🏞｜general-with-images

[royal monolith]

roosters have very sharp beaks - you tell him

[clever oar]

Nagens for dinner today

[royal monolith]

he's in an operating theater surrounded by sharp knives. be very, very quiet...

[clever oar]

ok I'll be quiet like a ninja

[quartz ravine]

Stable Diffusion 3.5 Medium alongside other open-image base models. This model only requires 9.9 GB of VRAM
Anyone have a good comparison between XL/2/3/3.5?

I'm curious how much of an improvement 3.5 is

[royal monolith]

that sounds like a good project for you :)

[round olive]

notebook girl with name keli style Barbie

[royal monolith]

the channels to generate in are the Artisan channels. start by reading the information here #artisan-faq

[sterile kiln]

Bilbo Baggins, SD 3.5 Large > Flux1Dev > Upscaler

[nocturne oak]

My lonely NES is killing me...

[nocturne oak]

& eye!

Eye must confess; eye still believe.

Still, Bill leave.

[jagged badger]

eye

[steep sleet]

Flux RF Inversion

[viral frost]

I gave a super random prompt, and all Flux wants to generate is macro photos of insects.

[scenic yew]

1.5(less then 1b) knows the most art styles I believe but has very very bad prompt following, and the base model is really low quality. got a lot of community support, lots of finetunes, ip adapters, and controlnets. Finetuned models are great at their specific domains though.

2(less then 1b) didn't get much community improvement, should have slightly better prompt following(still very bad) then 1.5 but not that great image quality either. Finetuning it helps generate better images but its not as good as 1.5.

sdxl(3b) has better prompt following(not great compared to modern) and better image quality, pretty bad text rendering but could write single words unlike above, got a lot of community support and has lots of finetunes, ip adapters, controlnets and techniques like lcm

sd3 medium(2b) was ok, should have more detail and prompt following then above but honestly the aesthetic was worse than sdxl, and humans were horrible especially on different poses, but decent text rendering, however very little community support

sd3.5 large(8b) is a considerable improvement, knows lots of art styles, great prompt following, nice image quality, good text rendering. seems to have lots of support. sd3.5 medium(2.6b) is alright, the aesthetic quality is worse then sdxl but prompt following is better, not too great text rendering, humans are similarish to base sdxl.

[languid pebble]

Has anyone tried to add lying Sigmas to Cascade Ultrapixel?

[quartz ravine]

prompt? That's cool

[quartz ravine]

thx

[maiden pike]

Create a freddy krueger with realistic body

[floral fern]

Billions must make the official Catholic Church mascot a lora

[twin jolt]

I am trying to create images like the following, but they always come out blurry. I don't have any problems with other images with all the same settings. I also sometimes get blurriness with photos of white sand beaches. Is it the brightness? What is it about this prompt? "flat white background, 3D Unreal Engine render of an old green wooden chair"

[nimble mason]

https://github.com/ClownsharkBatwing/UltraCascade guessing you've seen this

but this is much more ready to use than the other implementaitons just fyi

[spice kiln]

i have two gpu's one with vram 12 gb on with 16 gb. is there any possibility to run stable diffusion video using these two. its would be a great help. i am new learner .

[clever oar]

cant do normal background with x ray man

[languid pebble]

Get 200 credits for the best voted new A.I. https://www.recraft.ai/invite/W7Usp3Pk31

[languid pebble]

Need to find my most complex prompt and try it there. They say it can do loooong texts

[clever oar]

works with -"in cafe"

[clever oar]

what is it do

[languid pebble]

A new A.I. Model best voted by users at the moment.

[clever oar]

thx maybe i try it later

I think the XL model is the best because it is available on moderately expensive video cards with good quality

[languid pebble]

XL should be fine ...

[clever oar]

more freedom)

[clever oar]

i can use 1.5 only XD

[clever oar]

need breakfast 🍔

I just woke up

[wispy nest]

recraft looks really strong yeah

might need to run a different model as refiner over it sometimes cos the photorealism is a bit off sometimes

but it does layouts very well

[languid pebble]

Made with recraft

[wispy nest]

ye that's really good

text all correct and high photorealism

[languid pebble]

Flux didn't do the prompt that good ...

[steep sleet]

ReCraft

E x c e l l e n t T e x t on this ReCraft!

[reef atlas]

It's paying. We can't use it locally for free...

[steep sleet]

Errrrm I've got a Free account so far ...

[languid pebble]

If you open account using ref link you get 200 instead of 50 credits. The referer gets 500 credits ... so better refer

[dry blaze]

You like a result, you want to tinker with it, and met with this. This will take a while to install nodes for
Especially if i have to uninstall "conflicting nodes" and install them manually lol.

[worldly quest]

Iwjs in skisuwsuiaus

[reef patio]

Купальник для художественной гимнастики

#artisan-1 купальник для художественной гимнастики

#купальник для художественной гимнастики

[steep sleet]

Rhythmic Gymnastics

[subtle crater]

New Discord stuff.

[clever oar]

you very like gold😎

[languid pebble]

Thanks. Super Flux with my LoRa 🙂

[fleet goblet]

"Super Flux"

The Patreon workflow?

[languid pebble]

When you don't render in 30 Steps but 3 times 10 Steps

[fleet goblet]

What workflow does that?

Literally none of them I've seen re-iterates the same image three times at 10 steps.

[languid pebble]

Selfmade Workflow .... bad luck I don't know how to add the Detail Demon

[clever oar]

Mojo

[fair remnant]

is the text also ai

[languid pebble]

Yes! Flux Model!

[robust sparrow]

some results with Mr P Brawl Stars LoRA that I trained for fun on weights.gg

Didn’t expect Flux.1-dev lora training to be good with just 8 images lol

[fair remnant]

wow that's actually crazy

[languid pebble]

There's a new model that can do even better

[fair remnant]

do you use comfyui

[languid pebble]

Yes but this new model is available only online ...

[languid pebble]

https://www.recraft.ai/blog/recraft-introduces-a-revolutionary-ai-model-that-thinks-in-design-language If you want I can send you an invite. 200 Credits for you (insted of 50), but also 500 for me ...

[scenic yew]

I tested it a bit, flux dev actually performs better.

One of the worst things about recraft is that the text is positioned horribly and looks like you photoshopped it on the image.

Good realism tho but prompt following is also pretty eh imo compared to things like sd3.5 L and Flux dev

[languid pebble]

You can also place text and an image manual and let the A.I. bring it together

[sterile kiln]

the song must not be the same

[clever oar]

vegetable fashion week

[plush ember]

Hello! I am new to this community, im not sure where to ask for help. How would one go about creating such images? The base model for all of these is SDXL, and I can see that the creator has used specific checkpoints, but my generations are no where near what is shown in the image.

[royal monolith]

what are you using for prompts?

[plush ember]

I use comfyui, and usually just manually input the prompts

I was experimenting with image to image upscaling using ksampler and it added way much more details (upscaled twice)

[plush ember]

Is that the standard practice to add details into a base image?

[royal monolith]

there are lots of things, especially if you're using comfyUI, that you can add in that can enhance an image.

[plush ember]

Could you tell me what else are some good practices?

[royal monolith]

one thing you probably want to experiment with is changing the number of steps. and changing the cfg values. also, the sampler and scheduler you use will ahve a huge effect - and some models will not work with some samplers or schedulers.

[languid pebble]

All you should be able to see on CivitAI what others use

[royal monolith]

yeah. read through civit and see what people are saying they are doing and useing

[plush ember]

I have been using Euler, samplers converge after some amount of points right? Will adding more steps increase the details to a huge extent? I was under the assumption that more steps will have diminishing returns on details beyond a point, and my gpu is pretty shit

[royal monolith]

euler isn't always the best choice

don't assume. experiment

[plush ember]

Yeaah, I have been doing that till now, but couldnt wrap my head around how could I possibly get more details into the images

[plush ember]

Alright, will do

[royal monolith]

that's why you do a grid. just use the word 'apple' and then try generating it with various numbers of steps. or various cfg values, and so on

and study the changes in it

[languid pebble]

There are always a lot of things to try ... like Detail Daemon ... SuperFlux ... you can see some of the add ons in youtube videos

[plush ember]

I could achieve these results just by increasing the steps and playing with the basic variables I assume?

[royal monolith]

never assume

i nothing else, contact the guy that created those and ask him what he did

[plush ember]

Thank you, I havent heard of these will take a look into it 😁

[languid pebble]

And I don't think you will get the same picture ... different graphic cards, software versions ... etc.

[plush ember]

Did that haha, waiting for his reply

[plush ember]

Yee sure sure, I wanted to know how to achieve such details, not the exact images

Thank you both

[languid pebble]

And every Model is different ....

They also often need a different kind of prompting ....

It's not as easy as they might tell you

[royal monolith]

what he said - thus my admonition to explore

his images look to me like he has hundreds of hours in figuring out how to make it work

"it" being ai generative imagry

[languid pebble]

Started spring 2022 😄

[royal monolith]

hundreds of hours :) i'm just a few months ahead of you

[wispy nest]

everything the same except sampler

same seed even

LOL

[languid pebble]

For any reason I can't add nodes at the moment ^^

[wispy nest]

comfy without nodes is pretty rough

[languid pebble]

I have a lot, but can't install 4. I think it's cause they have different requirements ... but not 100% sure

Or 4 Node Packs

Clowny is pretty deep into that topic ....

[wispy nest]

you can get a decent amount of the benefits if you use DPM++ 2SA and then put Eta and S_noise as high as you can

DPM++ 2SA can keep up with clown a lot better than euler

[royal monolith]

git clone

[languid pebble]

I think Flux is better with DPM++ 2M but all that stuff is condusing me ... getting to muc

[royal monolith]

flux is better put in the darkest possible drawer and locked away

[languid pebble]

Manual doesn't work, too ... waiting for new Comfy, now ...

[royal monolith]

what nodes are you not able to add?

[wispy nest]

I do like DPM++ 2M a lot as well

[languid pebble]

I think one was the WAS?

[royal monolith]

clown?

[languid pebble]

No ... not from Clown ...

[wispy nest]

clownsampler

[languid pebble]

WAS Node Pack

Suite

[royal monolith]

has he actually created a new sampler and named it after himself?

[wispy nest]

haha it works with a second node called sharksampler 😂

[languid pebble]

😄

[royal monolith]

oh this is much too good :)

and a batwing scheduler?

[wispy nest]

that would be good yeah

[languid pebble]

And Lying Sigmas ....

[wispy nest]

he tends to use a big noodle chain of schedulers I don't use that bit
I prefer manual sigmas to schedulers

[royal monolith]

and entire comedy branch of math

[celest sigil]

I think it's clown scheduler and shark sampler. Something like that anyway.

[wispy nest]

also the names change over time (sadly)

there's some other stuff in the repo I haven't even tried yet

[vagrant dust]

I would recommend you to search for the tile controlnet if you want more detailed images...

[plush ember]

Will do, thank you!

[vagrant dust]

Example of the tile controlnet left base generation and right the tile upscaled one

[plush ember]

Its past midnight here, so will have to see how this goes tomorrow haha

[plush ember]

Btw, how intensive is creating a 3d model from an image? I have gtx 1650, with 4gb vram

[royal monolith]

there aren't any good at home open source text or image to 3d AI generators

[plush ember]

CRM seems pretty impressive tho? Im not sure if its open source or not

https://github.com/thu-ml/CRM

[plush ember]

Are they not good at the mesh generation or material generation more specifically?

[royal monolith]

even the paid ones aren't that great at mesh geneation yet

you'll be better off just using blender for at least another year

[plush ember]

Shit, here i was thinking i could maybe get this to give some passable results

[royal monolith]

you might be able to. try and see

[jagged hedge]

2dgame background pixel sky cloud

[nimble mason]

Here is the image you requested.

[fallow vessel]

what does this option do? it appears to have no effect on my clip skip slider usage.

[royal monolith]

eyeball teddy train, eyeballs falling, eyeball rain, ancient sad eyeballs

[celest sigil]

I guess it doesn't know Thunderbirds very well...?

[royal monolith]

doesn't know a lot of the old kids shows

[celest sigil]

It got the font correct though 🤔

[royal monolith]

what model are you using?

[celest sigil]

I'm trying several Flux variants

[royal monolith]

oh. well - yeah, not surprised flux doesn't know.

but it knows some. it knows there are things that fly

[celest sigil]

PixelWave Flux

[royal monolith]

elmobirds?

[celest sigil]

It knows the slogan too

Here's another 😄

[royal monolith]

that's ... i have so many questions...

[celest sigil]

I have no answers 🤷🏻‍♂️

Only more questions!

That was also a Magic Roundabout prompt.

[royal monolith]

what's magic roundabut?

[celest sigil]

An old kids TV series

[celest sigil]

This one it knows!

[rain gazelle]

Nice 🙂

[rain gazelle]

I like this one, too.

[clever oar]

Netflix film for 18+

[rain gazelle]

Tried a couple. I like that bottom one. 😛

[clever oar]

good game

now i playing in enternal

[pale steeple]

guys, where can i learn nsfw art?

[languid pebble]

Youtube or Maybe better: CivitAI

[pale steeple]

is there anywhere i can discuss it?

discord channel?

[languid pebble]

not sure whether it's forbidden to discuss that ... but for more learning a chat isn't really the best place. We are not paid teachers. We help each other a bit ...but this is not a course

[vagrant dust]

Try unstable diffusion discord. But avoid the nsfw niche channel 🙂

[pale steeple]

any link to it?

[royal monolith]

just search google

[pale steeple]

what is this blur

or what's in the background image

how cna i clear this up?

[royal monolith]

the blur in the background or something else?

[pale steeple]

yeah

using pony realism with comfyui

why do i generate these blurs when i am using pony realism?

[candid pollen]

I had a similar problem when using a "wrong" sampler (IIRC DPM++ 2M) with pony realism in A1111. Changed to SDE sampler and it worked. Changing to a different checkpoint (Cyberrealistic pony) also worked.

[pale steeple]

i am working in comfyui and don't have sde sampler

[royal monolith]

try something other than euler and karras

[pseudo turret]

Paint a picture of sunset and lone ducks flying, autumn water together in the sky

[gaunt ruin]

Children's picture book, in the style of Hayao Miyazaki, forest, small animals, little boy, night, starry sky

[nimble mason]

Here is the image you requested.

[nimble mason]

Here is the image you requested.

[languid pebble]

Quantum physics good morning coffee

[celest sigil]

Another firefighter gone rogue

[hidden whale]

using which model this was generated?

[untold zodiac]

boys lets say i wana train a pony lora on the ben 10 omnitrix badge

i want it to learn the logo and the body suit styles

and i got 2 folders one with 88 images and the other with 250 images

how many repeats should i use for the 88 images one?

like this

body suit and the badge itself

[celest sigil]

Photo op!

[steep sleet]

"Now, if she could only turn those into water cannons!!!" 🥳

The two spires I mean! 😉

[steep sleet]

Cut 'em up, she can!

Flux.dev.fp8, Wraith_BW LoRA

First date

[celest sigil]

Last date

[steep sleet]

Surgeonettes

[royal monolith]

quoth the raven, "never more!"

[languid pebble]

Do not open the box!

[steep sleet]

Flux and Silhuflowart2 LoRA

[cloud sorrel]

which sorcery is this? which model can handle text so well? 3.5?

[languid pebble]

Flux ...

[cloud sorrel]

nice

[languid pebble]

3.5 should be able, too. But I am more used to Flux

[cloud sorrel]

i am stilll stuck to xl and lightning, i guess i should try some new models, but i am afraid it will be to slow on my mac

[languid pebble]

I have no idea about Mac, but for Flux there are already a lot of smaller and faster versions ...

[cloud sorrel]

how much longer flux iteration takes compared to xl with same sampler?

if for example you have 10it/s with xl and 5it/s with flux, i can guess what i could expect

[languid pebble]

I never really cared ... I'm going for quality. But I am sure you can find info about that using google

[cloud sorrel]

i will try anyway, i just wandered if you know

[languid pebble]

😄

[cloud sorrel]

when your mac costs like 2 decent gaming pcs, and you have 3-5 seconds per iteration you want to break it 😂

[languid pebble]

My Workflows are damn crazy using Super Flux with Detail Daemon and my own Lora ... it's not really fast here ...

[languid pebble]

Just give it a try ... it's not that much work.

[royal monolith]

just say no to macs?

[cloud sorrel]

i need mac for work

[royal monolith]

your mac got nvidia gpu?

[cloud sorrel]

there is no mac with nvidia

[royal monolith]

just say no to macs

(at least for AI)

[languid pebble]

I don't know the M4 ... I wonder how they can share only 16GB ....

[cloud sorrel]

i have m3 pro with 18GB of unified ram, and it is slow

[royal monolith]

what you need, is an nvidia GPU with 16 gig VRAM

system ram shouldn't be used for this

and an integrated GPU is always going to be slow

[languid pebble]

Macs are also used for video cut and so on and they can handle that .... I think they have a special architecture ...

[royal monolith]

that's video. but the current AI technology is all written for nvidia, python, and cuda

macs/apples were always the leaders in graphics, but this is totally different

you're just getting graphics as the end result

[languid pebble]

I have never tried that, so I have no idea whether there aren't good solutions for Mac, too

[royal monolith]

the funny thing is, you can run stable on an iPhone, but not an android. yet you don't want an apple product for your desktop for it

[vagrant dust]

Even the models converted for the apple ANE (neural cores) are not faster. But if you have a notebook which can work 8-10 houes without a power cord compared to 400w consumption there must be a difference 🙂

[royal monolith]

yes, but that's hardware, not software

[vagrant dust]

But the shared memory is quite nice for the the larger llms with 96 gbyte shared "(V)RAM".

[cloud sorrel]

m2 ultra with 192GB of unified RAM is insane

[royal monolith]

but slow

[cloud sorrel]

its faster than my m3 pro with 18GB

[vagrant dust]

Yes it is but still even with the actual torch version with support for the mps backend of apple you won't be fast. Even if you use it for parallel processing (8 images at the same time...) the GPU Cores are limiting the performance

[cloud sorrel]

more unifed ram is like having more vram on GPU, so it works faster

since 2/3 are reseved for vram

2/3 of 192 is 128

so 128GB for VRAM

[royal monolith]

but it's not an nvidia gpu, so ...

you're just going to be slow

"An Nvidia GPU is preferred for AI image generation because of its specialized "Tensor Cores" which significantly accelerate the complex mathematical calculations needed to run deep learning models, allowing for much faster image generation compared to a standard CPU, making it ideal for generating high-quality AI images quickly; essentially, Nvidia GPUs are designed with parallel processing capabilities that handle the massive data processing involved in AI image generation very efficiently. "

as opposed to normal graphics work

[steep sleet]

... just not long strings of text?!

[royal monolith]

depends on the text

[steep sleet]

Llanfairpwllgwyngyllgogerychwyrndrobwll-Llantysiliogogogoch

[cloud sorrel]

wtf is this?

[royal monolith]

going to scotland?

[cloud sorrel]

i cant click on this

[royal monolith]

where'd you get that from?

[cloud sorrel]

civitai

and if i click on that nothing happens

[royal monolith]

ask civit then

their site is always glitching

[cloud sorrel]

i found it under anouncments

but click on notification does not take you there (as it should)

[heady lodge]

What's the best civitai model for monster design?

I'm looking for one that pertains to the designs of silent hill

[steep sleet]

Yum This Pumpkin Spice syrup is just what I need in my egg-flip!!!

[cloud sorrel]

take your pick

[heady lodge]

most of these are just images of people walking through a spooky fog

[royal monolith]

it tried

[royal monolith]

why do you need a lora? all the AI base models do monsters just fine with just a prompt

[steep sleet]

Its the name of a town in North Wales

[royal monolith]

sorry, i did know that. not sure why i thought scotland

[cloud sorrel]

and what about non english characters?
СРБИЈА
ČĆŽŠĐčćžšđ
예쁜 여자
漂亮的女孩

[languid pebble]

The people ther just call it: Lianfair ... that's what's a commercial says here

[royal monolith]

go run some through sd3.5 and see

[cloud sorrel]

i am asking about the flux

but same aplies to 3.5

[royal monolith]

okay well go run it the non-english characters through flux and see?

[languid pebble]

AFAIK Asian Characters it's more hallucinating ...

[cloud sorrel]

thanks, any idea about non english Latin characters? or Cyrillic ones? same as for Asian?

[royal monolith]

since the Ais are trained on english text, probaly no non-english is going to work well. you'd really need loras for those and i wish you lots of success

[languid pebble]

I just prompted an asian city, no specific charecters. I can't read them but some asien friends told me they don't exist

Maybe you have more luck prompting a specific text, but have to try ...

[cloud sorrel]

i am just curious, i have no specific need

[languid pebble]

There are some asian models, too ....

Made for prompting in an asian language so they should replicate that ....

[spring mist]

Hunyuan-DiT

[languid pebble]

https://docs.google.com/spreadsheets/d/1vViQh45B-txERc5wnDVh_aNfHWLuEIrOCz7_HGXpPZM/edit?gid=0#gid=0

[heady lodge]

I mean there's different types of monster styles I guess, I figured I might need a lora that captures a specific artstyle

[royal monolith]

you might have to train one for that

[spring mist]

might not also

[heady lodge]

I'll just use ponydiffusion

[barren minnow]

A cute girl with white flower in hairs.

[royal monolith]

read the information in this channel: #artisan-faq

[heady lodge]

What happens if I use the same seed but change the prompt a little?

[radiant grail]

a realistic HD image of inside of the store full of prams, cribs, and bicycles

[royal monolith]

it changes the image. that's the reason for locking the seed down, so you can tweak the prompt or settings.

[royal monolith]

you can't generate in this channel. read the information in #artisan-faq

[celest sigil]

Here are your images...

[languid pebble]

How you make em throw up?

[celest sigil]

https://tenor.com/view/gag-gif-17244252139129953341

[languid pebble]

Schrödingers Coffee

[opaque glen]

We lie down together amid the cacti. The succulent leaves closest to the ground are a marbled grey, as if turned to stone, and we become absorbed by them, feeling our way around their rounded contours with our fingers. As we gaze up, following their odd tear-shaped forms bundled together, the sprawling double cypress tree – two trunks locked in an embrace – claims our attention with its swaying branches splitting into ever finer branches and twigs ending, here and there, in clusters of cones. Initially, we can’t really tell if it’s the wind that’s causing the canopy to stir and sway that way. It forms a dark shifting frame that we enter and get lost in as one does in a forest.

[viral frost]

It’s just accidental. I’m retrying these prompts I had done months ago with Invoke, SDXL, and the nightmare prompt node, and every so often I find one that gives wild, varied results.

[steep sleet]

Flux Turbo LoRA w/f with 2 x KSamplers, 2 x Upscale and Sharpen

[celest sigil]

Finish him!

[celest sigil]

"The players are literally melting in this heat!"

It was at this moment in time that The Joker learned why Batman was known as a sore loser...

[steep sleet]

Flux Turbo + LoRA

[celest sigil]

Batman: "I'm the one that wears the cape...!"

[wispy nest]

I remember when I spammed batman images because it was the only human I could get the original SD3 medium to make accurately

[viral frost]

I have a 64GB M3 Max and use Flux. It’s a bit hard to describe the generation speed for you: one, because I use an adaptive ODE sampler for quality, but it doesn’t have a fixed number of samples and the time per sample is slower than normal samplers, and, two, because it’s been fluctuating up and down as I update PyTorch with new nightly versions. But it’s running as fast as I’ve ever seen it right now at about 6.5-7 minutes for a 1MP generation with bosh3. If I were to use DPM++ 2M, I’d estimate it would take about 5.5 minutes for 50 steps. I’m running the full-fat dev, not a quantization.

[wispy nest]

yeah speed comparisons can't really work because there are way too many variables

CUDA and torch version matter a lot

and then whichever combination of TensorRT, TorchAO, Torch.Compile, FP8/INT8 Matmul people are using for acceleration

or the opposite, whichever combination of GGUF formats is slowing them down

[royal monolith]

batspam... yummy

[young ridge]

Made a fake "Planet Earth" style doc using Midjourney, Runway, Suno and Elevenlabs... Check it out if ya like. I think it turned out kind of cool. Everything is AI except the text and the editing. https://vimeo.com/998739582?share=copy

[unreal dune]

"A promotional image for a tea shop located on a bypass road, named 'Milestone Cafe.' The cafe has a cozy and inviting interior, with large windows that provide a view of passing vehicles and greenery outside. A prominent 'Milestone Cafe' sign is visible, showcasing the logo and inviting customers from the bypass. Inside, warm brown tones, wooden furniture, and soft lighting create a relaxing ambiance. Tables and chairs are arranged for comfort, and shelves display various tea selections. The overall mood is welcoming,

I want a apple image

[hallow basin]

Are all this images made by ai

[celest sigil]

Are you asking yourself?

They look bad enough to not be AI, unless a layering workflow was used.

[celest sigil]

Urban camo

[wispy nest]

I feel like the model messed up here in term of logic but I like it more

[cosmic elbow]

This is quite an old SD generated image. SD has changed quite a bit since then and it is hard to replicate this type of camera angle I find and a figure so far away...

[wispy nest]

nice colours

[sly tangle]

Reposting this here, if anyone can help 🙏

[royal monolith]

just looks like an oil painting.

[sly tangle]

I mean the character style, not so much the medium. Is there an artist known for painting characters in that style

[royal monolith]

not as far as i know. to describe it, it's just an oil painting of a girl with brown hair. a fairly common oil painted look in fact

[sly tangle]

For example WLOP has a particular character style

[royal monolith]

no

even google lense doesn't think it looks like any thing in particular

[sly tangle]

I agree this looks super common, which is why it's hard to figure out, but I could've sworn there was a relatively popular digital artist that made characters/eyes shaped like that, and people were making loras styled after it. Maybe I'm hallucinating tho

[royal monolith]

you might be thinking of Margaret Keen, but hers were big eyed kids

[wispy nest]

there is some term for non-anime graphical digital novel or comic art but I forget the name

there are civit loras that use the term

ah I found it

on civit its called Western Comic or Western Animation

[proven wasp]

Samdoesarts is your answer

[sly tangle]

Thank you, that was in fact the name I was thinking of, but couldn't remember

[proven wasp]

[celest sigil]

It's actually not far off the prompt. She's transforming into data.

[wispy nest]

ah that makes so much sense

okay the model was right then

I thought you prompted woman in front of city

[coral spire]

画一只小狗

[languid pebble]

Good morning coffee

[steep sleet]

SD3.5L Turbo Llama3.2

[languid pebble]

"Today, we are adding new high-resolution capabilities to FLUX1.1 [pro], extending its functionality to support 4x higher image resolutions (up to 4MP) while maintaining an impressive generation time of only 10 seconds per sample. Higher Resolution, No Compromise in Speed FLUX1.1 [pro] – ultra mode: This option enables image generation at four times the resolution… "

[pale steeple]

what is the easiest way to extend an image sideways?
make it from portrait to widescreen for example

[languid pebble]

Outpainting AFAIK

[pale steeple]

how do you do outpainting?

[languid pebble]

There are webservices for that. I don't really do that often. I'd search at youtube ...

[vagrant dust]

Local tools like invokes canvas or comfyui support outpainting.

[clever oar]

👋

[royal monolith]

SD3.5 medium. prompt: a white man wearing a white tee-shirt and a black man wearing a green jacket running away from a small sports car. the car that is fairly far in the distance, in the background, is on fire.

[clever oar]

no mutation ?strange

[royal monolith]

mutation?

[clever oar]

extra hands and legs

fingers

[royal monolith]

no, none of that. SD3.5 large and medium do a very nice job of not doing that.

[clever oar]

when it first came out and I checked I had problems with it

[royal monolith]

you are sure you're thinking of sd3.5 rather than sd3-2b-medium?

[clever oar]

i use 3.5 large

[royal monolith]

maybe you're using different settings than i am. the sampler, scheduler, number of steps, and cfg will have a huge effect on whether you get those mutations or not

[clever oar]

you use it local?

[royal monolith]

yes. for that image, i'm also using skip layers. it's sd3.5 medium. sampler: euler_ancestral, scheduler: linear_quadratic, skip layers: 22,23,24 scale: 1 steps: 36 cfg 3.7 a man and woman gazing into each other's eyes. She is stroking his cheek, his hand rests on her shoulder, sunset.

[clever oar]

low cfg ,in sd 1.5 my sfg is 7

[royal monolith]

i never every go that high with cfg. ever. maybe 5 but that really is rare. yes, low cfg

and low steps but not terribly low

[clever oar]

you like creative ?

[royal monolith]

you restrict too much, you get issues

you might want to experiment. lock the seed down. lock everythign down, then do a series of gens where you just change cfg by one decimal place at a time, starting with cfg 3

and working up to 7

and then study them

[clever oar]

i dont undestand whats update im my sd , now i cant use ip adapter =out of memory -before i use it witout problem

[royal monolith]

are you sure you're using the IP adapter that was trained for the model of stable diffusion you're trying to use?

[clever oar]

how many ip adapter version?

[royal monolith]

ipadapter is a model. and just like all other models, like loras, it will only work with the base model it was trained on. you can not use IPadapter for SD 1.5 with anything other than SD1.5 base or SD1.5 checkpoints. if you want to use SDXL, you would need an IPadapter trained for SDXL. same with SD3.5 - you would need an ipadpater trained for SD3.5 and it's not out yet

[clever oar]

itr for sd 1.5

i use 1.5 models

it work before

[royal monolith]

are you out of disk space?

on your system drive

[clever oar]

low disk space

500 mb

you think it take many space for it?

hm

i dont check it

[royal monolith]

you have to have enough diskspace on your system drive for the system to run as well as quite a few other things, even if you have stable running on some other drive mostly

check your disk space, check your swap space, and if those are fine, post in the #🤝｜tech-support channel and give the exact errors

[clever oar]

i hope it help

[clever oar]

i will clean my harddrive

[hard quartz]

Any idea by anyone if there is any Image to Reflief/3D COnverter Model?

[shut galleon]

draw a paperwall with a hello kitty

[nimble mason]

cool as f image btw adreitz

[nimble mason]

Here is the image you requested.

[languid pebble]

Their last selfie ...

[clever oar]

alien robo evolution?

[spring mist]

Henry

[languid pebble]

Looks like ... time traveller 🙂

[clever oar]

I hope the method of reproduction has also evolved🙂

[languid pebble]

They just eat the earth and reproduce themselves

[clever oar]

What kind of coffee do you drink, instant or ground?

[languid pebble]

Senseo 🙂

Expensive but easy, tasty and I only have 1 cup a day

[clever oar]

capsules?

at one time I had a 0.5 liter mug and I drank one such mug of coffee every day in the morning🙂

[languid pebble]

No capsule made of platic or alu ... coffee in a paperfilter

[clever oar]

I’m designing my room in SD , it’s beautiful and interesting, but probably very expensive🙂

Denoising strength =50 .
if more then this is no longer my room

ip adatper still now work

sad(

[clever oar]

jewelry store on the face?

[languid pebble]

pretty expensive 😄 Old prompt translated for Flux

light on the nose not pretty ...

Using a special Flux Version only the trainer and I have and a technology to improve ...

[clever oar]

Mojo

[languid pebble]

Ohhh.... forgot my LoRa

[clever oar]

I recently listened to several songs made in suno, I really liked it

[languid pebble]

I have created some, too

[stuck totem]

i waifu-ized my elden ring character lol (using Sakimi-Chan flux LoRA and some inpainting, before and after)

[clever oar]

how long does it take to learn to do something good?

[languid pebble]

https://www.youtube.com/channel/UCaIN2kn0zofdQlHPeg15HIQ You also need luck 3 days maybe?

[clever oar]

some song like real

[languid pebble]

If you wanna use them commercial you need a subscription ... but you chatgpt analyse songs that fit to the style you want and use it ...

[clever oar]

subnautica

[clever oar]

no need ,just for fun

[languid pebble]

Give it try with free credits than ... should be enough to create 4-5 good ones.

[clever oar]

i cant use it every day>?

[languid pebble]

Than you will have a good idea how to use it

[clever oar]

one song

one song in day

[languid pebble]

If I remember right you start with 500 credits that will be renewed every month ...

[clever oar]

👍

[languid pebble]

So sure you can focuse on one song every day ...

[clever oar]

what you know about voice cloning, can i found something like elevenlabs for free?

[languid pebble]

Yes ... Search for RVC on youtube ... but didn't try that combination yet

[clever oar]

good result

tried the bark yesterday, the result was so bad

[languid pebble]

I have made a model of my voice and used it with a vnese song. It worked pretty good ...

[clever oar]

where you train it?

bark?

[languid pebble]

but if the voice in the song you wanna change has been manipulated it doesn't work good

[languid pebble]

RVC

[clever oar]

i read about it little bit

[clever oar]

good tatoo for night drive on bike

[celest sigil]

Be seen! 😄

[clever oar]

mass effect in jewelry store🙂

[languid pebble]

;-P

[steep sleet]

Ukiyo-e painting with Gustav Klimt influences. Beautiful fairytale princess discovers a big shiny, golden compass in dense, dark forest, gnarly trees, lush green vines and colorful flowers, full of magic and mystery. Dwarves watch in amazement. Sparkling light floats in the air, adding a sense of mystery and fantasy.

[haughty roost]

Trying out the use of generating landscape wallpapers in different styles.

[charred sun]

Hi, everyone 😀 .

@languid pebble Hi....

[languid pebble]

Welcome! Nice picture!

[charred sun]

@languid pebble Thanks, bro....🥹

[scenic yew]

bark is pretty bad now, there are far far better alternatives

voicecraft - best voice cloning ability but no streaming support and a bit slow(around 1x realtime on gpu)
xttsv2 - ok voice cloning ability, very very fast streaming(0.1 sec) and very fast(6x realtime on gpu)
styletts2 - ok voice cloning ability, no streaming support, but incredibly fast(8x realtime?)
fish-speech - good voice cloning ability, very fast streaming(0.15 sec) and very fast(5x realtime on gpu)

Also, you can look at glm4 voice, it doesn't support voice cloning but it can generate speech with multi-emotion(angry, sad, whisper) and different speech rate like gpt4o.

[charred sun]

@scenic yew Hey, bro..... I am new to discord . Can you give me link to any user manual to how to generate images in here ??? I am not going to lie.....this discord thing and chatbot staff are going over my head 🫨

[languid pebble]

https://discord.com/channels/1002292111942635562/1237461679286128730

[charred sun]

Service car of RoboCop in pinkish neon light 😇

@languid pebble A thank you gift for you, bro.....

@languid pebble Don't take it otherwise....but your name reminds me of a character, a cartoon character

[clever oar]

thx can i use something in your list online ?

[scenic yew]

There are many hf spaces for the models, just search them, you can try here: https://huggingface.co/spaces

best quality is usually locally run but its decent in spaces too

[clever oar]

I tried several spaces but the result was not very close

[scenic yew]

can you give the voice file? which voice do you want to clone?

[clever oar]

I tried to clone different prizedents😎

[languid pebble]

You need a voice without any background sounds ...

[clever oar]

i local install cogue tts

but not try clonning

i test it online

bad

[clever oar]

not close like elevenlabs😎

[wispy nest]

cogue tts is not close to elevenlabs yeah

assuming cogue tts means coqui-ai/TTS

[scenic yew]

can you give some voice file that you tested with xtts

[clever oar]

i delete it , maybe you show example

[clever oar]

how?

[languid pebble]

https://hailuoai.video/

[clever oar]

it works now with images?

[languid pebble]

Yes ...

[clever oar]

oh!!!

i dont know it

[scenic yew]

with donald trump

[clever oar]

not bad

tts-xtts2-multi i try this

[scenic yew]

let me try with xttsv2

[clever oar]

no close -and distortion, voice is not recognizable

maybe wrong speaker

[scenic yew]

xtts default settings are pretty bad, you need to change with them. You can make xtts even laugh and do sounds like ugh by changing settings. I'm not sure if this is the right channel to discuss about xtts though.

[languid pebble]

I am a nobody here ... but as long as there's no other deep discussion ... just do it!

[clever oar]

not worry there we ask about food😆

image or video or sound are parts of one whole

[languid pebble]

All about A.I.

[clever oar]

yep

there posting video

[reef atlas]

Why did you use exactly? I didn't really understand.

[scenic yew]

fish speech
hf space: https://huggingface.co/spaces/fishaudio/fish-speech-1
model: https://huggingface.co/fishaudio/fish-speech-1.4

[clever oar]

oh is very fast

but why on korean lol

oh this better and not support my lang((

[reef atlas]

I tested it with a sample of my voice, but the result wasn't convincing

[scenic yew]

you usually need 10+ sec for fish-speech, experimenting with the params can help too, Voicecraft is better for lower sec but the demo has a error right now.

[reef atlas]

Do you have some settings to recommend?

[scenic yew]

depends on the sample, but usually
set iterative prompt length to 0
set top p, temp to low if its a very long sample, and opposite if its shorter, should be 10 sec at least

try to keep repetition penalty 1.0 except if you get large spaces

[reef atlas]

Ok. Thanks.

[sterile kiln]

Xtts is really good y, i dev this with it (partially) :

[rain gazelle]

Tried running an animation with mochi. Stretches the capabilities of the 3060.

[quiet vine]

hi..any1 know which artist that has this art style? found this on pinterest and description said it used nijijourney

[languid pebble]

Try to search on civitai ...

[quiet vine]

😅 how do i do an image search on civitai? does it has feature like google image?

[naive pewter]

Black horse man fusion with anime effect

[thin grotto]

hello there im new and i have just one question for now, see i have made this character in forge UI with a PonyXL model and a bit of inpainting and now i want to make a character sheet or reposing her for a costum lora, i watched over 10 houres of tutorial but nothing worked for me. does anyone have a idea or a working workflow for cumfy UI (yes i have that too)

[naive pewter]

#Black horse man fusion with anime effect

Black horse man fusion with anime effect#|

[quiet vine]

how do i make the whole body instead of only the head to face the camera? what prompt should i add?

[languid pebble]

It sounds like the name of a model. You could search for that

[rain gazelle]

the sweatshirt came out great on this one

[celest sigil]

This one too

[rain gazelle]

the lady looks the same too

[celest sigil]

Same description 🤷🏻‍♂️

[rain gazelle]

"blonde lady, sonic sweatshirt, masterpiece, absurdres, greg rutkowski"

[royal monolith]

don't use greg unless you are doing sword and sorcery paintings. that's all he paints

[rain gazelle]

it was a joke 😉

[still rose]

What model and loras for cars been looking on how to do cars properly

[celest sigil]

Flux with bio-ink lora for the glowing patterns.

[still rose]

Bio ink on civit AI? Also one more question how is the prompting formula for cars with Flux

[celest sigil]

Yes, on civit. Just prompt for the car you want, it's pretty simple.

You don't need the lora for cars though.

[still rose]

Ok thx, I can specify certain on the car that I want different?

[royal monolith]

yeah, just use the specific make and model in your prompt

[still rose]

Thanks I’ve struggled with AI and cars that I’ve never tried to do it on flux

I will try now when I get home

[celest sigil]

A car with tuning modifications and sleek body-kit was in my prompt.

[still rose]

👍

I’ll try specific things on the car if that doesn’t work out all to well I’ll keep it simple like yours

[still rose]

True

[sterile kiln]

Is it Buldozer AMD architecture?

[languid pebble]

Not sure about that ^^

Maybe just a bug ^^

[royal monolith]

flings rotten tomatos

[still rose]

The headlights still annoy me how odd it looks

[royal monolith]

they don't look nearly as odd as some of the ones i see on cars on the road

[clever oar]

crystalwizard

clean space on hard drive not help🙁

[royal monolith]

:( post the errors in #🤝｜tech-support and maybe @dry crow can figure out what's going on

[clever oar]

i dont know what is error -its only out of memory

i was in tech support with this question

not help

[quiet vine]

what prompt should i add to make the person standing straight as if shes taking passport photo?

[royal monolith]

you could try adding passport photo to the prompt

[quiet vine]

im trying to make this pose but the body refuse to face forward 😅

heres another 1

[orchid garden]

Abdalfattah el sisi 3d character

#🏞｜general-with-images Abdalfattah el sisi 3d character

#artisan-1 cat

[heady lodge]

so this happened to me when i tried to generate a medieval warrior with "plate armor"

[royal monolith]

well - apparently he thought it was dinner time ;)

[heady lodge]

thanks pony

[earnest phoenix]

king

king god

King Charles III, formerly known as The Prince of Wales, became King on the death of his mother Queen Elizabeth II on 8 September 2022.

In addition to his official and ceremonial duties in the United Kingdom and overseas as The Prince of Wales, His Majesty has taken a keen and active interest in all areas of public life for decades. The King has been instrumental in establishing more than 20 charities over 40 years, including The Prince's Trust, The Prince's Foundation and The Prince of Wales’s Charitable Fund (PWCF).

His Majesty has worked closely with many organisations, publicly supporting a wide variety of causes relating to the environment, rural communities, the built environment, the arts, healthcare and education.

King Charles III, formerly known as The Prince of Wales, became King on the death of his mother Queen Elizabeth II on 8 September 2022.

In addition to his official and ceremonial duties in the United Kingdom and overseas as The Prince of Wales, His Majesty has taken a keen and active interest in all areas of public life for decades. The King has been instrumental in establishing more than 20 charities over 40 years, including The Prince's Trust, The Prince's Foundation and The Prince of Wales’s Charitable Fund (PWCF).

His Majesty has worked closely with many organisations, publicly supporting a wide variety of causes relating to the environment, rural communities, the built environment, the arts, healthcare and education.

King Charles III, formerly known as The Prince of Wales, became King on the death of his mother Queen Elizabeth II on 8 September 2022.

In addition to his official and ceremonial duties in the United Kingdom and overseas as The Prince of Wales, His Majesty has taken a keen and active interest in all areas of public life for decades. The King has been instrumental in establishing more than 20 charities over 40 years, including The Prince's Trust, The Prince's Foundation and The Prince of Wales’s Charitable Fund (PWCF).

His Majesty has worked closely with many organisations, publicly supporting a wide variety of causes relating to the environment, rural communities, the built environment, the arts, healthcare and education.

[celest sigil]

Above post is from a 6 year old, or a meaningless bot.

[opaque sleet]

I've tried new IP-Adapter for Flux, it's really great https://x.com/ShakkerAI_Team/status/1855962063165891069

ShakkerAI (@ShakkerAI_Team) on X

IP-Adapter for FLUX.1 is here on https://t.co/YcvIezXSsP. A new generation of AI creative tools is coming! Helping you generate images in any style you want. #shakkerai #ipadapter #flux

Twitter · Nov 11, 2024, 1:13 PM

[raven barn]

For some reason my prompts I’ve been using consistently for awhile drastically changed the art style it generates for the worst. Any idea why that could be?

[languid pebble]

New model? Different settings?

[celest sigil]

You must have changed something, give us a clue.

[quiet vine]

what cause the small glowing specks to appear?

[quiet current]

try using a different VAE

[quiet vine]

the only option in VAE is auto or none

[quiet current]

you can get other ones from civitai / huggingface / etc

VAE quick guide :
1/ What is a VAE.
It's a part of the stable-diffusion pipeline that encode/decode information from tokens to latent space and from latent spaces to pixels. Aka it transforms math statistics in pictures
2/ Where do I put my VAE ?

• VAE with .vae.pt, .vae.ckpt, .vae.safetensors extensions go into the models\Stable-diffusion folder
• VAE with .pt, .ckpt, .safetensors go into models\VAE
  3/ How do I use my VAE ? Three possibilities :
• Either you name it similar to another one of your model (eg : Anything-V3.0.safetensors + Anything-V3.0.vae.pt), by doing that it should automatically load the VAE when you load the associated model.
• You manually load your VAE by going to Settings -> Stable-Diffusion -> sd_vae and selecting your VAE
• You add an easily accessible VAE dropdown at the top of your page to quickly switch back VAE by adding sd_vae to your Settings -> User Interface -> Quicksettings list

In civitai you can filter VAE

[quiet vine]

thanks! that fixed the problem 😁

[celest sigil]

"Do you want to buy a snow globe?"

Now, at home...

[steep sleet]

3.5L Turbo

[raven barn]

@languid pebble after digging it appears this is the problem

My new code is using this site to generate the art. https://api.stability.ai/v1/generation/stable-diffusion-xl-1024-v1-0/text-to-image

The website generating my previous art is: The website I’m using: https://api.stability.ai/v2beta/stable-image/generate/sd3

I’m attempting to fix it now, will report back

[languid pebble]

Yes ... different models need a different type of prompting

With the latest models you can prompt more in natural language and more complex

[royal monolith]

because the newer models use t5xxl - but it can also cause you issues and run off down rabbit holes more easily

[languid pebble]

Works fine here 🙂

[raven barn]

Yeah, it works fine with my other app but my new app is having so many issues getting it fetch the api. Had to take a break from trying lmao

[mild tusk]

hoi i iahve a question since im not familiar
im using illustro and my image is blurry here my setup unsure if you guys would know how to fix this?

[celest sigil]

I don't use A1111, but make sure you have a valid VAE selected.
Change the width and height to 1024x1024 (presuming it's an SDXL model?).
Sampling steps to 25.
Hires steps to 20.
Denoise to 0.4.
CFG to 5.

[languid pebble]

Hangover coffee 😄

[languid pebble]

Cheeky robot ^^

[languid pebble]

So lazy that I prompt my signature 😄

[royal monolith]

good way to invite forgerys

[languid pebble]

They don't have my A.I. Model ...

[royal monolith]

they don't need it, forgers just need your signature

[languid pebble]

Nobody willing to do that with me 😄

[wispy nest]

trimmed it short to keep things sfw

sometimes the hands are not ok sometimes it works perfect ; ) its custom i2v

[languid pebble]

A.I. and hands ... even harder with video ....

[wispy nest]

i will try and retrain on bigger models that are based on flux , but it takes 6 a100 to train, thats 14$ / h

probably 80-100$ per lora.

[languid pebble]

Pretty expensive ...

[wispy nest]

Im betting on the fact that there are thirty people out there

[languid pebble]

Do you need them to get that cash back?

[wispy nest]

thats why im doing it in first place. $

[languid pebble]

Ahhh... a friend has made a Flux based model ... good luck for me nobody else has it local ...

[wispy nest]

lora or model?

[languid pebble]

Model ...

LoRa I can do local ...

[wispy nest]

nice

yeah flux lora can be done on 24gb vram

I am not sure exactly what you trained

did you fine tune cog

[scenic yew]

Mochi-1 is surprisingly good with hands especially for open source even with anime which its not trained for but obviously not near perfect

[languid pebble]

Pretty good!

[scenic yew]

This is probably a better view of realistic hands with mochi

[languid pebble]

But only 5 seconds and pretty blurry

[royal monolith]

the background is supposed to be blurry - that's how you can tell it's in the background, not right behind the subject

[scenic yew]

the blurry part is because of vae tiling, you can turn it off but it will use massive extra amounts of vram, there are quiet good gens, some I made and some other people, but yeah limited to 5 seconds sadly, and no image2vid yet. The devs are saying it should come before the end of this year though.

[languid pebble]

Can you tweak it reducing the FPS?

[scenic yew]

Yeah I think so, but that will just make it slower or faster.

[languid pebble]

SVD you can creaty with 3 FPS and render the missing frames later ....

If you are happy you can also SloMo it by 2 times

[languid pebble]

#🏞｜general-with-images message Tweaked SVD

[clever oar]

good night

[scenic yew]

Oh you mean like that, I don't think so right now. But you could probably use some other tools, and when img2vid comes, you could probably extend videos.

and yeah you can slowmo it too

[clever oar]

good night cofee

[languid pebble]

Nini dicordos

[languid pebble]

We'll see

[viral frost]

What a majestic... dog?

[celest lance]

Anyone have adobe illustrator and can just quickly resize some things in an image for me for 10 bucks?

[viral frost]

@languid pebble Good evening coffee

[languid pebble]

Good morning coffee 🙂

[fickle heath]

three-dimensional ball made of sheep fat jade in Hotan, Xinjiang, with a white and round png picture in the background.

[royal monolith]

read the information here #artisan-faq

[fleet goblet]

I wonder if I should do cog img2img

easily could with florence2

[fleet goblet]

welp

[fleet goblet]

'

[fleet goblet]

I think I have a good img2img setup now

I can use wd14 tagger, florence2 and ollama.

A digital illustration of a cute, anime-style girl with long, wavy brown hair and large, expressive purple eyes, smiling with a playful expression. the girl is positioned in the middle of the image, inside a cube-shaped light box, which is placed on a wooden floor in a dimly lit room. the background features a cozy living room with a dark couch and a rug, creating a warm and inviting atmosphere. the lighting is soft and warm, casting gentle shadows that accentuate the girl's features. the box itself is square in shape and has a cartoonish, kawaii style with a soft, pastel color palette, adding a touch of whimsy to the scene.

[royal monolith]

not a bad description. The original prompt, however, was a little shorter: a photographic 3D cube with an anime face painted on one side

[fleet goblet]

That prompt was auto-generated by your image.

using Florence2 Promptgen 2.0.

Then fed at 0.5 denoise.

[royal monolith]

right. i understood that and I think it did a really good job

want a couple more images to test it with?

[fleet goblet]

Sure. It probably won't be perfect with all of them.

This is with Pixelwave Flux finetune btw

And the Vit-L improved clip.

[royal monolith]

these should be somewhat complex for it. let's see how well it does

[astral osprey]

A minimalist rhino head in a faceted, geometric style with sharp angles and a prominent horn. The design uses bold, black shapes without a background to create an abstract yet recognizable silhouette.

[gentle girder]

A cozy bedroom in morning sunlight, warm sunbeams through white curtains, unmade bed with cream colored bedding, a sleepy cat stretching, coffee cup on bedside table, minimalist interior design, soft lighting, lifestyle photography style, warm color palette, high-end interior magazine aesthetic, 4k, detailed, --ar 4:5 --style raw --v 6

[languid pebble]

Good morning coffee

[humble vector]

add realism

[median jacinth]

A digital illustration of a cute, anime-style girl with long, wavy brown hair and large, expressive purple eyes, smiling with a playful expression. the girl is positioned in the middle of the image, inside a cube-shaped light box, which is placed on a wooden floor in a dimly lit room. the background features a cozy living room with a dark couch and a rug, creating a warm and inviting atmosphere. the lighting is soft and warm, casting gentle shadows that accentuate the girl's features. the box itself is square in shape and has a cartoonish, kawaii style with a soft, pastel color palette, adding a touch of whimsy to the scene.

[royal monolith]

read the information in #artisan-faq

[clever oar]

mass effect

[languid pebble]

Good morning coffee

[fickle fjord]

Good morning coffee

[winged stag]

Fortnite... Flux model is awesome.

[nimble mason]

what's fortnite

[clever oar]

Good morning green tea

[winged stag]

a childs game, there is a skin for arcane jinx.

[pure surge]

Author: @gleaming fox
MotionPrompt: $freefire_michael_jakson_emote
Command: </random:1272485059353640963>
Background: from video
Model: v2-turbo
Explore more features at viggle ai

[clever oar]

from flux to haluo

[sterile kiln]

I did this when Flux came out (that's me in the photo, looking chubby)

[clever oar]

I like how Kling makes it the most

[languid pebble]

5 Seconds ... ... ...

[clever oar]

before kling can 10 second

[languid pebble]

Just a fun thing ...

[royal monolith]

kling does 10 second videos

[languid pebble]

5 Seconds is more SVD .... yeah it can follow a prompt .... but 5 seconds isn't enough

[clever oar]

not now in free account

[royal monolith]

that's becasue you have a free account. you have a demo account, you only get a taste - they're not going to change that

[clever oar]

we have no choice, models with this quality are not available locally

[wispy nest]

sometimes mochi does ok

[clever oar]

The runway can video to video

https://www.youtube.com/watch?v=5Z5ytQwDPRE

[clever oar]

runway?

[royal monolith]

meta

[clever oar]

what you use promt or video or image

[royal monolith]

prompted an image and then told it to animate it

[dusty warren]

How do you use this type of open pose with already predetermined shape? What type of controlnet should I use? Directly using Openpose_full result in black image

[rain root]

Hello! I'm new to this server, although not new to Stable Diffusion. Would like to ask for advice or ideas if anyone has. I'll attach here a couple of pics to start with. This is a OC I have created and keep using for many different gens. I've trained a Lora (then re-trained, and trained again and again) and by this point I'm achieving really good consistency with her features etc. --> I don't necessarily need to prompt anything about her appearance, the lora usually handles it all. Or I can just promp something about hairstyle or expression etc.

I've kinda accepted that the results are somewhere between anime and photorealism. Like...quite realistic but you can tell it's not photorealistic. It might be that my lora is just making a sort of a bottleneck here, since the images used to train the lora were not photorealistic, the outcome with the lora can never quite get there. Right? I include in my prompts things like "photorealistic" or things like that. I am not complaining, but just wondering. Woudl you guys easily come up with any ideas on how to increase the level of realism, textures etc.?

I use flux and tend to use rather simple and short prompts. I can include one example below. Should I go for more complex prompting to achieve realism?
Photorealistic picture, cinematic style, a beautiful mature British woman. She's wearing casual outfit with white ribbed tank top, oversized flannel shirt worn open, stonewashed blue jeans, red converse trainers, large wooden beads necklace, bracelets, round sunglasses. perfect hand,HDR, intricate details

[nimble mason]

That third one has the tiniest tie I've ever seen 😂

[languid pebble]

Good morning coffee

[rain root]

Haha yeah. I guess if/when ladies wear neckties, they're not supposed to be very long. But still. 😄

[warped totem]

一只猫猫

[opaque saffron]

生成一个狗

[clever oar]

good morning 生成一个狗

[vagrant dust]

She wears a tie? Maybe the tie just appears smaller due to the rest of the composition 🙂

[scenic yew]

you should probably remove the "photorealistic picture, perfect hand" part and maybe the intricate details if you are fine with it and maybe add "shot from iPhone, shot in 2004"

should help a bit, or you can just apply a lora like this one: https://civitai.com/models/689192/aesthetic-amateur-photo-flux-dev

[ornate osprey]

Guys i need help please
do u guys know How to change anime art style without change the whole in AUTOMATIC1111, i mean like not changging the face, hair, costume, just like change the art style
like this art : https://www.instagram.com/animazing.art/
i try a couple method but not work

and do u guys have recomended setting like adding some tab VAE, or like adding some tab clipSkip.. Thanks btw

[rain gazelle]

use different models, loras, artist tags in prompts, etc.

[rich nimbus]

#1237460438229450772 a car

[rain root]

Thanks for the tip! I tried modifying the prompt but it didn't really make much of a difference. I think the lora is the reason; it's been trained with not-photorealistic images so it affects the outcome. Ah well. I'm okay with that.

[rain root]

😉

[ripe cedar]

I keep getting OOM when I try to inpaint on a1111 with SDXL. What alternate? I just need to be able to use grounding DINO with it

[royal monolith]

switch to swarmUI rather than a1111?

[ripe cedar]

Ok I'll check it out

[icy leaf]

stable foundation

[nimble mason]

looks like it's taking a leak

[languid pebble]

coffee machine drone 😄

for the real coffee to go ...

[languid pebble]

2 mouths to feed 🙂

[nimble mason]

thats Schlonky Bong, the forgotten nintendo character

[languid pebble]

I see the reasons 😄

[royal monolith]

prompt : A close-up of a 2D solid colored Cel shaded...

@wispy nest

[shell valve]

a cute girl'

[whole plover]

I meed a Various cartoon shapes of animal soft rocking horse