🏞｜general-with-images

[south temple]

oh god

you get an acid trip

I should turn the CFG down

about a 5 minute render for video stuff

[sour ice]

looks like something I'll see in a fever dream

[south temple]

using that as a prompt

[nimble mason]

freaky lol

[wild sorrel]

AI might be interesting for some horror stuff

[nimble mason]

Oh I've made plenty of that

[nimble mason]

I have a whole collection of raw eyeball food imagesx

Soups, salads, you name it I made it

[south temple]

60 steps!

[crisp stream]

@shut sinew

[south temple]

oh boy am I about to play and ruin things lol

whats a detailer do?

post img?

[nimble mason]

It's not easy to get these types

Especially hard is having spider legs instead of human legs

[south temple]

good to know

[nimble mason]

Loving the new comfyui ipadapter nodes

[south temple]

ComfyUI-IPAnimate

interesting

https://github.com/Chan-0312/ComfyUI-IPAnimate

[wild sorrel]

that's just a food from the future, that's not horor

I meant more of a like weird video stuff, distortion , something that doesn't really happen in real life, but looks creepy and weird

[nimble mason]

Did you see the spoiler ones I sent right after that

[nimble mason]

Give me an example image of what you have in mind

[wild sorrel]

yea, pretty horror-y, but that's just images, I'm thinking about videos, that could be more fun with AI weirdness
and not just scary looking stuff \ gore e.t.c, but more of a like tense atmosphere, weird things happening

[nimble mason]

Yeah video is a hell of a lot harder cuz the computational cost is so insane

[nimble mason]

This is as far as I ever went into animation

[wintry owl]

a pretty girl

a pretty girl , chinese girl

[nimble mason]

Here is the image you requested.

[south temple]

neat

[severe wind]

Wow that’s insane good job

[south temple]

I like that

how does one copy the workflow from a video?

[nova nebula]

Who can generate some photos of Asian handsome guys?

[south temple]

ill drop in your prompt

[south temple]

thats cool

[languid pebble]

Old prompt, new animation 🙂

[nova nebula]

nice

[languid pebble]

Cool one!

[languid pebble]

Thanks 🙂

[south temple]

FINALLLY got this bloody thing workin

[nova nebula]

cool

[south temple]

Well thats neat

Mojo, I know I keep asking… but are you comfy?

That’s some clean animation stuff

[languid pebble]

Stable Diffusion Forge ...

[south temple]

SVD?

Or animate?

[languid pebble]

SVD

[south temple]

Maybe I should play around with SVD a little more

[languid pebble]

80% are fine ...

[south temple]

I sorta wrote it off until I could figure out how to steal the last image rendered an batch render at least 2-4x

You using img or txt to SVD?

[languid pebble]

IMG ... I am more a prompter 🙂

[south temple]

Does it have to be a specific sized img ?

[languid pebble]

1024x576

[south temple]

Curious if you can see if it will pick up on this 🙂

Might need a resize

[languid pebble]

Try ... always better than listening to me 🙂

[south temple]

I’m at work now till 10 so, my playtime is over for a few hours.

The mission I think is going to be SDXL in animate

[languid pebble]

I'll try a small one ...

[south temple]

Take your pick 🙂

[languid pebble]

I used the normal settings for video output I told you ... and no fixing afterwards

[south temple]

Hmmm SVD has loads of camera action

I wonder if it will take a prompt to keep the camera center.

[languid pebble]

I think it depends on the picture ...

[south temple]

It’s almost like SVD was designed to simply add to the frame… and then push like a dolly zoom

[languid pebble]

You know my corridor animations?

[south temple]

Negative

[south temple]

Yeah

Clean AF

[south temple]

Man I want these to bounce, or be longer 🙂

[languid pebble]

Maybe add some mushrooms to make A.I. hallucinate more? 😄

[south temple]

That’s kinda the mission

I like nature visuals

[jovial tiger]

@heady sorrel

So these are 2 raw as you promoted, and 4 through my llm prompt expander

[eternal comet]

We are free ong

I have made a lora training setup

That scrapes data for lots and trains

Lora

[south temple]

Mushroom people, Deep forest, glowing spores, glowing roots, magical

[languid pebble]

Imagine dragons ...

[south temple]

That took a while I bet

[languid pebble]

2 Minute?

[south temple]

Dang

That thing is good at rotations and stuff

[eternal comet]

What gpu

🤔

[languid pebble]

4090

[cyan shoal]

Papers?

[languid pebble]

Papers?

[clever oar]

free dom\

[south temple]

5090ti…. Might have to pull a trigger on that one

[clever oar]

9090 titi

1 second video generation

video cards are developing, but graphics in games are not very strong

[languid pebble]

Will need some more months ...

[eternal comet]

https://tenor.com/view/a-train-edit-a-train-the-boys-a-train-the-boys-edit-gif-26341809

GTA 6:

The textures alone gonna bring even a 4090 to it's knees

[clever oar]

I think this is unrealistic because it will take too much time and it’s money

make graphics like in real life

[languid pebble]

Problem is they have to serve many different plattforms ...

[clever oar]

and plus the games are being adapted for consoles

[languid pebble]

A game only developed for 1 GPU would be way better

[clever oar]

gta 6 nice trailer but how it will be in real

[eternal comet]

Rage engine doesn't disappoint

I wouldn't be surprised if the games engine uses transformers and ai

[languid pebble]

Damn .... can't stop ...

[south temple]

I need to find a 1.5 that has some of the same production as my SDXL 😦

Turns out animate SDXL is not the greatest lol

If anyone has some recommendations for a checkpoint SD1.5 that is nature based. Would appreciate it

[south temple]

Also… this new IPA is stoopid!

Love the 8bit

lol 8-bit murder scene

[south temple]

What’s the story little kld? You lost your pup?

The unified loader is new.

I think

[eternal comet]

cascade

is a ram monster

[south temple]

That a program?

[eternal comet]

the outputs are magnum opus tho

especially the hands

[south temple]

It looks fun

[eternal comet]

big ass ai model

two models

[south temple]

What is a clip-vision?

[eternal comet]

prior

and then

main model

its like having a huge refiner

its two models

it doesnt even fit on a 4090

entirely

you have to offload one into sys ram

then do first model

then do second model

[south temple]

Still talking cascade

[eternal comet]

i didnt use clipo

i just used cascade

[south temple]

I know lol.

Sorry, I got confusing. There’s a new IPAdapter that just released and I was learning it

In comfy

https://youtu.be/_JzDcgKgghY?si=nrkT-XlnMFB85HCe

[eternal comet]

]

[barren axle]

its been stuck at here for 30 mins now.

what should I do?

[south temple]

Whoops

[barren axle]

yea. its says installing requirement. but its been stuck there for a while now.

ive got somewhat of a high end pc. shouldn't take that long

[south temple]

It might be done

I think when its dont is just … sits.

[barren axle]

should in close or restart?

[south temple]

Close

Installing comfy?

[barren axle]

idk much about stable diffusion to know what that is. xd

im installing 1.1.1.1 autmate though

[south temple]

AH!

Different UI

[barren axle]

is comfy better?

[south temple]

Im not gonna be much help installing that.

Comfy is different

Depends on how deep down the rabbit hole you wanna go

Comfy is a node based setup which can be complicated and easy. No matter what…. You are in for an addictive ride

[barren axle]

i clicked enter and something happened.

nvm. we are a go again. xd

[blissful oasis]

using the kohya ss trainer notebook in colab. i used mount drive btw and left everything else as is

[eternal comet]

no

you will fuck enviroment variables

then ud have to do --nocache-dir

[gusty cloak]

I believe I run out of VRAM when using SDXL + Canny 🤔

A1111 estimates that a generation will take 4+ minutes, which is literally not possible for a 1024x1024 photo 😅

These take seconds usually

My GPU must be in pain 😢

[barren axle]

thanksfully. I didn't do anything. it started on its own. xd

[barren axle]

what gpu u got?

[gusty cloak]

4070 Ti, 12 VRAM

[barren axle]

damn... thats not good

[gusty cloak]

Mind you without a ControlNet this issue isn't present

[barren axle]

ur making me worrying. xd

for my gpu

[eternal comet]

what gpu

how

is that

pain

im using a 3060

💀

im suprised with its performance tho

it can run 1024 x 1024 easily

unless u havelaptop gpu ur config prob wrong

[nocturne oak]

Cascade with SDXL refinement stage and the better of the 2 outputs. Prompt is May the Force be with you.:

[eternal comet]

💀

this shit is so fun bro

[gusty cloak]

I have a desktop PC, though like I said I was also using a ControlNet

SDXL Canny

[eternal comet]

oh

lol i use shit ton of loras

still hasnt done that for me.

idk

check config

is it offloading

[gusty cloak]

Config as in?

config.json?

[nocturne oak]

There's no place like home.

[nimble mason]

a lot of genuine charm in that image, love it

[south temple]

2080 super. Render god here

60? 2060

Horse, horse farts so hard it blows over a town, the horse is happy, the town is not

[eternal comet]

i hooked it up to the internet

real

[hazy warren]

that yoda is pretty good

[hazy warren]

would you live there?

[nocturne oak]

Seems like a beautiful place, eh?

[hazy warren]

ye

[nocturne oak]

My mama always said life was like a box of chocolates. You never know what you're gonna get.

[zealous wolf]

生成一张全球高清地图

[hazy warren]

yum

[zealous wolf]

generate a global map

[hazy warren]

global map of chcoolate

[nocturne oak]

Magic Mirror on the wall, who is the fairest one of all?

[open grotto]

https://tenor.com/bcbxc.gif

[nocturne oak]

A martini. Shaken, not stirred.

[nocturne oak]

To infinity and beyond!

[jovial tiger]

I think mine's a vampire.

[nocturne oak]

It looks inpainted.

[jovial tiger]

yeah, but it came off the press that way. 🙂

[minor needle]

https://tenor.com/tKgaYjwJq16.gif

[nocturne oak]

You may say I'm a dreamer, but I'm not the only one. I hope someday you'll join us. And the world will live as one.

You miss 100% of the shots you don't take.

[twin remnant]

- Wayne Gretzky
- Michael Scott

[eternal comet]

average day

[deft bison]

robotic

[south temple]

ooo what set is that from ?

[eternal comet]

please do not announce to the servee

[south temple]

anyone got a prompt?

[heady anchor]

barack obama surfing on an ocean made of gold coins,the surfing board is shaped like donald trump

[south temple]

sent. this will take a moment to render... godspeed

[south temple]

first pass is about to pop

[south temple]

128 frames might have been alot

final render is in prog

[jovial tiger]

Here is the image you requested

[south temple]

absolutly backing down to 32 again

lol

[covert pagoda]

Sick flip yo, double board trick n everything

[south temple]

oh wait for what is about to pop out

10 minutes to render that at 125

[heady anchor]

whats your gpu?

[south temple]

3090ti

fe

[mellow orchid]

Guys, I need advice on model training. I trained a model of a girl as the first 2 pics. The same girl others trained as the last 2 pics. I'm wondering why is my result so blurry. I used 300 pics, 3 repeats, 10 epochs. I don't know how to make it more realistic.

[hazy warren]

gonna make you sick lol

[south temple]

trying to work on slowing that motion down

[south temple]

better

[rancid drift]

coffeehead artwork

[hazy warren]

will it do realistic stuff?

[south temple]

yerp

sure will

[languid pebble]

Good morning ^^

[clever oar]

lcd

[covert pagoda]

what

[gusty cloak]

I think they meant cuphead 😂

[covert pagoda]

😆

[dawn seal]

Because everyone wants a little tongue with their strawberry jelly(fish).

[clever oar]

canny test

i need shark with stawberry skin)

this strawbery wish shark shape😃

[arctic laurel]

Mark Twainsworthy

[clever oar]

is canny?

ip adapter

[arctic laurel]

it’s a mixture of tools—check out Become Image, it’s a comfy workflow on git

lemme find the link

https://github.com/fofr/cog-become-image

[clever oar]

can change texture skin with it?

[arctic laurel]

I believe so

[wispy nest]

i tried craiyon ai art generator today and i have to admit it’s terrible

it’s rlly good at making my favorite celebrity tho

[arctic laurel]

she’s kind of like a human puffin tho, like her features are naturally exaggerated

so that tracks

[wispy nest]

yes her face is naturally beautiful the only part abt her that’s fake is her body

[arctic laurel]

yeah that’s the tragedy of our times…the artificial bodies culture imposes on others

[wispy nest]

right

[arctic laurel]

i wish there could be more balance

😦

[clever oar]

Melted chocolate

[eternal comet]

Hungry

[clever oar]

lime

[eternal comet]

Loll

[clever oar]

watermelon

can i generate 4 promts by one click generate?

[grizzled sage]

in automatic1111, prompt from text file script

[gusty cloak]

A tiger cat!

I know tigers are already cats but you know what I mean 😂

[clever oar]

yes i test mix with canny

[spark ice]

#🏞｜general-with-images

[clever oar]

?

[cyan shoal]

https://huggingface.co/spaces/FoivosPar/Arc2Face

okay not bad, made him look a little feminine but still

[wispy nest]

young beyonce

[clever oar]

Look at me!

[cyan shoal]

Ashley! LOOK, AT, ME..

[wispy nest]

iv never seen such a terrible sight

[wispy nest]

nickiii

[clever oar]

recolor test

[clever oar]

mojo give me hibrid animal

[covert pagoda]

what?

[clever oar]

hybrid of turtle and jellyfish

[languid pebble]

😉

[languid pebble]

Will need some minutes ...

[jovial tiger]

feel free to give me another, I'm working on my regional prompter automation.

[clever oar]

cat want to be frog

[clever oar]

what is it

tool

[mild jay]

How are you combining animals like that?

[jovial tiger]

I'm using regional prompter. but for simple stuff you can just say. cursed cat-frog hybrid

[mild jay]

regional promoter confuses me honestly, you can get good stuff out of it but there is a steep learning curve.

[jovial tiger]

yeah, it's why i'm using automation to create it.

that said, you can get really easy simple stuff out of it.

[mild jay]

here is the very first imagei got out of region promoter i made it yesterday.

[jovial tiger]

This is just "Trump's head ADDROW Pikachu's torso ADDROW robot legs"

[mild jay]

oh i see

I can also assume your using an sdxl model unlike me.

[jovial tiger]

you're only on sd 1.5?

[mild jay]

my hardware is weird and im afraid of my pc exploding.

[languid pebble]

Should work with regional prompter, too

[mild jay]

I might need to test sdxl just because, is there a model you would suggest for general use by chance?

[languid pebble]

SDXL ... zavychromaXL is nice ... but always depends on what you need ...

[mild jay]

Im honestly not sure what i need thats why i like trying out models that can do a little bit of everything.

[languid pebble]

It can do a lot ...

[languid pebble]

Have a look at Civit.ai ... you will always find some example pictures ...

[mild jay]

will do ill also look at the model you suggested.

also here is an unrelated image.

[languid pebble]

description for the model is pretty important, too for best results ....

[mild jay]

yeah i guess...

[languid pebble]

I always open it when I switch to another model ... can't keep it all in my mind 🙂

[mild jay]

Yeah its hard to remind yourself of things sometimes.

[jovial tiger]

so that 1.5 512x512 resolution can also be frustrating with regional prompter. Sometimes for the complex stuff, I have to jack it up to 1920x1080, because 1024x1024 isn't big enough to fit all the details it wants to put in there.

when you're splitting up the image into 6 regions for example, you won't get doubling because each subject isn't ever more than 1/3rd of the screen, even at 1920x1080

[mild jay]

You know ive been wanting to make a comic or something with ai but im not sure if its possible.

[mild jay]

i see.

[languid pebble]

Better do it picture by picture ... but consistance is still a problem

[mild jay]

Yeah i know and seeds seem like a good start or image to image.

[languid pebble]

AFAIK MJ has kinda consistant characters ...

[mild jay]

what in the world is that?

[languid pebble]

Midjourney ... a service for A.I. Picture creation ... but commercial

[mild jay]

oh yeah that, i think ill stick with free sd (stable diffusion)

[languid pebble]

They will find solutions soon, too. It's a crazy time

[clever oar]

lool my transformation😃

original

😄

wtf

i no need like that

[wispy nest]

I have stopped saving my promts to prevent duplicated images of my art in future with all my modifications would make it nearly impossible to make the same or close to same image lol

[clever oar]

alice best game

[languid pebble]

Alice is part of the prompt, yes 🙂

[jovial tiger]

@nimble mason

[languid pebble]

Time for bed. Have a good one!

[eternal comet]

Shall I send some of my Loras gere

Here

[eternal comet]

whats the prompt

[jovial tiger]

it actually won't let me paste it here. says it's against the rules or something. maybe the formatting.

[eternal comet]

do "cpp"

`

` 3 of these then put cpp

then another 3 to close

[jovial tiger]

A torrent of squirrels, rabbits, and birds cascading from above, in the style of a surreal oil painting ADDCOL, Chandeliers shaking and swaying amidst the falling animals, in the style of a surreal oil painting ADDCOL, Wallpaper and curtains fluttering in the chaos of the scene, in the style of a surreal oil painting, ADDROW, Shocked, well-dressed patrons recoiling and shielding themselves, in the style of a surreal oil painting ADDCOL, Plates, glasses, and silverware scattered across white tablecloths, in the style of a surreal oil painting ADDCOL, Animals landing on and scurrying across the fine dining tables, in the style of a surreal oil painting

ok there we go

finally finished my claude 3 based ai regional prompt automation.

The input prompt was "flood of animals falling from the ceiling over a fine dining restaurant, much to the shock of the patrons"

[wispy nest]

definitly something i would make and thats why i love it

[eternal comet]

gimme something to train lora for

[minor fractal]

sd3

it knows what a nendroid is

[crisp stream]

Is it capable of doing painted stuff as well? Only did see realistic/3D stuff in the exampleimages on Civitai

[last lava]

https://tenor.com/TW2f.gif

[wispy nest]

more nicki photos

[eternal comet]

natural selfies

lora training

[south temple]

this is epic

[hazy warren]

oh cool, how did you do this one??? this one looks so good!

[gusty cloak]

I was actually doing selfie shots myself today 🤔 (Some of these are the same seed, but with a switched up ethnicity)

Right! I know exactly what you mean

Oh my God 😭

Can my model do ice cream? 😂 I have to try

I still have A1111 running in the background

Some of these should go to #1019361238234443776 😅

[hazy warren]

good balencing

[gusty cloak]

I feel like my model wasn't optimized for ice creams either 😅

Let's assume this is a caramelized orange slice 😂

[hazy warren]

whats the point of A.I withou ice-cream

[nimble mason]

i heard there's ice cream in there

[gusty cloak]

1.5 checkpoint (1st) vs. XL checkpoint. I guess eating ice cream is hopefully something to look forward to in SD3 😂

Not sure what's up with the quality in the second one

[nimble mason]

actions are a big weakness of SD

[hazy warren]

mountian needs to be made of ice-cream

[hazy warren]

god knows when that will ever get here

[steady kettle]

hey guys, how do I create images?

[hazy warren]

get out a pencil and paper

[steady kettle]

?

[hazy warren]

start drawing

[steady kettle]

what command should I using?

[hazy warren]

the write command

[steady kettle]

In midjourney is "/imagine"

Here ...?

[nimble mason]

Beep boop! I am a bot! Prompt me with "Prompt! :: your prompt here"

[steady kettle]

In this chat? To create images?

[nimble mason]

Beep boop! Yes!

[grizzled sage]

ice cream licking prompt on lavi-bridge t5 adapter

[steady kettle]

I know you are not a robot

[gusty cloak]

You see, the big brain move is to use Dall-E 3's admittedly good prompt adherence and then pass the output through Canny + some img2img to get the SD feel

Sorry, this took some time to cook 😂

[steady kettle]

The prompt doesn't work

[gusty cloak]

Stable Diffusion is open-source, you need to run it on your computer locally

[nimble mason]

Beep boop! Just try a prompt!

[eager prism]

How is it possible to create images?

[gusty cloak]

Stable Diffusion is open-source, you need to run it on your computer locally

[nimble mason]

Beep boop! Prompt me with "Prompt! :: your prompt here"

[steady kettle]

Wtf... I've used it before, with a prompt in this Discord. But I dont remember how

[gusty cloak]

I suggest diving into using a web UI as this gives you tons of control with 0 restrictions 🙂

That is, if you have a good GPU

[grizzled sage]

meh. dale doesn't impress with actions too well either

[eager prism]

/prompt! :: circle blue

[gusty cloak]

At least the ice cream is touching his mouth 🤣 Not a piece of caramelized orange either

[grizzled sage]

whats the prompt?

i'm gonna run it through t5 to sd15

[nimble mason]

Beep boop! Here is the image you requested.

[steady kettle]

Like I said, doesnt work

[gusty cloak]

Perfect, because it uses 1.5 keywords as the first model was 1.5:
1man, 23 years old, Indian, eating an ice cream, brown eyes, short black buzzcut hair, very small beard, tight white t-shirt, depth of view, hiking, mountains

[grizzled sage]

the bots dead zed

[nimble mason]

Here is the image you requested.

[steady kettle]

stop man

[steady kettle]

no one believes you

[nimble mason]

Here is the image you requested.

[gusty cloak]

When you don't open the image preview, it almost looks like he's holding a cigarette 😂

[steady kettle]

I just want to know the command to create. Help me please

[gusty cloak]

As I explained, there is none

[grizzled sage]

I think the community would have a TON of fun with lavi-bridge. I wish i could use it in a UI. I wish i could port this project to an extension

i wish i had a girl with a phone i would call her

[steady kettle]

But I remember. Do not exist anymore?

[gusty cloak]

Not on this server as of writing this. But as I said, I strongly encourage you to explore using SD locally

[nimble mason]

Here is the assistance you requested.

[grizzled sage]

your memories were implanted by a false bot

[steady kettle]

No... Like a MidJourney, here 's the same

[grizzled sage]

its like total recall. synthetic memories

[eager prism]

Exactly, but how is it the command to do that?

[steady kettle]

Sure

[grizzled sage]

how would you know the difference between an implanted memory and a supposedly "real" one

[nimble mason]

Beep boop! Reply to me with "Prompt! :: your prompt here"

[gusty cloak]

By knowing the difference 👀

[grizzled sage]

the difference is just another implanted memory. now what

[gusty cloak]

You implant the knowledge

Embody AGI yourself

Anyway enough bullshitting from my part 😂

[grizzled sage]

what like the matrix? this is real life we're talking about here not the movies

[feral crystal]

guys, is this normal? this took 9 minutes to reach 99% then it gave me the error "OutOfMemoryError: CUDA out of memory"

I have RTX 3060 12Gb VRAM. + 16Gb RAM

Web UI launched with these arguments: --xformers --api --autolaunch --skip-python-version-check --theme dark --no-half-vae --medvram

[feral crystal]

Realistic Vision V6.0 B1 model

[gusty cloak]

I was messing with you, the truth is you can't know which memory is implanted and which isn't

[steady kettle]

Before it was just writing "/prompt create........", here in Stable Diffusion discord. There's a IA bot

like a midjourney discord

[nimble mason]

Beep boop! Try using Forge.

[eager prism]

/prompt create mattress with two density

[grizzled sage]

maybe my posts weren't flippant enough.

[grizzled sage]

i'm gonna go lie down. i'm bummed that there's no nodes or extensions for lavi-bridge again

[feral crystal]

Ok I will install it. Hope it fixes it.

[gusty cloak]

Apologies, my intention was to make a lighthearted joke, not to disrespect.

[grizzled sage]

[eager prism]

@nimble mason how did you do it?

[nimble mason]

Beep boop! I am a magic bot. Please prompt for another image.

[eager prism]

/prompt create : mattress bipartite with two different density

[hazy warren]

Theses an argument you are@meant to add to prevent memory leaks but I forgot what is it

Also, u don’t need 100 sampling steps

[gusty cloak]

All of this has inspired me to create more images for hanging prompts:

[feral crystal]

Forge is fast but with SDXL models it's the same as A1111, takes lot of time.

[nimble mason]

Forge is much faster than A1111 with SDXL

faster than even comfyui for some reason

on my 4090 it's just over twice as fast

[feral crystal]

I see, this took 2 min, 12 seconds to generate. Is it normal? Can I try adding arguments to webui to increase speed?

Realvisxl V4 model

[feral crystal]

@nimble mason

[jovial tiger]

So something I mentioned in DM, if I switch models in a1111, it slows down to half speed until I restart the service. once I restart, it's back to full speed again.

[nimble mason]

depends on your gpu

it is pretty frustrating that performance is such a problem with a1111

it's THE reason i started using comfyui in the first place

[feral crystal]

I do have comfyui, but I need to familiarize myself with it, that's why I'm using A1111 it's simpler

[nimble mason]

it is and it isn't imo

[jovial tiger]

man, for all the times this automation fails, sometimes it comes up with awesome stuff

[lofty frost]

amazing

[gusty cloak]

Did you use a ControlNet for this? 🙂

[arctic laurel]

just my thumb and index fingers

j/k—it’s ’Become Image’

and a little SUPIR upscaling a

[arctic laurel]

i would feel really uncomfortable eating my muffuletta sandwich in front of that giant bird

[covert pagoda]

the bot is down

[arctic laurel]

Last one but this is the best

[open umbra]

#🏞｜general-with-images

[wispy nest]

#🏞｜general-with-images 画太阳升起

[languid pebble]

Simple SVD if it recognizes a full body person it sometimes tries to make it walk

[hazy warren]

😮

[languid pebble]

Painted stuff also works sometimes ...

[languid pebble]

And sometimes not 😄

[hazy warren]

I looked at some vids on Reddit, its actually pretty decent.
how long can u make the vids?

[languid pebble]

The normal way 4 seconds only ...

[hazy warren]

aww, kinda small 😦

[languid pebble]

Usual an good working setting is: 15-25 frames (more doesn't work well) and 6FPS. It should be possible to do a Video2Video later ... I'm using other tricks ... SloMo can make it longer, too

[hazy warren]

so 25 frames max? 4 sec video?

[languid pebble]

25/6FPS ... if you do more FPS it will be shorter

[hazy warren]

ah ok, is there a setup tutorial?

[languid pebble]

Many at Youtube. Stable Diffusion Forge comes with preinstalled SVD AFAIK

Pretty happy with forge ... if you are on A1111 you might wanna try it ...

[heavy wing]

#🏞｜general-with-images

[lean talon]

cat donut

[languid pebble]

Art ... 😁

[lean talon]

this workflow is nuts https://civitai.com/models/372584/ipivs-morph-img2vid-animatediff-lcm

instant animorphs

[hazy warren]

create a video of a person swimming in lava

[languid pebble]

Never tried that ... but a waterfall made of colors doesn't seem to be a waterfall for the A.I. and longer ...

[hazy warren]

lava swimming, lets see it!

[languid pebble]

Even a 4090 is limited ...

Working on this:

[hazy warren]

😮

[clever oar]

ice cream shark)

[worldly wolf]

Whatchu guys think -am I pushing Sdxl to its limit yet?

[clever oar]

mojo more hybrid picture

[languid pebble]

I'm working on my projects ...

[clever oar]

oh

[languid pebble]

You could try Leonardo.ai ... you'll get free credits there every day

[clever oar]

difussion master pic

nice hybrid)

[clever oar]

color test

[languid pebble]

It has park assistance ...

[clever oar]

I didn't immediately understand what you were talking about lol until I saw the collision

[midnight bough]

CLEAN UP ON AISLE MY PANTS!!! 🤯😳😍😩🤤

[nimble mason]

Weird shit = awesome

[clever oar]

cyberpunk restyling

[jovial tiger]

Wow that's really neat. Do we know what they used?

[nimble mason]

not sure, all i've used for video is animatediff so that's my only guess

[clever oar]

he give link below

[jovial tiger]

very cool, thanks. the downloading begins (there's at least 10 models referenced in that thing)

[clever oar]

I wanted to download too

[clever oar]

not a standard use of a neural network; I make design options for my rooms in the house using photographs

[grizzled sage]

i think the only standardized use of neural networks so far is brains

[dawn seal]

Scooby Doo, Where Are You??

[jovial tiger]

prettyt trippy stuff

[nimble mason]

that's awesome as hell

[jovial tiger]

gonna try a couple of similar shark pictures next.

[nimble mason]

that's wild

[jovial tiger]

yeah it works best when the 2 pictures have only 1 subject in them, otherwise i think it has trouble latching onto what to center on.

this next one is gonna be neat, less subjects.

[clever oar]

it for comfy only?

[jovial tiger]

yeah it's a comfy workflow

[clever oar]

😭

[nimble mason]

just use comfy lol ppl make it sound like you need to be able to solve partial differential equations writiing with sticks on the beach... it's really not bad at all

espec with premade workflows...

you don't need to know much about how comfy works to be able to use a lot of stuff

[jovial tiger]

i dunno man, i had to go all over the internet and in tons of directories to assemble all the models this thing needs

wow

[nimble mason]

oh, that one i haven't looked at

[jovial tiger]

the one i'm doing now, girl in a hawaiin shirt... it even makes her move around and dance, with no prompting at all from me.

[clever oar]

lol a111 is my religion

[jovial tiger]

i was like you, and for straight image gen, i still use it, but it's really limited for anything other than txt2img and a very mild img2img

there's a world of clipvision to explore that does stuff that no txt2img can.

lol this one i'm doing now has this dancing girl and then a tiny little snippet at the end of godzilla.

[nimble mason]

and especially the new ipadapter that's comfy only... holy hell is it something else

[jovial tiger]

ok i understand now, the first image is the primary, and the second pic is the stuff that mildly gets morphed in. it's not a 50/50 split.

yeah this thing is all ipadapter.

[nimble mason]

i haven't looked into it at all yet, but SDXS could make things especially interesting

people on reddit said they were generating at nearly 300 FPS on a 4090

yes, nearly 300 images per second

[jovial tiger]

so what's sdxs?

[nimble mason]

that's about all i know... supposedly around turbo level quality and resolution, i think

[jovial tiger]

SDXS is a model that can generate high-resolution images in real-time based on prompt texts, trained using score distillation and feature matching. For more information, please refer to our research paper: SDXS: Real-Time One-Step Latent Diffusion Models with Image Conditions. We open-source the model as part of the research.

[clever oar]

the truth is that for comfy I don’t need to be a programmer and configure a million windows

[jovial tiger]

ok i see. they have a demo up. so definitely useful if you want it to generate based off a live camera feed or as like a music visualizer or something. the quality is very low, but if it's high frame rate moving as a response to something, it would probably look great.

[nimble mason]

nice workflow

gotcha

[jovial tiger]

wow really neat.

works really well with vector art apparently

[nimble mason]

sdxs? or the animation?

ah

[jovial tiger]

sdxs

i wish there was a way to get these mp4's to loop. gifs do but they're huge.

[nimble mason]

interesting... so with low quality, does that mean low density of assets? or just shitty details

yeah

and they have the 256 bit limit

i think webp can loop

[jovial tiger]

yeah the demo of that sdxs is seriously low quality. 512x512 and it's majorly blury and crappy looking. as i said, with very high frame rate that could make it look way better, but it's garbage as a single photo

[nimble mason]

i have a workflow somewhere for making looping webp with animatediff

is it a huggingface demo?

[nimble mason]

i swear those look awful most of the time anyway... skeletal settings

[jovial tiger]

yeah webp loops, although not in the feed

hmm

[languid pebble]

You might wanna have a look here -> https://www.purz.xyz/

[jovial tiger]

oh well, looks like gif is it

risky click of the day. 🙂 what is that site?

[nimble mason]

layer blend modes are real interesting with ipadapter

[languid pebble]

Ohhh... wait ... not the workflow page ...

[nimble mason]

what makes me sad is we don't have a curves node

hard to believe that's the case, but it is afaik

[languid pebble]

Dang ... I had those workflows somewhere ...

[nimble mason]

if you only have literally one thing for image postprocessing it's that

input, three outputs with different tricks with blend modes before fgeeding it into ipadapter

[jovial tiger]

would be neat to put those layer things through this kind of animation

[nimble mason]

more shit you'll never get with a prompt alone

[languid pebble]

That's the one I have been thinking of: https://discord.com/channels/1076117621407223829/1215481171148939304

[nimble mason]

at least for now, so i guess the word "never" is inappropriate 😛

blended the input image with itself via "add" at 100% opacity for the neg prompt, and "color burn" at 25% opacity for the pos prompt

[jovial tiger]

it's neat how it makes an animation between them.. one was an eye, another a granny with headphones. so she leans into to look and then it zooms into an eye

wow

so this works really well with 2 of the images from one of the 9 set renders.

that way they're rather similar and it does animation between them.

@nimble mason

i wonder if there are other black and white animated gifs that we could use.

this is doing this one.

[nimble mason]

good idea - optical illusion images could be really cool

[jovial tiger]

ok trying this one, let's see if it works

[calm canopy]

Rorschach blots could be very interesting and fun

[jovial tiger]

so it looks like it doesn't want to do anything with the others i'm trying.

maybe it's not smooth moving enough.

it just generates regular animations between the frames.. which are certainly impresie, but not like that circle effect.

[grizzled sage]

theres tons of these style of animations out there on stock video sights. "transition animation masks" or other kind of animated masks

[jovial tiger]

like this

[nimble mason]

ipadapter with exposure adjustment

[grizzled sage]

another one of these i got kickin around

[grizzled sage]

aren't the exposure nodes basically just post processing?

[nimble mason]

yes

is there any way to make the node search window in comyf bigger?

[jovial tiger]

yeah, i even changed out the load video node to a local upload. it won't do anythign with any of the mp4's or gifs I'm getting off the internet.

the only one that works is that original circle one.

[grizzled sage]

i was playing with touch designer, trying to figure out animations of this style in it. i might just revert back to adobe products or blender since thats my foundation

i've seen people making animated region maps too and that's pretty interesting

[jovial tiger]

this reminds me a lot of infrared photography

[lean talon]

if you play with the strength in the qrcode control net you can get other mask videos to work

[nimble mason]

i'm actually using images with inverted luminosity to get the strange lighting in the center

[lean talon]

https://media2.giphy.com/media/v1.Y2lkPTc5MGI3NjExeXdzazFxZGllZXI2eG1yajJxb3ZwNGFuYzdoODZtZWIyN2pmNjh1ZSZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/xTiTnfckn5j6EgTUAw/giphy.gif

[jovial tiger]

whoa

[lean talon]

yeah, if only this was faster lol, i've been playing with streamdiffusion since it dropped in december, but this is so much more visually interesting than flickery neural soup

[jovial tiger]

wow you're right. i upped teh strength to 0.75 from 0.45 and it's working now.

[jovial tiger]

https://civitai.com/user/ipiv holy cow this guy has tons of these.

he has ones that are way more coherent too. so just mild animation to bring a still to life.

wow... he's got one that takes any text input and makes a simple motion thing out of it. really cool.

@nimble mason

[teal dune]

is it possible to decode prompts from midjourney or recreate an exact copy. Especially i want to know promts from artist gokuryo.

[languid pebble]

Ask the creator?

[teal dune]

He doesn't give

[languid pebble]

Cause it's his work? Maybe better create your own prompts ...

There A.I.s pic2promt ... not very good but they might help you.

[nimble mason]

if you want an exact copy, copy the image... lol

do you mean something close?

[jovial tiger]

midjourney does its own prompt expansion. i don't know of a way to get that expanded prompt.

[teal dune]

almost exact promts)

[languid pebble]

Clicked it to see animation 😄

[jovial tiger]

have the prompt for that or is it clipvision? I'd like to throw it in this thing.

[nimble mason]

thats ipadapt not CV

[jovial tiger]

this other workflow is really good for creepy stuff.

[jovial tiger]

To quote Thanos: Fine, I'll do it myself. 🙂

better as gif. half ressed.

[teal dune]

I think should add lora

his version so cute <💖 3

[nimble mason]

mine is cuter

[teal dune]

😆

[lean talon]

this mask worked pretty good https://media0.giphy.com/media/v1.Y2lkPTc5MGI3NjExd3FoYmszMTloZ204dDhuc25kYnp2Zm54N29oa3RudnJ0MWtnOG14ayZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/26BGOpbRi4ZtkZ1kI/giphy.gif

[lean talon]

working in comfyui feel like

[south temple]

why did this grid thing happen?

[gusty cloak]

Do you have some sort of tiling?

[south temple]

looking

I ran like 4 hours of render and got all poop for it lol

I wouold assume.... its somthing in this upscaler

[nimble mason]

yup

that is your issue

[south temple]

lol what tho?

[nimble mason]

cat asllep in my arm so im gaoing to just wave my one free arm useless/y lol

[south temple]

Can’t figure out which setting I need to turn down or off

[clever oar]

comfy looks like you're not using a neural network, but creating one😃

[cyan shoal]

SD3 image:
In a grand castle, a plump orange cat with emerald green eyes gazes out at the horizon. The castle walls are covered in intricate carvings, and the cat's fur is a deep orange hue. The scene is set in a lush green forest, with the sun casting a warm glow on the cat's fur.

generated by a beta tester called CrystalWizard on twitter

I showed them SuperPrompt-v1 and provided an example and they actually generated it

SD3 was rightfully confused as the output prompt was set in 2 PLACES AT ONCE! (forest + castle??)

[calm canopy]

That cats in a pretty darn large castle, its got a forest inside of it. And another castle!

[cyan shoal]

yeah

SuperPrompt v1 isn't exactly smart, its a 77M model

[calm canopy]

Otherwise tho, pretty decent detailing

[cyan shoal]

especially for the preview access models, which are often blurry or lowres

compared to whatever comfy and lykon are using 😫

[calm canopy]

The lack of color bleed is nice too. With 1.5 models that I'm used to, putting emerald eyes or orange fur will make a ton of things orange and the image will have things that sorta almost look like actual emeralds.

[lean talon]

still couldn't put the cat in the castle

[nimble mason]

comfy is waaay more powerful for a reason

[cyan shoal]

just looked at his twitter and it seems he might've used my prompt again!

[calm canopy]

Tbh the AI did a good job of the setting based on what it was given: 2 settings, a castle and a forest

That one is much better

[lean talon]

damn

[clever oar]

Can I quickly generate something without setting up many windows?

[nimble mason]

yeah just drag in an image with a workflow

[calm canopy]

Yeah just pull up someone's workflow and throw in your prompt

[cyan shoal]

yeah if the prompt was less confusing:

In a grand castle, a plump orange cat

scene is set in a lush green forest
And if the model's weren't random, then it would have been more consistent

[calm canopy]

No node knowledge or editing needed

[south temple]

these were great

[nimble mason]

did you figure oul ur issue

[south temple]

nope

rendering now

no idea

[nimble mason]

i think youre sending sdupscale all the frames as a list

instead of one at a time

[south temple]

trying one more thing

[nimble mason]

or something like that cuz it looks like its treating each frame as a tile

[south temple]

right!

i hit somthing while not paying attention

[nimble mason]

work nowL?

[nimble mason]

vthats really food

good

[arctic laurel]

😄

[nimble mason]

cat still asleep in my arm

[arctic laurel]

it’s ‘Become Image’ along with a little SUPIR

[nimble mason]

and iu use dvorak kb layout so typing one handed blind lol

[south temple]

this one was coming out great!

[jovial tiger]

Other than the vae of course, they keep posting stuff that SDXL can do just fine on its own. Nothing that really shows off sd3.

[cyan shoal]

I didn't even ask him to post using that prompt, but it was kind of him to do so

nice image btw

[arctic laurel]

it almost looks like the hind leg wants to be a second tail

[cyan shoal]

?

[nimble mason]

i want to see sd3 with someone eating ice cream

[calm canopy]

Yeah I think there's a certain setting you need to check in order for ultimate SD upscale to not create a grid of your image (or just use control net tile)

[cyan shoal]

I think someone did that

[arctic laurel]

no more pixar ice cream?

[cyan shoal]

but it looks like whatever current finetunes can do I think (the ice cream)

realistically im gonna be using sdxl turbo as a daily driver anyway

[jovial tiger]

Sure, that's a good point.

[south temple]

i wanna annimate these

[jovial tiger]

having the base look like that is pretty great.

[cyan shoal]

I hope we can tone down bokeh and other stuff

[cyan shoal]

I don't want another anti bokeh lora that makes everythign super bright

[calm canopy]

The crazy part of sd3 is the natural language prompting tbh, image quality is already crazy insanely good on other models

[lean talon]

text handling too

[lean talon]

being able to actually make readable text opens so many new creative possibilities

[jovial tiger]

definitely.

[cyan shoal]

I'm less interesting in text stuff, unless its about memes, then sure its gonna be hilarious

[south temple]

larger batch size might smooth these out ya think?

[cyan shoal]

but prompt adherence man

you can describe people much more closely if the model itself doesn't know

[cyan shoal]

nice

[south temple]

smooth motion on the cat... but how does one slow down the morhing images in the backround?

cuz that light is bouncing all over hell

[lean talon]

inside the waifu generator

[cyan shoal]

I love the sunlight shifting about

[calm canopy]

Yeah it's gonna be wild to be using natural language in prompting after mastering the concept of prompting with "(masterpiece:69.69), best masterpiece award-winning artwork by gold medal artist" but it's going to be super fun seeing how sd3 reacts to descriptive/detailed prompts

[south temple]

this would have been great

it the tiling were not a thing

[calm canopy]

Could always cut each frame out, mask the background, then place a static image of the background over it all. Or at least, a slower and better version of it.

[clever oar]

I'll be gone for 5 minutes...

[south temple]

lol

that cats motion is NUTS

still tiling...

[gusty cloak]

May I ask what kind of adapters you used for this? I'm looking to combine faces sometimes!

[arctic laurel]

yep, this is just the Become Image workflow. There’s a cog for it—hold plz.

[arctic laurel]

https://github.com/fofr/cog-become-image

[arctic laurel]

I would describe the temporal fidelity in this clip as “shroomy”

[crisp stream]

Looks quite 3D though 🙂