🏞｜general-with-images

[royal monolith]

why is that crazy?

[languid pebble]

Nobody knew them and they are better than SAI ...

[royal monolith]

they ARE sai. they are the original Eluther SAI scientists and developers, and the actual creators of stable diffusion itself.

[hearty violet]

hi guys, what upscaler is indicated nowadays? Is to make this imagge bigger and softhly... In this example I used 4x-UltraSharp but change a bit the original image, have someone better?

[languid pebble]

They have been ...

[royal monolith]

they are. stablity.AI was originaly put together withe a group of Eluther scientists and researchers - the team that is now black forest.

it's their code and tech

[languid pebble]

I'm playing with RealESRGAN_x4plus.pth now

[hearty violet]

I will try this one thanks ^^

[languid pebble]

But crazy what they can do without the millions of dollars now ....

[hearty violet]

here?
https://huggingface.co/spaces/Marne/Real-ESRGAN/blob/main/RealESRGAN_x4plus.pth

[royal monolith]

the model that is flux is the SD3 model we teased out last spring. they don't need to train it, it's trained.

and they have the money if they need it

[languid pebble]

AFAIK they are German ... not a lot of risc capital here ...

[royal monolith]

they're still friends with emad. and everyone else

[languid pebble]

Shure ... but it's a bit crazy why SD3 is such a fail and all the good pics you can see here or on Civitai made with flux ....

[royal monolith]

SD3 is not a fail. what we tried to tell this community a number of times, an dwhat this community ignored, is that 2B medium, which was open sourced, is NOT the version of SD3 they saw teased out. that it's a low end model, that it was released to them specifically to ensure people that stability.AI is yes, interested in open source (this community was full of derogatory remarks about sai not open sourcing sd3 already) and to give them something to play with. that it is beta, and unfinished.

and they are still working on the model that is to be the actual release.

[languid pebble]

I'm sorry ... but a release is a release and no beta ...

[royal monolith]

lykon and alex BOTH tried to tell people that a lot of steps were skipped jsut so the community could have what they were demanding. the model was never intended to be something people would go out and train and go into production with. but this community, as always, just called them names, called them liars, and ignored what they said.

[languid pebble]

I never got that point ....

Can be my fault ...

[royal monolith]

really tired of this. it's beta. people were told that it's beta. it was last minute decision by management not to use the beta flag and they very quickly admitted that was a mistake. and that was made public too

but EVERYTHING posted on this discord gets argued with, people are insulted, told that they are lieing, the developers writing the code are told they don't know what they are doing or what they are talking about

to the poiunt that they aren't even on this discord any more

and this one fuller than most

[royal monolith]

sorry for snapping. it's not your fault. you've never been anything other than reasonable

[languid pebble]

Well, I'm not bad with SAI ... I just have the few information I have ...

[royal monolith]

i'm one of the beta testers. i was the one that did most of the tease posts last spring. i did a massive amount of rapid testing on sd3 2b when it released and the devs have all those results

[languid pebble]

Everything OK. I'm always happy to get more information and learn.

[royal monolith]

i tried to post here how to make it work, i got flamed

at that point, i decided i didn't care if people could make it work or not

[languid pebble]

I'm hoping for a better SD3 ...

[royal monolith]

it's being worked on

however they're going to release it when it's actually finsihed and not when the community demands this time

[languid pebble]

Sorry if I sounded rude ...

[royal monolith]

you didn't, i was just letting you know they are working on it

did you not see me say that I am the one that made most of those posts? and i cherry picked nothing.

what was released was NOT that model

that model is now Flux

[languid pebble]

BTW why is you Flux so small? The original dev was 24 gig?

[royal monolith]

you don't listen, do you.

and this @languid pebble is eactly what i was talking about. you say something, and the members of this discord tell you that you don't know what you're talking about, that you're lying, etc

this discord is a toxic mess

[languid pebble]

I listen to everyone I think who is cleverer than me 😄

[languid pebble]

Don't care about it ... just do YOUR thing

[royal monolith]

and i'll try this again. the SD3 model that we teased out last spring is now Flux. it is exctly what we said it was. the SD3 2b medium model that was open sourced was never advertised as the other model. the community are the ones that 1. said it was and then 2. went ballistic with all sorts of baseless incorrect assumptions about it and the issues it has

[languid pebble]

But Flux is Black Forrest and not SAI?

[royal monolith]

1. stable diffusion, all models are open source 2. the code belongs to the team that is now black forest, they took it with them.

that's what they released as flux

[crimson dawn]

SAI was BFL until they all left

[royal monolith]

right, becasue what you saw on the site are from the LARGE model, not the 2b medium model. do you not listen?

[crimson dawn]

only the beurocrats and support staff remained.

[royal monolith]

there are devs there, but not the Eluther scientsts and researchers. they're all black forest now. SAI has always been an incubator

[languid pebble]

It's a story ... hmmm ... doesn't sound good

[royal monolith]

ye it can and yes it did. and i'm starting to see why you got baned on bananadoco and other places.

no you didn't. you don't have access to its weights

[crimson dawn]

incubator is a good way to define the legacy. it's been a ride. but the kids have left the nest

[royal monolith]

that's what SAI has always been - kind of like y combinator

you're a great fit here.

[crimson dawn]

like "Aviato" 🤣

[royal monolith]

anyway, @languid pebble - you can DM me if you want other information

[wispy nest]

I need to explain
you get confused between the models sometimes
you trained the 2B param model
which is 10 GB in size when both clips and T5 are included
this is still known as SD3 medium, not SD3 large which is 8B param

the 8B param model, with both clips and T5 included, would be larger than 10GB in size

[royal monolith]

he won't listen to you, he'll jsut tell you that you're lying, and then get argumentative and puff out his chest

[wispy nest]

the weights for the 8B model were never released
only the weights for the 2B were released
the versions you have been seeing are the following:
2B only
2B with both clips
2B with both clips and T5

the 10GB one, which you sometimes refer to as the large one, is 2B with both clips and T5

[royal monolith]

see? he doens't listen. he's one of those that has decided incorrect facts in his mind and to him they are the real facts.

don't waste your time on him

his channel is also sus

if you actualy knew what you were doing, you could get the same high quality images out of SD3 2b medium that you can get out of flux. that you can't says volumes

[crimson dawn]

you can use it on the api

[royal monolith]

odd - i could swear i saw you telling people to watch your videos

[crimson dawn]

https://www.youtube.com/@Dice_Ai_Development

[royal monolith]

based on what you've said in this discord and this conversation, i'm having a hard time believing the '10 years' of experience coding AI

[crimson dawn]

https://youtu.be/2XPNLM9_IYo this is funny. you're looking at the cc value thinking they're using an api. dude they just use a regex

big hero 6

[royal monolith]

considering that stable diffusion only released 2 years ago, what were yo ucoding before that

[crimson dawn]

#🆕｜sd3 message

[royal monolith]

that's got nothing to do with 'what were you coding before 2022" that is AI related

[crimson dawn]

you ended up hammering those people's sites, just for them to see a bunch of junk in their queue lol. The credit card declined message comes up always. they're not processing anything

it's literally just a catch all wide net. the testing for what they got comes later

if you gonna hustle the hustlers you gotta think like a hustler. remember, they're lieing to you about everything

so boring

[royal monolith]

so i take that to mean you didn't actually code anything AI related before 2022, so you don't really have 10 years of experience coding AI

what actual coding experience do you have?

[crimson dawn]

when you roleplaying but your dice are smooth.

[royal monolith]

D100s are actualy like that

[crimson dawn]

my highest is a d60

[wispy nest]

wow nice mine max out at D20

Kolors prompt: photo of a D100 dice

[royal monolith]

D100

(real d100)

[wispy nest]

as usual Kolors is pretty but not very smart

[crimson dawn]

proper die with good labels is hard for flux rn too

[wispy nest]

I tried to translate into Chinese
Yībǎi miàn tóu de zhàopiàn

looks like I triggered the unconditional

whoops

ah yeah this is a different seed its just the unconditional

[royal monolith]

that could be an interesting double exposure composite

[wispy nest]

the vast majority of the time I prefer unconditional outputs to people's conditional outputs TBH

I love the soft dream-like look

[pale steeple]

any ideea why these errors occur?

[steep sleet]

Dice being dicey again? 😄

[languid pebble]

Good morning cocktail ... 🙂

[unique condor]

Wasn't he that dude that was scaring people with old doxxed IPs (old restorecord data leak data from a while back but might still have regional enough data for course geolocations) the other night when he was getting roasted by people calling him out on stuff?

Surprised the mods here didn't ban him

[languid pebble]

Not really what I was trying to get, but cool 🙂

[wispy nest]

they might be gone now
their comments are deleted and they don't come under search

[languid pebble]

Sounds like I missed something. But I am not into drama ....

[steep sleet]

He can be quite the fraud! However, he is extraordinarily helpful 🙂

[languid pebble]

Hungry? 🙂

[wispy nest]

I accidentally made R2D2 cooler

original in the back

also a cursed version

[languid pebble]

A strange mood in these pictures ... diffused light?

[viral frost]

bizarre unbelievable action photograph of a very damaged gricksack screamingoleakes action!!! photographily chaotic noodles!! extreme chaoticism clifcracks diagrams views and the like great display cg so; sh could at the same it-boobs time be (moreussian:1.5) high resolution!? makes small potatoes fuzzyouls realistic lighting horrific scary digital rendered 3d diagram titleackle*( best quality technical 1937 photograph )); intricate sloppyooky carnage with haunted quiche fight island

[wispy nest]

the mood comes from FreeU

[viral frost]

bizarre unbelievable action photograph of a collage impossibly dressed scary ape enjoying the felment spagdog futurism concept artwork working on extremely very near libertarian aboul Chachi, realistic human space exploration scifiornm nubbins, orrery anti giallo penna, cosmic jv herebot, crunchwrap

[wispy nest]

these ones are more extreme

[viral frost]

bizarre unbelievable action photograph of floating cavorting seaweed spagdog shifting gears from film grain, cinematic lighting!!! nadvacks!! alien bagpipes boogler and skeletonswers decrashionism garmonwarskues (sloppy drippuc:1.1), intricate controller turnchaangles!!!!xffmj

bizarre unbelievable action photograph of a great, free,umpy Trump jebaitingobitansonweaking Prort Troel from the no-talent giant library, cosmic, happenstimely, sloppy marinara etench Super Sleepawayandy, cosmic, de-burbed,arnold, dusty

bizarre unbelievable action photograph of the big tightropealingaman anthe warrior, wars thatGivingover with p raldines jvslr cinematicastic popeyes hat vtuber stanky bonegoat, chaotic lighting, d.nyne clinicalcoloredmmx, finely detailed render, cocoaene shreeking macro [ resetchpanson photograph]

[sturdy pewter]

which thumbnail is best guys for flux lora training tutorial

[languid pebble]

There are some good ... 🙂

[wispy nest]

I think this one
the ones with split colours for the text are nicer

[languid pebble]

I like on the left side the third from the top ...

[wispy nest]

ye that one is nice too

[languid pebble]

Took the red pill?

[sturdy pewter]

ty so much

[wispy nest]

Star Wars if BB-8 ate all the food

really wish star wars had giant thick BB-8s like this rolling around

[brisk knoll]

BB the Hutt?

[unique condor]

I saw the text before the image and in a split second I thought I was going to scroll and see some dummy thicc meme version of bb-8

[wispy nest]

lol

[tired nymph]

league of legends sion smoking fat cigar that looks like q with green beany reage style
would somebody do it for me?

[wispy nest]

To those that dont have 24gb vram

[nocturne oak]

https://tenor.com/xdsm.gif

[keen echo]

it looks like this with every model and prompt ive tried so far, just a different texture

this is what a forrest looks like, its all so dark and just looks like an old painting

[languid pebble]

I'm coming in peace ^^

[gentle blaze]

Rothschild family

[languid pebble]

The last tree 🙂

[pale steeple]

can anyone link me a comfyui workflow for img2img for flux please? would really appreciate it!

[steep sleet]

Cannot find these nodes in Manager 😦

[languid pebble]

Let me check ...

Try to update all and search again

[steep sleet]

I have d/loaded each node - in custom_nodes folder
ComfyUI is not picking them up

[languid pebble]

I've installed missing nodes via Manager .... have you restartet your A.I.?

[steep sleet]

After updating all, they still do not appear in Manager

[languid pebble]

Strange ...

[steep sleet]

I search for tensorop - not there
Search for Florence2 - also not there

[vagrant dust]

I guess in your config.ini (custom nodes > folder from manager) the security level is set to normal.

[steep sleet]

Attempting a manual install via the URL "You do not have the level of security to do this operation" is the error ... ?

[vagrant dust]

This leads to the situation that some custom nodes do not appear on the list

Either you change the security level (and increase the risk of using malware custom nodes) or you install the nodes manuel with the shell.

[steep sleet]

As admin and non admin - same message? I wonder if its because I need a Forence2 key?

[steep sleet]

Where would I change the security level? CMD as Admin?

[vagrant dust]

As mentioned in the config.ini file within the custom node folder of the manager. You won’t need admin etc. for this.

[vagrant dust]

strong
doesn't allow high and middle level risky feature
normal
doesn't allow high level risky feature
middle level risky feature is available
normal-
doesn't allow high level risky feature if --listen is specified and not starts with 127.
middle level risky feature is available
weak
all feature is available

[steep sleet]

OK - config file!

[languid pebble]

Thanks Lord ... I didn't know that 🙂

[vagrant dust]

So either normal- or weak should allow you install. But remember to change it back as there are some repositories out there with malware nodes.

[languid pebble]

TBH I never changed anything there ...

[steep sleet]

Config.ini reports security_level = normal

Change to weak?

[vagrant dust]

Yes if you want to install manually via github links

[steep sleet]

OK, it has now installed

I shall change security level back to normal later - thanks for the heads up!

[steep sleet]

Florence2 i2i 1st image

🥳

[final tapir]

Can anyone help me how to generate something like this?

[steep sleet]

Florence2 i2i 2nd image

[pale steeple]

thanks 🙂

[languid pebble]

Got it from Dice ... it's flux schnell

[steep sleet]

Florence2/Flux i2i

[steep sleet]

Flux i2i w/f - KJ Nodes "Get Image Size and Count" - its not wired in!
Where would one place it at all?

Great Eggspectations

[languid pebble]

Strawberry tropical? 🙂

[steep sleet]

Florence2/Flux i2i

[viral frost]

bizarre unbelievable action photograph of the cannelloni and bones confused scene beneath a cloud Numa N laughing agent has extremelyoleyepisode trotting stanky gibs, huge pilesare out trees: caustic old pond biff humor samples fast colors! neon paint everywhere intricate!! delicious : close up sharp focus texture jj abog day!! something gobs! something beasted! intricate

[jovial tiger]

here's a challenge, is anyone aware of a model that can do the middle finger? had a friend try it with my stuff and flux/sd3/sdxl couldn't do it. seeing how uncensored so many of these models are, it's weird that we couldn't find one that could do it.

[wispy nest]

mesh_graphormer, also known as HandRefiner is a rly good hand control net in general

but it wasn't trained on middle finger either most likely

you need a diffusion model, a control net, and a detailer model set (some combination of stuff like yolo and segment anything)
to all know what middle finger is, for this to work well

cos if you take juggernaut, mesh_graphormer and Impact pack hand detailer you can already get rly good hands

reason I say juggernaut is they simply trained it on a lot of hands lol

according to some old reddit comment they made

[soft oak]

Help me generate a master image

[shell sleet]

mmmmmmmmmmmmmm thatsa fail

[lavish garden]

Prompt: https://civitai.com/images/26765873

"Revelation" - Kai Ran (oc) | Pre-timeskip

[gleaming night]

https://media.discordapp.net/attachments/1158054263457595564/1278955153206022188/00000.png

orb.

hmmm just realized the shadow is bad

this was something i made in like 20 minutes really quickly

I'm a beginner

[languid pebble]

Good morning ... ähm ... drink

[steep sleet]

Florence2/Flux i2i

[slate owl]

is this kling?

[languid pebble]

Stable Video Diffusion ....

[thick flint]

Flux

[steep sleet]

Florence2/Flux i2i

[dry blaze]

Makes me wanna stick to 5 or less training images and not 11 lol

[languid pebble]

Not so creative today 😄

[dry blaze]

Think i'm getting the hang of lora making lol

Made a snake tongue

[nocturne oak]

YOU DIED.

[sterile kiln]

me as a sith x)

[nocturne oak]

Your reign of evil is over, sith.

[sterile kiln]

Well, it's time for a fight

[sterile kiln]

Did you make a Lora of yourself too?

[nocturne oak]

Yeah. 🙂

[sterile kiln]

hahaha

[nocturne oak]

That's why I had responded...had to. 😄

[sterile kiln]

working well?

[nocturne oak]

Yup.

[sterile kiln]

same, very, very similar to

[nocturne oak]

I've got an image brewing for the next response. 😉

[sterile kiln]

and i have an idea 😄

i do a versus with jedi and sith, and with inpaint...

[nocturne oak]

Oh, this image is coming along well...

3rd phase rendering now.

[sterile kiln]

i use Forge, really prefer A1111 style

[nocturne oak]

To each their own. No hate.

Have to run a small inpainting to fix a hand. Not surprised. 😄

The image is otherwise too good not to use.

[sterile kiln]

yeah, with flux it happens a lot less often though

[nocturne oak]

No doubt.

Ugh...so close...2 inpaints not quite right.

[sterile kiln]

trying to have 2 blades clashing, really hard

[nocturne oak]

This will have to do.

[sterile kiln]

bjorn ironside jedi x)

prepare to inpaint, i have something

inpaint your side, and then I'll try to animate it x)
@nocturne oak

used this prompt to inpaint, if it can help :
face of a man, a sith lord, the luminescent light of a red lightsaber illuminates his face, and he stares fiercely at his adversary, <lora:$ebl4_Flx_v1:1>

[nocturne oak]

Give me a short bit here.

[sterile kiln]

?

[nocturne oak]

A bit of time to get the inpainting done.

🙂

[sterile kiln]

ah x), 👍

i will prepare next frame, when you die

obviously

[nocturne oak]

What guidance are you using?

[sterile kiln]

guidance? sorry i'm not english

[nocturne oak]

There's a guidance setting for the prompt in flux.

[maiden axle]

#✍🏼｜rules-and-tos شخص رهيب

[sterile kiln]

for the base image? (I made a base image without LoRA to be more precise, then inpainted it with my LoRA)

on the left, a Jedi knight with a blue saber, blond, short hair with a thin braid at the back of the head, dressed in traditional Jedi garb, fights, and stands lightsaber to lightsaber with, the 2 blades clash, on the right, as a Sith lord with a red lightsaber, long black hair, beard, dressed in traditional Sith garb, The 2 have closed faces and are in fighting stance, fiction, in Coruscant a futuristic megalopolis with “Star Wars” movie architecture with gigantic buildings, spaceships by the hundreds flying in the sky in the background, stormy weather, dark clouds, volumetric shadows

[nocturne oak]

Guidance is a numerical setting linked with the prompt itself.

[maiden axle]

#👥｜roles

@nocturne oak

#📣｜announcements

#✍🏼｜rules-and-tos photo for test

[sterile kiln]

on the left, a Jedi knight with a blue saber, blond, short hair with a thin braid at the back of the head, dressed in traditional Jedi garb, fights, and stands lightsaber to lightsaber with, the 2 blades clash, on the right, as a Sith lord with a red lightsaber, long black hair, beard, dressed in traditional Sith garb, The 2 have closed faces and are in fighting stance, fiction, in Coruscant a futuristic megalopolis with “Star Wars” movie architecture with gigantic buildings, spaceships by the hundreds flying in the sky in the background, stormy weather, dark clouds, volumetric shadows Steps: 30, Sampler: Euler, Schedule type: Simple, CFG scale: 1, Distilled CFG Scale: 3.5, Seed: 2275577246, Size: 1152x896, Model hash: bea01d51bd, Model: flux1-dev-bnb-nf4-v2, Denoising strength: 0.5, Hires upscale: 1.55, Hires steps: 40, Hires upscaler: Latent, Version: f2.0.1v1.10.1-previous-480-g3b9b2f65, Diffusion in Low Bits: Automatic (fp16 LoRA)

I don't know if this will help

[maiden axle]

#🏞｜general-with-images

[nocturne oak]

hrmm...no. But I'll keep working on it to see if I can get it matched up.

[maiden axle]

#🏞｜general-with-images san in morocco

#🏞｜general-with-images/san

#🏞｜general-with-images@san

#🏞｜general-with-images #san

💯

[dry blaze]

@tropic shell Hoi, might need assistance here

[maiden axle]

Imagine/ san

[maiden axle]

#newbie شمس

#newbie

[sterile kiln]

الرئيسيات

[maiden axle]

كيف ابحث

من فظلك ساعدني

كيف ابحث عن صورة مثلا

@sterile kiln

#✍🏼｜rules-and-tos photo de san

[sterile kiln]

t'as pas l'impression que tout le monde parle anglais ici?
Vas spamer tes merdes ailleurs

[maiden axle]

Ok tnx

Can u help me

[sterile kiln]

start by learning to write. And ask for mp.

[maiden axle]

@sterile kiln Vous devez être soigné le plus tôt possible

[nocturne oak]

Almost there...small adjustment needed, but the head/face is in.

Here we go.

[sterile kiln]

HAhaha! good 5am teamwork bro

[nocturne oak]

Consider the fact that using AI, you and I have now faced off in the Star Wars universe.

[sterile kiln]

small upscale :

[maiden axle]

Help

#help

Help/

[nocturne oak]

Very cool. Fun stuff to combine the self-LoRAs to create something like this. Took minutes to do what might otherwise take hours or days before all of this.

I put it into my keepers folder. 🙂

[analog panther]

@dry crow help me

no matter what UI i use , i get this

[dry crow]

Come to #🤝｜tech-support

[acoustic trellis]

hola

[deft finch]

a futuristic cityscape at sunset with flying cars

[languid pebble]

Good morning coffee!

[steep sleet]

^..^<

Forge

[sterile kiln]

👍 🙂

[clever oar]

🖐️

klingai 99% problem

wait 2 days

[clever oar]

mojo it work for you?

[languid pebble]

Never tried cause I don't speak Chinese

First date ... 😄

[clever oar]

its every lang...

[languid pebble]

Whenever I wanted to do text2video it switched to chinese ^^

[steep sleet]

Forge img2img

[sterile kiln]

Guys, Forge now works well with Q8 + T5 Q8 + LoRA ❤️
Ty lllyasviel
https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/1634

[languid pebble]

Cool!

[crisp halo]

https://tenor.com/blYDH.gif

How does the tool work, how do you make art?

[languid pebble]

Have a look here: https://discord.com/channels/1002292111942635562/1237461679286128730

[crisp halo]

thanks

[wispy nest]

stable drama
https://preview.redd.it/did-anyone-catch-the-full-tweet-deleted-before-i-could-read-v0-klce6zhuq1md1.png?auto=webp&s=3c8aa2a333c3db0e1fa8f57bb8c3090fbc78868e

[languid pebble]

1.5 is so yesterday ....

Flux is today ...

[languid pebble]

Hi @discordos

[clever oar]

HI @languid pebble

[languid pebble]

Nice to meet you here!

[clever oar]

CogVideoX-2B block my promt

[languid pebble]

Bad censorship ....

[clever oar]

high level of censorship

cutting his beard is bad words for him

bananabot

[languid pebble]

Need to sleep ... good night!

[crystal flower]

https://civitai.com/images/1098364

what am I doing wrong?

[crystal flower]

Is there a checkpoint and lora that can create an image like this?

[rough jackal]

tell me, I installed
flux1-dev-bnb-nf4 but all the pictures are blurry. The image is visible, but it's not clear at all. What else have I not installed?
Forge with CUDA 12.1 + Pytorch 2.3.1 also installed
4070TI Super

[languid pebble]

Wrong VAE?

[rough jackal]

There is no VAE at all

[vagrant dust]

Looks like low step count. As it seems to be the dev not the schnell model you would need 20 steps

[rough jackal]

and in which folder should vae be installed?

[languid pebble]

Models/VAE depending on your UserInterface

Some more information here: https://tensor.art/articles/760080809324596639

[rough jackal]

flux1DevVAE_safetensors.safetensors, Sampling steps 20 last step turns everything grey

[sterile kiln]

What are you using? Forge? Comfy with a workflow?
Models+Encoder?

[brisk knoll]

yay, not only did i get a good steady workflow for SD3, but i fixed the session persistence in the Garage. Definitely a good weekend for dev work

[wispy nest]

this looks cool

[brisk knoll]

thanks, its on my GH if you want to try it too

[wispy nest]

I took a look, its a nice idea yeah

good to have things all together

[steep sleet]

Flux Pulp Sessions LoRA

Flux Toy Camera LoRA

[sterile kiln]

tough day for me

[clever oar]

advertising for presidential debates?😃

[steep sleet]

#Florence2/Flux + LoRAs

[clever oar]

man mistook dress for pants😃

[crisp stream]

more like HL 😄

[clever oar]

HL?half life?

[civic pasture]

what model and prompt??

[crisp stream]

WF is included in the image where you can see the Stable Cascade models being used (including @nimble mason ´s stage B finetune):

and usually not sharing prompts like that 😉

[steep sleet]

"Trans-y!" 🙂

[civic pasture]

i mnade this!

[steep sleet]

Florence2/Flux + LoRAs

[crisp stream]

oh and it´s img2img

[steep sleet]

Cannot find wf in PNG?!

[crisp stream]

@civic pasture

no prompt, no input image

[steep sleet]

Agree, but someone mentioned a w/f inside the PNG?

[wispy nest]

I've been having trouble getting comfy workflows out of images too yeah

that was the best part of comfy cos people very rarely share their workflows, but you could take them from the images

[crisp stream]

@steep sleet @civic pasture it's using this Invictus Redmond v1.1 stage C fine-tune:

https://civitai.com/models/316681/invictusredmond-stable-cascade-stage-c-finetune-generalist-model

ClownsharkBatwing fine-tune stage B:

https://huggingface.co/ClownsharkBatwing/CSBW_Style/blob/main/cascade_B-lite_refined_CSBW_v1.1.safetensors

Currentlly cannot find teh link for the stage A fine-tune, you can use the standard one instead as well though.

And yes @steep sleet the WF was missing from the image because I clipboard-pasted it instead of the original PNG. Thee image I posted then contains the workflow, just like this one 🙂 :

[crisp stream]

@steep sleet could/did you get it to work?

[steep sleet]

Afk

[steep sleet]

Just trying it now ...

???

[wispy nest]

he is from the future

where they have 64 channel VAE models

[crisp stream]

guess you haven´t got the proper models yet?

[steep sleet]

I have so many - bf16, lite etc - none of them work

[crisp stream]

you would have to use the Stable Cascade base models or the ones I linked (there are some others around that work as well, though they have to be Stable Cascade based)

[civic pasture]

first thing ive made with ai, Dream ai for animation generated origonal image with stable

[haughty roost]

First time trying video ai. This was made through cogvideox-5b in the top huggingface space. Didnt turn out too good but still pretty realistic looking if you ignore the people

[unique condor]

I got cogvideo 5b running on an 8gb card earlier. Slow as balls using CPU sequential offloading, but it works. Just have to make sure to use tiles vae and vae slicing or else you'll oom

[haughty roost]

Well for some reason I can never get ai stuff whether it be LLM's or image generation models via comfyui to go quickly on my rtx 3060, even though I've seen people with similar tech get super quick results whenever they generate stuff

[unique condor]

Want to say it was around 16sec/it @24 frames and 25 steps per frame. Something like 50 sec/it with 50 steps and 49 frames

I'm using a 2080 FE

But I just wanted to see if it could work at all. The cogvideo 5b is kind of a prototype and doesn't have frame interpolation yet, so it's really jerky without some second pass for interp

[river sigil]

Create me an image with nude or pastel colors, that says "La Real Gracia" and is simple to use as a profile.

[haughty roost]

@river sigil

[still rose]

U think for photorealism SDXL models or 1.5 is better?

[crisp stream]

Cascade 😀

[still rose]

For realism can I run Cascade on a 8gb vram AMD GPU?

[crisp stream]

good question, not sure

[rugged zodiac]

#picture make a picture of a cat

[royal monolith]

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

[rugged zodiac]

ok，thank you

[steep sleet]

Florence2/Flux

[steep sleet]

Diversity!!! ⭐ ⭐ ⭐

Florence2/Flux_GGUF

[steep sleet]

Senorita

[opaque rapids]

特色

[rain ocean]

the third whatsapp reich

[steep sleet]

C.Guile 😦

[crisp stream]

Btw I don´t know how it is practically like there a very specialised fine tunes for 1.5/SDXL, so while Cascade is potentially of better quality I don´t know what the availability of specialised finetunes/loRAs compared to 1.5/SDXL looks like.

[still rose]

Yh fair rn running Flux

[still rose]

Flux is pretty fun

[sterile kiln]

It makes you want to see... the discs under the logo 🙂

[still rose]

I see some unfortunate issues with the gen I think it can be fixed easily

But other than that it's pretty good imo

Especially the fact i can run it on my macbook and it take 2 to 4 min per gen on a m2 pro 16gb ram model

Which is the Dev 8 bit model

[sterile kiln]

Are you using LoRAs? What model & text encoder are you using?

[still rose]

So I was using real lora V1 I forgot who it's by I need to check again, T5xxl_FP16 and rn I'm alternating between the DEV fp16 and fp8

I prefer ofc the fp8

[sterile kiln]

Comfy or Forge?

[still rose]

Forge sometimes and other a application quite popular on IOS for Stable Diffusion "Draw Things" pretty much IOS auto1111

If u don't need all the Forge features then I recommend Draw Things for iphone, Mac, iPad most optimised

[sterile kiln]

Forge is faster.
Try this models :
Flux gguf Q8 : https://huggingface.co/city96/FLUX.1-dev-gguf/resolve/main/flux1-dev-Q8_0.gguf?download=true

Et le text encoder :
https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/resolve/main/t5-v1_1-xxl-encoder-Q8_0.gguf?download=true

Because the basic models are far too heavy.

And with Forge, you have to have theses settings for optimal perfs :

[still rose]

Ah I didn't know that model of Flux exists does it lack in quality or text rendering or is it relatively the same consistency?

[sterile kiln]

GGUF are closer than dev1 + FP16, but much lighter, you save a lot of time loading, then it/s

[still rose]

Ty I was possibly considering trying to get Flux on Zluda with my 6600 xt 8gb but I know it will prolly crash so maybe I'll try this with other optimisations 👍 also try it on my mac

I'll let u know the speed

On forge and if Draw things allows me to download custom models of Flux like that I'll try it on that but first forge

[sterile kiln]

6600 xt nope, maybe with Schnell version, but very slowly if it works, because of Vram, and RocM

[still rose]

Yh fair zluda is slower than ROCm on Linux also yh 8gb of vram is iffy

I'll get a 3090 used soon tho

[sterile kiln]

i just bought used 3080ti 🙂

450€

[still rose]

To replace my 6600 xt does well for 1080p gaming but now I do a lot more AI 6600 xt isn't the best

[still rose]

Nice I can get 3090 with 2 year warranty refurb for £580 GBP

[sterile kiln]

AMD/Intel GPUs, just forget AI

[still rose]

Idek how Intel GPUs even perform on AI 😭

[sterile kiln]

Cuda, is a must have

[still rose]

My fren got a 4070 super and he's enjoying flux on comfy

So I'm there with my 6600 xt like 😕

[sterile kiln]

Tell him about GGUF models, he will enjoy very faster

[still rose]

Will do ty I'll send him the links he will enjoy it

3090 ti used is like £600 to £800 I don't think that's worth right?

3090 seems to be the best price to performance £550 to £630

[sterile kiln]

It is, 24 GB Vram, its this or... sell your body to buy a 4090

[still rose]

True but I mean in compared to normal 3090 price wise

Ik 3090 ti is faster but is it worth the extra £150 ish to £220 faster I do not know

[sterile kiln]

they are only used today, so, It's the law of the market

[still rose]

Yep the prices for me used between the both is odd

I think I'll be fine with 3090

Regardless I get a performance boost in gaming and AI 😁

[sterile kiln]

in France, on https://www.leboncoin.fr/recherche?category=15&text=3090&kst=p&pi=6c28f78d-892a-4551-973f-e3629568d0b9

[still rose]

In UK 🥲

[sterile kiln]

I was on a 2070S before the 3080ti, and the AIs are a delight, but even more so in gaming... BF2042, Baldurs Gate 3, and lots of other games: Full ultra, 55~60°C...
Nothing to do with the 2070S

[still rose]

Although Ebay seems to be doing a bit better

[sterile kiln]

then you need a strong power supply

[still rose]

Yep I wonder how insanely better 3090 will do vs my 6600 xt as the 6600 xt raster performs like a 2070 super

[sterile kiln]

the 3090 can go up to 600 Watts

[still rose]

Yep ik I'll be getting a 750 to 850w rn I'm running a 5600 as a cpu quite low TDP CPU

I'll get 64gb of 3200mhz ddr4 ram for £100 since good kits for DDR4 is quite cheap

[sterile kiln]

It's very short, be careful, it's borderline. Don't be fooled by the TDP, the TDP is not the maximum consumption.

[still rose]

Ik but yk it's a 5600 as my cpu it naturally does take much power I'm not running PBO on either or OC so CPU power I shouldn't worry it's the GPU I should be careful on

Or do I save money and get 3080 ti? @sterile kiln like u did

Alr nvm it's sim price to a 3090 I'll just get 3090 UK used market is dodgy 😭

[autumn turret]

do consider 850, I was about to get a 750 then nosed at the 850 only to find it was cheaper lmao (corsair)

[still rose]

Yh 850w are like cheaper or same price in the UK also

On pc part picker it shows that anyways

[autumn turret]

yeah I'm UK too

[still rose]

W

[sterile kiln]

No, for all these AIs you need much more than the 3090. That's twice as much Vram.
Personally, I had to fall back on the 3080ti because I couldn't really afford it.

[autumn turret]

https://tenor.com/view/sips-tea-sip-gif-25933597

[still rose]

That's fair

Iirc raster for gaming 3080 ti is 3% slower than 3090

So isn't that big of sacrifice on gaming vs 3090 unless u play with RT then diff story

[sterile kiln]

However, beware of the founder edition, which should be avoided like the plague, as they are noisy and can be damaged by their faulty thermal pads.

[still rose]

Ah yes I've heard about FE cards for a while I always go for 3rd party

[still rose]

@still rose discord saving it as I'm on phone

[haughty roost]

God I love imagen3, wish there was a local version though

[still rose]

@sterile kiln

[sterile kiln]

GGUF checkpoint and GGUF t5?

[still rose]

Unf I couldn't get it run both gguf and NF4 idk if it's a ez fix or not I don't really know how to fix SD UIs on Mac

[sterile kiln]

As I was saying, you need the GGUF checkpoint, and the GGUF text encoder too. nf4 is a single quantized file with: ae.sft, flux1d and t5fp8.
nf4 is not as good as GGUF Q8.
And of course it's normal that it doesn't work by mixing the 2.
I've put a screen above for the configuration you'll need.

[still rose]

I've done those but it gave me
"TypeError: Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead.
Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead."

[sterile kiln]

on Forge?

[still rose]

Yep

[sterile kiln]

You probably not respected theses settings :

And obviously, an fresh updated version of Forge is required.

[still rose]

When I went to the tutorial for forge on mac from CS1o I pretty much copied those settings ill assume the auto installer didn't download latest version?

I'll try doing a Git pull command

And report back

[sterile kiln]

if you've just installed it yes, if you have an old installation...

[still rose]

That makes sense ty for telling me

[still rose]

One more question how do I git pull update on mac

[sterile kiln]

Maybe :
cd path/to/directory
git pull

in a regular terminal

[still rose]

It says already up to date. Interesting

Let me try the settings again

Maybe I missed something

[sterile kiln]

I don't know how it works on MacOS, if you've just installed Python (and if you have the right version, by the way?), nor do I know if your branch is the same as the basic one on Windows... Etc, lots of potential problems

https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/1103

[still rose]

MPS issue seems to be a unf problem rip

[sterile kiln]

MacOS 🤷

[still rose]

I got it bypass with a backend fix now the problem is I'm out of memory 💀

[sterile kiln]

unified memory is a very good thing, unless you're short of it.

🤷

[still rose]

16gb of it on my macbook

[pale steeple]

i am experimenting trying to make my own workflow and learn comfyui.
why is this not working?

[tall sedge]

i installed comfiui but why it not change?

[tall sedge]

why i can't see the image i created?

[royal monolith]

drag the save image node out larger

[tall sedge]

i have a bit confused with this
and accidently click on unload model, can you help me?

tried click on update, installed all things by update all/up conffyui,installed misssing

[royal monolith]

that's the comfy manager. the save node is at the very end of your work flow and is labeled Save Image - make it larger so the images you generate will have a place to display. You'll also find them being saved to /comfyUI/output

[tall sedge]

i alr did that sir,

is that ok now?

ty sir

i think it can track img promt like auto1111, hope it right

@wispy nest i mean idk if comfyUI can do like this then sit all setuped on tracked img to text to text label like this

and imo the clip i watching smt liek that, idk if you can comfirm it?

[wispy nest]

oh I see
yes it can do that

it can be a bit tricky to setup but it is possible

[tall sedge]

also how to install "flux" and what does it do to comfyUI sir?

i tried on youtube but it seem the file have been deleted

[tall sedge]

i hope i can ask you how to do it in future, now struggle on confyui tutorial so ==!

also idk how to fix this, is my cecurity too high or?

alr find it but what level should i put in to
high, low, weak, middle, normal?

[wispy nest]

go to config.ini and set security_level = middle or low

sorry I mean normal, normal- or weak

information is here: https://github.com/ltdrdata/ComfyUI-Manager?tab=readme-ov-file#security-policy

[tall sedge]

alr done sir, ty forr help

[tall sedge]

urgh, can you fix this sir

lmao smh it ok now ==

[tall sedge]

also do i need install all of them, it from custome node manager

[wispy nest]

no you don't need them

[tall sedge]

ty sir

[finite cedar]

I cant click this dropdown

When i click it it just makes me start typinh

[clever oar]

try change browser

apdate ui

[tall sedge]

good think i now can send img prompt by using load img, but smh it not what i want, about the color, what i still missing and how to get it sir

original img here

[crisp stream]

The issue I have with Flux (just like with SD3) is it not really being able to correctly display art styles. Everything tends to look either realistic or plastic 3D like. Rough flat painted appearances are hardly possible in that context.

Reason I am generating with Cascade again.

In Flux as well (until I get bored of its "plastic coating")

[still rose]

for me its just 3d plastic for realism

😭

[tall sedge]

is there;s anything i need to take off, install more?

trying to make anime stuffs so

[shut sinew]

You do not install swarm ui as a custom node afaik, it's a separate launcher that uses ComfyUI as a base and gives it a different UI

Also please use #🤝｜tech-support for tech questions

[tall sedge]

yeah i untick and uninstalled it

[tall sedge]

ty sir

[steep sleet]

Yes, it is difficult to escape its "Fluxiness" 😄

Yet excellent when it works well

[languid pebble]

Long neck with Flux ^^

[tranquil mural]

как тут рисовать?

[steep sleet]

HeT! Artisan

https://discord.com/channels/1002292111942635562/1237460438229450772

[shut sinew]

I have a similar pic

[languid pebble]

Cool ... it's been a prompt censored by another plattform for any reason.

[thorny ingot]

“A beautiful front yard of a grand house, filled with lush green grass and a variety of colorful flowers carefully arranged in vibrant patterns. The scene features dense trees trimmed into perfectly round shapes, adding elegance to the garden. The sunlight casts soft shadows, enhancing the vivid colors of the flowers and the greenery.”

[steep sleet]

Your image as requested

[steep sleet]

Florence2/Flux img2img

[rugged stone]

Yesterday was the first time when I generated an ai image

[crisp stream]

I see, well 🙂

[crisp stream]

have you got an example image?

[languid pebble]

Reminds me of Uma Thurman in Kill Bill ...

[steep sleet]

#🏞｜general-with-images message

[crisp stream]

they look good, yet havent got any flat painted appearance with visible brushstrokes

[crisp stream]

for more natural realistic images I can recommend working with according input images:

Input image (generated in Stable Cascade)

[still rose]

cascade or flux?

[crisp stream]

Flux with the input image below from Cascade

[crisp stream]

well not this one but the one below the first image

[wispy nest]

do you mean like IP adapter?

what is input image?

[crisp stream]

image to image

[still rose]

i did find a lora that fixed most if not all thats dodgy plastic skin

[nimble mason]

Here is the image you requested.

[rain river]

Can anyone generate for me an anime version of Norm from The New Norm? It is extremely important.

[nimble mason]

Here is the image you requested.

[rain river]

Wow, they're absolutely nothing like what I wanted! I don't think the AI understands Norm. Thanks anyway.

[steep sleet]

Anyone have a random prompt generator w/f for ComfyUI at all? 😄

[steep sleet]

Florence2/Flux.Dev.SingleFile img2img

[visual hemlock]

Old GPU out 😄

NEW gpu added 🙂

To my secondary PC 🙂

Currently getting AI setup and hoping to start testing 😄

[steep sleet]

Dynamic Prompts in ComfyUI (Mobius checkpoint)

Rolls Royce farm vehicle 😄

Sneau and ice

[civic pasture]

4070?

[visual hemlock]

Sadly no 😛 An 3060 😄 12 GB

[steep sleet]

Moebius checkpoint using Searge SDXL nodes

[wispy nest]

I looked up Searge nodes but it said this Detailed Documentation (TBD)

lol

I almost never find node packs with docs

[nimble mason]

i wonder if that's using the correctl clip L

[steep sleet]

It seems to be marginally better/slightly more detail than Flux1.Dev (or am I just imagining things?) 😄

[still rose]

What's the new gpu? just read above my bad

I would get a 3060 12gb but for gaming it would be a side grade to my 6600 xt

[nimble mason]

no idea, we're discussing it over in #sd3 though

[visual hemlock]

I am running an 2080ti in my gaming PC and the 3060 now in my secondary PC

[still rose]

I'm still considering 3090 used

I don't want to build a second

I wanna just upgrade I don't have a use for 2nd system

[visual hemlock]

That is okay too 🙂

I am using ai pictures etc and selling so this was to get it away from my gaming PC

and handle all stuff on secondary pc. I host local GPT now there and running some home servers so 🙂

I have the use for it.. THe GPU only for Stable Diffusion and GPT ofc xD

[analog panther]

prompt: A warm, golden hour scene, reminiscent of Wes Anderson's nostalgic style, set in a cozy, pastel-hued cafe. A young woman of East Asian descent with porcelain skin, striking features, and an alluring gaze, sits at a small, round table, taking a selfie. Soft, diffused afternoon sunlight pours in through a nearby window, casting a gentle glow on her face, accentuated by a subtle fill light with a warm, yellow tone, positioned at a 45-degree angle to her left. Her eyes, an arresting shade of brown, sparkle with a hint of mischief, while her lips, painted a deep coral, curve into a captivating smile. A shallow depth of field blurs the surrounding environment, drawing focus to her face, with the cafe's pastel colors subtly visible in the background, incorporating a triadic color palette of soft pink, baby blue, and pale yellow. A slight film grain and warm color grading evoke a sense of intimacy and nostalgia, while the overall atmosphere remains light and playful.

Steps: 50, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 7, Seed: 2772991253, Size: 1024x1024, Model hash: d91d35736d, Model: juggernautXL_juggernautX, Version: v1.10.1-amd-5-gd8b7380b

result ^

why does it not output an image with "taking a selfie"?

[analog panther]

@royal monolith 150 steps

30 steps

Steps: 150, Sampler: DPM++ 2M SDE, Schedule type: Karras, CFG scale: 7, Seed: 1738328287, Size: 1024x1024, Model hash: 8421598e93, Model: AnythingXL_xl, Version: v1.10.1-amd-5-gd8b7380b

i see , what does the sampler do?

btw my display freezes at 100% completion , for a few seconds

any idea?

none

[wispy nest]

probably need tiled VAE decoder

[analog panther]

what does that mean?

intel 15 13500
32GB ddr4 ram @ 3200MT
amd 7800xt gpu 16GB vram

[wispy nest]

its hard to find ways to explain in an easy way
when you do text to image, you are actually running two models
the first model is a U-net and it outputs a latent, which is like a very compressed image
the second model, the VAE, turns the latent into an image

its possible that running the VAE is filling the memory on your graphics card up

using a tiled VAE helps with this

you have 16GB vram though so this seems unlikely now

[analog panther]

can i run 12GB fp32 models on my gpu?

[wispy nest]

just try it and see

[analog panther]

good idea

thats how i installed it

yeah

[wispy nest]

you might have to ask on AMD discord, or ROCM one if there is one

not sure about AMD

[dry crow]

VAE process. Can be fixed with the tiled VAE extension

[analog panther]

thank you

[dry crow]

You shouldn't xD

[analog panther]

this?

[wispy nest]

the other day I made an image so big it needed tiled VAE on a H100 😿

[analog panther]

💀 wtf

u have h100?

[wispy nest]

haha no no

genuinely my main interest in these models
is breaking them in funny ways or pushing them to their various limits
I don't particularly have a need for images

oh I agree that's partly why I use cloud

I make the workflows locally and test a baby version with easier settings
then open cloud server for a number of minutes to run it there in full

[dry crow]

Yes

[languid pebble]

Have you tried with cascade ultrapixel and how big?

[wispy nest]

not yet I keep meaning to, since I talk to Clownshark a lot about his project

I have actually been getting 5k images out of SDXL with just 2 passes

using the combo of FreeU, SAG, PAG, SEG, a block-by-block and scheduled version of Koyha Deep shrink, and this thing called Res-adapter

oh also the upscaler in between the two passes is transformer rather than ersgan

haha yeah

I'm kinda confused that this workflow works
cos on reddit and youtube I never saw people saying SDXL can do this

I guess the most popular thing is to use tiled upscale or external sites like magnific

[steep sleet]

Flux.Dev

[wispy nest]

there is a lot of bandwagon jumping yeah

a funny example is on Civit you see Pony tags in non-pony models

people are prompting Flux with pony tags even

[steep sleet]

"I am a prompt wrangler! Any prompt that gets within the range of my lasso I wrestle it to the ground. And then publish!!!" 😄

[wispy nest]

ah yeah I do see greg rutkowski a lot in civit prompts

[steep sleet]

Beksinski, Giger

[wispy nest]

the worst thing in civit prompts is the massive negatives

sometimes 150 token negatives

[steep sleet]

mat3o says the best negative prompt is aaaaaaaaaaaaaa aaaa aa aaaaa

[wispy nest]

there's a set of nodes that lets you explore nearby tokens that is cool
https://github.com/Extraltodeus/Conditioning-token-experiments-for-ComfyUI

[analog panther]

do realistic one

non animated

[languid pebble]

😄

Prompt: 3072

[analog panther]

how does it work

why doesnt it work for me

what prompt u gave?

[steep sleet]

It works well - good prompt coherence as she is seen taking a selfie

[steep sleet]

A warm, golden hour scene, reminiscent of Wes Anderson's nostalgic style, set in a cozy, pastel-hued cafe. A young woman of East Asian descent with porcelain skin, striking features, and an alluring gaze, sits at a small, round table, taking a selfie. Soft, diffused afternoon sunlight pours in through a nearby window, casting a gentle glow on her face, accentuated by a subtle fill light with a warm, yellow tone, positioned at a 45-degree angle to her left. Her eyes, an arresting shade of brown, sparkle with a hint of mischief, while her lips, painted a deep coral, curve into a captivating smile. A shallow depth of field blurs the surrounding environment, drawing focus to her face, with the cafe's pastel colors subtly visible in the background, incorporating a triadic color palette of soft pink, baby blue, and pale yellow. A slight film grain and warm color grading evoke a sense of intimacy and nostalgia, while the overall atmosphere remains light and playful.

[analog panther]

and negative?

[steep sleet]

Nothing

[analog panther]

how many steps

💀 it will just NOT give me a selfie pose

[steep sleet]

20

[analog panther]

yeah i get this pose

[analog panther]

60 steps

nah

[steep sleet]

She is very clearly taking a selfie here!

[analog panther]

nah

[analog panther]

this is better

[analog panther]

trying

but this is not FPV

how do u do FPV

not fpv

[analog panther]

but like this

first person view , but not that , my bad

[steep sleet]

FPV = full person view?

[analog panther]

nah nah

[steep sleet]

Closeup on face and chest

[analog panther]

i mean like this

how do u get that

also , how do u use a Lora in webui?

u havent seen tikok

[steep sleet]

Flux.Dev+LoRAs (rbcharc0al, Blue Future, BW Wraith)

[analog panther]

💀

[steep sleet]

D/load this png - it contains the workflow #🏞｜general-with-images message

[analog panther]

using webui

auto1111

[analog panther]

flux?

trying fp32 model

time to cook ||the gpu||

[steep sleet]

Select LoRAs in A1111 - they will automatically be inserted into the prompt

[analog panther]

how do i select?

[steep sleet]

Use the LoRA tab in A1111

[analog panther]

this? at bottom?

[analog panther]

oh i see

[analog panther]

fp16

💀 wait

they are same?

[dry crow]

yes

[analog panther]

but? why

[dry crow]

because just the precision is changed

and from 32 to 16 its not that noticeable

[analog panther]

oh??

[dry crow]

thats why most models are fp16 nowdays

[analog panther]

[dry crow]

because they are smaller and faster

[analog panther]

thats cool

[analog panther]

hand weird

[still rose]

Flux with Loras are cool

[south temple]

awefulyl quiet in here

[wispy nest]

Hello

[languid pebble]

Good morning coffee!

[steep sleet]

Florence2/Flux img2img with SD3 input images (SD3 first, then Flux)

Florence2/Flux img2img with SD3 input images (SD3 first, then Flux)

Flux forgot to add a driver!!! 😄

[fossil mason]

Sorry, something went wrong while trying to process genre. Please try again later.
Reference id: 4R8LBSRC

Sorry, something went wrong while trying to process genre. Please try again later.
Reference id: JPLFT9S6

[steep sleet]

Florence2/Flux img2img

[viral frost]

"Hello, friend. Would you care to try my delicious beef?"

bizarre unbelievable action photograph of an ((yak:1.1) drinking bloodventure (Compton:1.1):1.1) holding a bizarre deep dish meat surrounded in glowing whitehandler girl sections, intricate, (lethramphilly:1.1) viewing The Impossible (\u2013androids:1.1) video game offryly Lip a fiend forlorn mummy Tozzbot spelunker

bizarre unbelievable action photograph of the potted meat man during a glasgow astral gogurt carnage restaurant interior) photograph of the sloppy calamari kamaitachi on an plates countertop at work wearing a white and red as far as humanly bornasketball allowed malefin ceivable dwarfrikes milks a beandogpless fruit loaf per giomoghia noir

bizarre unbelievable action photograph of this, illustration, greebling: a mecha robot woman holding a ((huge:1.1) (monstroustrakboni:1.1) thingaped cartographer huge imagined cgi renderedanced fruitfly:1.1), chromes detailed hartebeest face photography detailedorgebydog, blurry focus marinara coveredin, grinning

bizarre unbelievable action photograph of a magical gogurt-covered central sky portrait magicalwartsfilm manwall apegrobb, ancient chud, greezy hihoh!endorust god rays, neon colored paperondon cart in background, sloppy ragpane, technicalks, jury render...showcase cled orbs andubabling amazeballs alien doggled our simulaped friends torturous lit gumdrop (Azathoth:1.1)

[steep sleet]

Florence2/Flux img2img

[subtle meadow]

Testing Flux in very different themes. Not always better than SD/SDXL. But one thing much more better in the Flux: the glassy fruits as ornaments:

[steep sleet]

Florence2/Flux img2img

[remote scaffold]

A futuristic cityscape with flying cars and neon lights, at sunset

image generated 2 images with model Ultra, using prompt Create an image that shows a step by step on how to draw a reading guinea pig with glasses and a book in a cartoon style, but accessible to someone who is just starting to draw., seed 1807284537, aspect 1:1!

/generate prompt: A futuristic cityscape with flying cars and neon lights, at sunset

!generate A futuristic cityscape with flying cars and neon lights, at sunset

[languid pebble]

Have a look here: https://discord.com/channels/1002292111942635562/1237461679286128730

[rain gazelle]

who remembers Titanic? 😛

[languid pebble]

He has found a bug ... ^^

[steep sleet]

A majestic holographic Ship of Theseus sailing through the cosmos, glowing in iridescent hues. The ship is illuminated by ethereal light, with intricate, transparent sails that reflect the stars. Surround it with shimmering galaxies and celestial patterns, merging with the ship as if part of its structure. As the ship glides through space, fragments of its body transform into stardust, symbolizing the paradox of its continuous renewal. Cinematic lighting, ultra-detailed textures, and a dreamlike atmosphere.

[crimson dawn]

and then flip out as if the model is broken and it's the entire world's fault and not theirs

[white pivot]

My new workflow for Hyper Flux is out:

https://openart.ai/workflows/elephant_present_36/hyperflux_llm_inpainting_upscaling_lowvram_v2/rzT092UZHZqdoVhwPhP6

[silent swan]

imagen de tienda exitosa en un barrio en bogota con luz cálida en un día soleado con clientes felices con abarrotes y alimentos bien organizados y una mujer joven sonriente amable con traje típico bogotano vendiendo hielo en cubos al lado de una maquina de hielo

[jade python]

Simple line drawing of a rabbit

[subtle meadow]

Hi, I have lot problem of Flux prompting. These days I re-create my older prompts by Flux, what was good in SD/SDXL. Some prompts better in Flux, primary realistic human portraits or simple human photography in several situations. But if prompt contains several styles or moods, cyberpunk, steampunk, futuristic, retro, rocker, mechanical etc.... and have to mix these styles with some person (or animal), the result is really bad compared to SDXL. Have I modify my old prompts for Flux or this type of model (the official dev and GGUF) cannot generate stylized images? Here are examples, "Woman from the end" (woman from the apocalyptic future world) so I think easy to find which 2 made by Flux:

[vagrant dust]

Well Flux is great for SFW Content the nsfw Part is kind of not trained. So horror, gore, blood, violence is not as present as it was in the SD 1.5 / SDXL world. So you would be more specific with the prompt and describe for example the skull etc. to achive these images with flux.

[vagrant dust]

Would be a flux approach to the image top left

"A cyberpunk woman with vibrant blue, pink, and yellow hair stands in an intense, post-apocalyptic environment. The character has a skull face makeup with, surrounded by mechanical enhancements in her body, Their expression is fierce and ominous, exuding danger making them look both futuristic and undead."

[subtle meadow]

I created half-nsfw and horror themes with flux (see 1-3 examples attached) but the images you created I dont really like. Her head is relative good, but under the stylized head her body is realistic and "cute" what mean not too consistent for the whole image (3 examples how I mean).

[vagrant dust]

Yeah as mentioned the usual selection of images seems to be based on some aesthetical design where the body of a woman is defined by chest and stomache size.... sadly if you want a more realistic over all image you would need to add specific details like stem, chubby, trained, battleworn, robotic parts,.... As by random it seems that there are 1000 Images of Models to 1 Images of your neighbour 🙂

[subtle meadow]

Yeah that interesting. I made Flux comparison test 2nd day, and the results sometime very shocking. For example, this car is named Trabant, really dont understand who trained Flux model by the photo of this old shit plastic car 🙂

[vagrant dust]

Well maybe the BlackForest guys went to see the spreewald and passed some Trabby 🙂

[subtle meadow]

Maybe they are taken photos on oldtimer show 🙂

[vagrant dust]

Pixart does not know the trabant, playground seems at least to know it is a car 🙂

[subtle meadow]

I used Flux GGUF for the pink Trabant 🙂

[steep sleet]

DallE Theme of the Day 😉

[steep sleet]

Trabants were made of wood 😄

[languid pebble]

Good morning cappucino

[steep sleet]

Searge SDXL

[graceful lynx]

Yeah that interesting. I made Flux comparison test 2nd day, and the results sometime very shocking. For example, this car is named Trabant, really dont understand who trained Flux model by the photo of this old shit plastic car

[white pivot]

I improved my upscaling group in my workflow

https://openart.ai/workflows/elephant_present_36/hyper-flux-llm-inpainting-upscaling-low-vram-v3/rzT092UZHZqdoVhwPhP6

[hushed oriole]

[wispy nest]

dall e3 is funny. I didn't describe the clothes of the kid, but I said "expensive and rich" for the adult
https://th.bing.com/th/id/OIG1.TMReHUuLUBrtjOs7_nUR?pid=ImgGn
What was the AI thinking to generate this?

[lavish garden]

Prompt: https://civitai.com/images/27925575

"Blood Moon" - Hellboy

[wispy nest]

am i the only one seeing this

[paper heart]

Is this dev fp16? Thanks

[steep sleet]

Ollama+Mobius

[sterile kiln]

10 buzz il you recognize him 🙂

[steep sleet]

Ollama+Mobius

[steep sleet]

Jacques Delors!

[sterile kiln]

Hahaha, close

you have 50%

[steep sleet]

Jacques Chirac

[sterile kiln]

👍

(and this is a real fact on the sign)

[steep sleet]

Ollama+Mobius