatch back holds more joyful characters. Four seagulls fly around, adding movement, as vibrant colors and energetic lines depict action and emotion through visual storytelling.

[grizzled sage]

yeah you can interrogate it, but dnk said it was so wrong and i suspect he had some sneaky prompting to describe a specific situation

[lost valve]

gonna have to do one for college yearbooks and do a side-by-side x)

[robust path]

Very shady, he got here just today and now he's showing this altright nazi stuff?
Block & report

[grizzled sage]

can't say for certain that there's anything qanon'ish about it. but they sure are being shady about sharing

i don't believe reports matter anymore. this shit is rampant on discord. my reports probably just go straight to the bin

[robust path]

I'm sure it is, why would he as a newby post it over here? It.s even Bing, not SD

[grizzled sage]

all i can do is try to get them to not be cowards and say what they mean.

[heady anchor]

plenty of ppl post dalle stuff here,nothin to do with nazis

thats a massive reach

[robust path]

Mods can make up their mond, I don't need to see this stuff

[heady anchor]

what stuff

[robust path]

Racist humor

[heady anchor]

ok where is it

[robust path]

I blocked him

[grizzled sage]

dog whistle's entire puprose is this. "POINT AT THE RACISM THEN!?" as if that's a giant defence

plausible deniability

[jovial tiger]

wait what? he posts a picture of a speeding bus and now he's a nazi? I don't understand what's happening.

[heady anchor]

yea where is the racist stuff

[grizzled sage]

88 is a pretty awesome number. two infinity signs. doesn't seem harmless at all does it?

until you realize it's code for HH

[robust path]

that's actually a good sign

[jovial tiger]

what's 88?

[grizzled sage]

white nationalism motto

[jovial tiger]

was there a picture with that posted/

?

[heady anchor]

but where is the 88,he just posted some pic of a bus with seaguls

[sly tangle]

Is the bus with people in it supposed to be racist imagery? I'm so lost

[heady anchor]

idk what that guy is smokin

[robust path]

Just look up 'stone toss'

[grizzled sage]

gaslighting people pointing at dog whistles. that's another

[robust path]

and i'm not a guy

[grizzled sage]

why do i even bother explaining dog whistles to people who obviously know whats going on

[robust path]

no more political discussion from me, I'm done this is sd images

[heady anchor]

dont see anything that says "stone toss" from dnk

[robust path]

Borrowed a mj prompt

[jovial tiger]

apparently DNK is just a bot that's being controlled by an evil puppet master.

[robust path]

That looks a lot like sd3

[heady anchor]

yea he touched my nono spot

[grizzled sage]

cascade is next level. could be that

[jovial tiger]

it's sdxl with regional prompter.

[grizzled sage]

thats next level too

[robust path]

I wish we had regional prompting in Draw Things

[grizzled sage]

just SaaS things

This week i enjoyed watching twitter MJ fans obsess over how game changing face swapping ip-adapters are. Meanwhile, open weight people have been playing with them for months

[jovial tiger]

yeah well, midjourney is really easy to use, so there's definitely a value to that.

ip adapter and all that has a lot of settings going on

[grizzled sage]

fooocus

MJ getting face swaps is them catching up, but the twitter hype presents it as if they're first to the goal

like a fat guy finishing the marathon

[robust path]

They just don't know any better

[grizzled sage]

forgive them father for they do not know

[robust path]

😂

[grizzled sage]

RunDiffusion are claiming they'll have a better model than sd3 out soon

[jovial tiger]

don't let anyone say that SDXL does text.

[grizzled sage]

one thing midjourney gets right. it sure is mid

heuheueheueheuheuh

[arctic laurel]

that’s some DREDD (2012) shit

[jovial tiger]

RuneFlusion. that's the company I'm for.

[grizzled sage]

friggin fast!

[jovial tiger]

a giant sign with "RUNDIFFUSION" written on it, in the style of cyberpunk ADDROW
a tall building , in the style of cyberpunk ADDROW
crowds of people with hands over their head cheering, in the style of cyberpunk

[grizzled sage]

https://twitter.com/DataPlusEngine/status/1768026143494377773

DataVoid e/acc (@DataPlusEngine) on X

who thought i would let #SD3 out do me 😂 ? The beast is coming!

Twitter · Mar 13, 2024, 9:27 PM

[jovial tiger]

regional prompting is really easy. no idea why i didn't do more with it before sd3 was announced.

WWHOAAA. datavoid makes seriously kick butt models like proteus.

[arctic laurel]

the ruffalo is real

[grizzled sage]

he works at rundiffusion these days so i think he's part of the juggernaut team

[grizzled sage]

bixby for life!

[jovial tiger]

wow

that explains a lot

[grizzled sage]

https://i.makeagif.com/media/5-06-2015/1K6RzU.gif can you hear it?

[wispy nest]

https://tenor.com/view/screaming-bird-alarm-clock-wake-up-gif-11858070275565777857

[jovial tiger]

ok sdxl REALLY can't do text.

[grizzled sage]

there are loras. harrlogos. its not bad at all. but yeah.. its bad too

[jovial tiger]

new york city street ADDROW
giant sign that has "DATAVOID" writte on it, held by hulk hands, photorealistic ADDROW
hulk arms holding a sign, photorealistic ADDROW
incredible hulk, huge muscles, clenched fists, angry face, photorealistic

that's interesting.. i wonder if I could do a lora for just one section.

[grizzled sage]

regional prompting extension does have the latent mode that allows loras in different regions

slower but works good

figured i should post an image if i'm chattering so much

[jovial tiger]

wow very cool, it worked. obviously the spelling is still off, but it made a big logo out of it

[grizzled sage]

NICE. yeah harrlogos does get spelling wrong 9/10 times but it does text better still

https://www.dailymotion.com/video/x2klp22 for those who couldn't hear it

[jovial tiger]

you may not like it, but this is peak hulk right here

[grizzled sage]

stupid sexy fistulk

[grizzled sage]

hidilydoodildy

[wispy nest]

😭

[grizzled sage]

trump isn't god emporer. dumb white nationalist bullshit.

[wispy nest]

💀

[jovial tiger]

I do tons of Trump stuff. Because it's funny. That's all there is to it.

[grizzled sage]

Trump is funny af. Elevating him to Leto status though? eeehhhh. nah.

more dog whistles. YOu've got to have read dune to know what it's getting at

[arctic laurel]

I don’t think folks realize how naked their intrinsic biases are when they create their AI art—so much of it is politically-oriented. It speaks volumes to the content of the subconscious mind. It speaks volumes about their day-to-day.

[jovial tiger]

And it's beautiful.

[carmine quail]

For the peasants in gen chat without images:

[jovial tiger]

When they mentioned the text model alone was 10 gigs.... awesome to hear.

[cyan shoal]

the text model itself requires 10GB of Vram?

[carmine quail]

I highly doubt it. Text models tend to be more compressed and it's only 5B params

[cyan shoal]

• int8 right now or even higher quantizations might come

[carmine quail]

Actually, I might be wrong. Mixtral 7x8B is 43B params and takes up 86GB, so maybe it will take 10 GB vram

[cyan shoal]

hmm maybe I could prove this VRAM usage by running Pixart on comfyui

[carmine quail]

But iirc T5 offers minimal improvements so it might not even be necessary

[cyan shoal]

idk if it's not a huge time loss I'll keep T5 just in case

not because of text, but because of the huge 512 context limit

[carmine quail]

Yeah, if it works it's totally worth using

[cyan shoal]

otherwise it's back to 77 limit

[carmine quail]

https://images.squarespace-cdn.com/content/v1/6213c340453c3f502425776e/19acc961-3e42-413f-b495-450803aba582/baseline_comp.jpg?format=2500w

[cyan shoal]

yeah I saw those

[jovial tiger]

The stability staff said the text model is 10 gigs. Not billion parameters. But then he said one could go down to 4bit quant, so that brings it down to around 1-2.

With a 4090 one could do the whole thing though

[cyan shoal]

yeah

well this is very cool you guys

8B offline with T5 is possible

[carmine quail]

T5 seems to help mostly if not only with typography, meaning text in the images, so it's not even necessary most of the time

[cyan shoal]

yes I read about it but I would like to use it just in case it would hinder prompt adherence (yes I read that it "techinically" doesn't affect it that much) and the 512 context limit

[carmine quail]

A bit of improvement in prompt following as well

[carmine quail]

Agreed. If I can use it I will use it

[jovial tiger]

Yeah the whole point is that kind of thing. Prompt adherence is what separates stuff like ideogram and Dall-e and everyone else

[cyan shoal]

the ferret jar example showed the ferret being in the jar all of the time, whilst it was only 1/4 without T5

[carmine quail]

The people who have to use the 800M param model will probably not be able to use T5 though

[cyan shoal]

the 2B will be an interesting story

[jovial tiger]

If I was still at my computer, I'd make a picture of a man unlinking the caboose that's on fire with a sd 1.5 users sign on it.

[cyan shoal]

T5 at 4-bit + 2B model at fp16 could be the king for low end

I don't know how good the 800M model would perform

considering its smaller than 1.5, even if the architecture it just that much better than Unet

[carmine quail]

I like sd1.5 because it has a lot of cool toys that work with it, but the poor quality images (ie hands) make it a hard sell over sdxl

[jovial tiger]

Where that'll be useful is embedded devices like on a phone or iPad

[cyan shoal]

SD1.5 is at its most final-ish form, with years of improvements built upon it, so its not always a fair battle

SD3 might tip the balance with it launching with controlnets and such

and of course, prompt adherence 🤑

[carmine quail]

Hopefully the controlnets are better than sdxl controlnets

[cyan shoal]

Yessir

[carmine quail]

If they take another month or two to make it work with controlnets I'd be happy to wait a little longer

Though I imagine my opinion isn't shared by many lol

[cyan shoal]

I wonder if 16GB System RAM will be enough for unloading and loading T5 and other models (at int8 or fp16)

[jovial tiger]

Probably will be fine

[cyan shoal]

I have 32GB so I don't have to worry about this

[carmine quail]

I think it'd be a good idea to upgrade to 32GB RAM if you want to use fancier nodes in comfyui or A1111

[cyan shoal]

but still I just want more people to experience SD3 offline

[jovial tiger]

Yeah, and if you set your nvidia settings to share system ram, it'll get slow if it goes over but it'll work

[wispy nest]

😍😍😍😍

beyoncé

[grizzled sage]

yup

[grizzled sage]

i keep hearing that controlnet's on sdxl are bad. open pose is one that i struggle wiht only sometimes. the rest are fine. i dont get it

depth is primarily what i use and the official stability control loras are what i use there

[grizzled sage]

i'm going to bet that the t5 will only link up to the 8billi model

[carmine quail]

That's certainly possible. I have no idea how it'll work

[hazy warren]

monster hunter

[hazy warren]

I wonder if you can make a realistic photo of Rosetta from mario

[tight silo]

GPT5???

Source: https://www.microsoft.com/en-us/microsoft-copilot

[wispy nest]

Can it make videos?

[tight silo]

Its not formally announced. It looks like a leak or a teaser

[jovial tiger]

gpt 4.5 is coming out in july

[tight silo]

U an insider?

[wispy nest]

How would I promt a visual image of sync like a jolt shock but diffreng

A photo of a human hand Hyperrealism multi angle 3d render 5 finger:;;:-- 5 fingers only realistic skin completion masters of art and imaging digital image wonder simulation real; gfx 90 smooth render complete fine art color

My second public domain hand promt

[hazy warren]

u buying?

[tight silo]

No, i want it in chatgpt

[hazy warren]

I kinda wanna make a hand now.

also a realistic Rosetta

[wispy nest]

Rosetta stone?

Lillith is quite wild too

[hazy warren]

the mario charater

[wispy nest]

The hand promt needs work that's a diffreng version but I told the original version to emad or who ever the guy who made stability is on a voice chat a year ago and his team basically said it's possible with a five finger text promoting but they don't know how

[wispy nest]

Oh warrio is funny too

[hazy warren]

I think the hand might need more shadows, also still kinda looks CG

[wispy nest]

I've tried human anatomy and versus medical books for reference but it provides more of a NSFW image like weird forms or muscle with no skin

[hazy warren]

yeah, i've noticed that SD randomly gives me NSFW sometimes, and im whattt, where did that come from.

[wispy nest]

Term feeding the neral network

Movies

Ext

Idk

[hazy warren]

I can see how it could happen, like the Seed is a random number, and it could hit and X formula to generate it

[wispy nest]

Type in random numbers for promts too it usually gives you a decent images 9 out of ten times wich is random could be something with seeding

[hazy warren]

dude, SD3 is taking so long to release, im totally going to forget about it when it does release XD

[wispy nest]

Hmm I wonder if it's too late to join the beta

List

[hazy warren]

I think I put in for it 2 weeks ago

[wispy nest]

Mad heads on that list

There users are growing

[hazy warren]

they probs only putting the popular youtubers on it, I bet when I get it, it will be like only a week before the release

[wispy nest]

Enuf time to get the good promoting out before the mass does

Have you seen the shock gore horror ai stuff 🤔

How does this fly on ig

I wanna do more weird stuff but idk man

[hazy warren]

😮

[wispy nest]

One of my friends always be sending me like random shock ai ig post with thousand sof likes

[hazy warren]

😮

[wispy nest]

What ps2 games could I use as a decent modification

For promt

digital nightmare scifi green planet solar screams of the next one robot horror vision pro VR silent night in the Galaxy imagine dream see explore futura 3d explore to fear no more my mind I will explore I'm high and high more shroom shroom room grainy 2008 green color tech film 1965 35mm

[jovial tiger]

It was on one of their press releases

[tight silo]

Really.. i hope so

I was hoping itd be this month

Cause Claude Opus is beating then rn

[eternal dome]

hi guys , i need some help with my stable diffusion sittings , currently i use --xformers --medvram --disable-nan-chek --no-half arguments , and these are my specifications

Device name DESKTOP-29E7NJC
Processor Intel(R) Core(TM) i5-10400 CPU @ 2.90GHz 2.90 GHz
Installed RAM 16.0 GB
Device ID 0E54BD6A-12FA-476D-A77C-6BC11CA42E4C
Product ID 00331-20350-00000-AA158
System type 64-bit operating system, x64-based processor
Pen and touch No pen or touch input is available for this display

i can make 1080p but it sometime gets weird because of the dionising

is there any better arguments or sittings ?

like this , looks weird but cool

this is with less dionising , but i feel it`s less detailed

[wispy nest]

I dont run local sorry fam

[eternal dome]

np

[wispy nest]

very cute image

[eternal dome]

thanks🥰

[wispy nest]

@vortex maybe can help u

[eternal dome]

Ok

[eternal dome]

@hazy warren

[eternal dome]

wow ! amazing !

[wispy nest]

thank you

simple promt tye dye glitch shroom

[eternal dome]

nice

[hazy warren]

😮

[hazy warren]

looks like a rainbow mushroom

[eternal dome]

i need help

[wispy nest]

taking two complex words that are colors creates a fushion instead of using colors also gen stone names provide a more surreal take

[hazy warren]

OMG! make a ice-cream boom!

[wispy nest]

ok

[eternal dome]

rate it guys , i want to know if it`s horrible or not 😅

[wispy nest]

9 out of 10 my mind wants more left frame

[eternal dome]

thanks !

[wispy nest]

sure yw

[wispy nest]

oh shes cute

[hazy warren]

me on mondays

[eternal dome]

haha

[hazy warren]

me before work

[eternal dome]

cool !

[eternal dome]

waaaah !!

[hazy warren]

I'd buy this

[wispy nest]

I had to change to explosion boom wasn't working also pinging from PC I'm on is not workable ty

[hazy warren]

do you think my MacBook Pro with the M3 Pro will be much slower than my 4080 for making images?

[nimble mason]

yes

[hazy warren]

😦

[nimble mason]

i like the idea of those things though

getting the bus speed up so ram can get kinda competitive with vram

[eternal dome]

😢

[nimble mason]

if they can get that even close... then we'll be able to build 256gb "vram" machines that can design nukes or something lol

[wispy nest]

No arnt those new chips

[nimble mason]

cascade

[hazy warren]

yeah, new chips

[wispy nest]

I'm not a apple user but they have nice chips yes

[wispy nest]

I'm feeling this one

[wispy nest]

Lighting reminds me of this one majestic vibes

[eternal dome]

wooooooooooow !!! legendary

[hazy warren]

I haven't done anime with SD yet.

[wispy nest]

There are ways you can write a anime promt that wil work with any setting if youu do it right with only changing the setting in the prompt streamlining animation in the Same style every time

[hazy warren]

😮

[wispy nest]

Appreciate you

[nimble mason]

Couple checkpoints I can recommend... Bluepencil, deepblue, carosello, quietgoodnight

[eternal dome]

bluepencil is XL right ?

[nimble mason]

Yep

I really really like those ones

Very diff styles too

[eternal dome]

thx, well try them soon

[nimble mason]

I have almost 240 checkpoints >_<

It's a disease lol

[eternal dome]

this is my friend`s

he is so good

i aspire to reach his level

[hazy warren]

😮

[nimble mason]

You can generate that with the right checkpoints and a bit of playing with the prompt

[eternal dome]

with less prompt and more details right ?

[nimble mason]

A specific type of prompt prolly

If you have one example you're generally good to go

Check out one of those checkpoints on civitai and look at example images

They include prompts

[eternal dome]

yeah

[hazy warren]

the question is, are you the artiest, or is the A.I, or is is the people who made the A.I

[wispy nest]

https://tenor.com/view/mind-blown-shocked-ashtonished-astounded-explosion-gif-4927161

lmao

I wouldn't even dare to answer that. It's just dumb fun for me c;

[hazy warren]

ye XD

[grizzled sage]

t5 adapter on cyberrealistic (sd15)

[wispy nest]

😂

[crisp stream]

have you tried n-prompting 3d with those?

[tired chasm]

no

[royal merlin]

一个农民