🏞｜general-with-images

[copper matrix]

their V1 release was well a small mess

[wispy nest]

yes it comes down to preference, but it'd be appropriate for me to say they both have their unique traits and styles, i guess i was leaning on those to like pony better, but illustrious model can be amazing for what it does

[copper matrix]

true that

[wispy nest]

the lines are vividly distinct with illustrious and very artistic, but lacks in background detail fill, pony fills in that area

[copper matrix]

i mean but you mostly generate with a flat color background (or atleast that we know of)

[wispy nest]

yes but sometimes i do some colorful background

a mix of complex pattern

[undone dock]

Eggshell bagel =/

[wispy nest]

how's this for sampler choice?

result

[copper matrix]

can you not scare me like that. i thought someone else posted lmao

looks great

[wispy nest]

scare you how

[copper matrix]

posting non women

thoughtyour account got comprimised

[wispy nest]

oh

i was trying out some food images

only sdxl model i use just for object related pics

cause pony can't make good objects

but then pony is great for subjects far better than sdxl

[undone dock]

If you're messing around with a 5.0 CFG, give dpmpp_2m_sde & exponential a try.

[wispy nest]

yep, was doing that too. nice results

actually chatgpt suggested me the same thing

lol

except it said *sde

[undone dock]

Combined they make amazing images. If you're also using Hypertile, I recommend modifying it to 1280/8/0/no

[wispy nest]

how many steps are you going for with dpmpp 2m sde / expo

[undone dock]

I use it with SD 1.5 models for SDXL quality 1080p generation. 16 steps

[wispy nest]

using this

this is the result

[undone dock]

Looks natural compared to the image above.

[wispy nest]

eyes look better

same setup, another example

too much cleavage maybe .. lemmer replace that pic lest fruit gets mad at me

much better

but honestly i posted the previous pic because of her sweet face and texture

it'd be very very good if society didnt body shame women and their beauty

[still rose]

i do like flux a lot

[celest sigil]

Hunyuan Video T2V

[true frost]

@wispy nest hey, unfortunately it doesn't work well, I don't know what the problem is, but I don't think this model is suitable for interior design. Can you tell me what I'm doing wrong?

[copper matrix]

You could change the model. Theres a few that try to work with interior design

If you think thats the problem

[copper matrix]

Though is that the only image you made or did you try it a few times?

[true frost]

@copper matrix I've tried many times but everything looks like a fairy tale... 🙂

[undone dock]

I'd recommend trying a different inpainting model than Deliberate. Looks like it was trained on a lot of fantasy elements. https://huggingface.co/XpucT/Deliberate

[true frost]

@undone dock Can you give me another good model that is suitable for interior architecture?

Does this only apply to Deliberate?

[copper matrix]

i recommend checking on CivitAI for any realism inpaint models 🔥

[undone dock]

Best one I can see is https://civitai.com/models/112690/alkaid-realistic-fill
But https://civitai.com/models/622597/interiordesignluluv10 might also work better
Just check it out, make sure to filter by the Model architecture you're using.
https://civitai.com/models?tag=buildings

[wispy spindle]

hello to you as well

[wispy nest]

does it have to be SD/SDXL? Flux Fill is so effective that even closed source struggle to compete with it
out of SD/SDXL stuff, Powerpaint V2 is very good

[undone dock]

I'm pretty sure I've seen a real t-shirt with this on it before.

[spring mist]

its not AI, shh

[wispy nest]

you should be able to find what your task requires with a bit of google search, there are tons of models out there to choose from, here is one link for architecture related work https://civitai.com/models/533040/realarchvis good luck

[wispy nest]

this is a pretty amazing AI generated video https://x.com/PJaccetturo/status/1892624681611755897

PJ Ace (@PJaccetturo) on X

This is one of the most realistic AI AI-films I've ever seen🤯

Nicéphore worked on this for weeks for the Paris AI Summit.

It's a story about the birth of cinema 130 years ago. What do you think?

X · Feb 20, 2025, 5:17 PM

[undone dock]

I don't know if this Discord server has a url filter, but here goes.
For future reference, anytime you link to X, you have to change the URL like so:
https://fixupx.com/PJaccetturo/status/1892624681611755897

[wispy nest]

i always remove embedding off the link 🙂

[undone dock]

Your link won't even show a video file, just an image. Mine will let people in the channel watch the video.

[wispy nest]

the link doesn't open a new tab to x page?

[undone dock]

You're misunderstanding. Why visit X when you can just watch it in the channel?

[wispy nest]

ahh i see what you mean

well i guess you are asking for ease of use, but what i intended was to share the link

[wispy nest]

why is she about to cry?

[true frost]

As usual, he's probably tired of playing the guitar and the constant pain in his fingers. 😄

she*

[wispy nest]

poor girl

[true frost]

Even AI knows that people have to struggle and get tired when they practice playing instruments, which is visible in their facial expressions.

@thorn forge nice light in your works

[thorn forge]

tanks

[true frost]

I work on interior design and combining my photos with generated backgrounds

[wispy nest]

cheering this up a bit

[pale steeple]

trying to install skyreels? how do i install these nodes? pls help thanks

[undone dock]

ComfyManager IIRC, has a section where you can scan a JSON file and it will tell you what Nodes you need and how to install them.

[pale steeple]

where exactly?

[copper matrix]

"install missing custom nodes"

middle row, second from top

[pale steeple]

already did this

[copper matrix]

your sure your comfy is up to date?

[pale steeple]

how to update comfyui?

[copper matrix]

...the "update comfyUI button"

[pale steeple]

these are the problem nodes

still getting these error after update

this is fun

[copper matrix]

a KJnodes

[pale steeple]

sooo?

[copper matrix]

did you try googling for "KJNodes"

[pale steeple]

yeah, but the results are something i can't wrap my brain at the moment, it's only github seems safe

[copper matrix]

i mean are you following a custom guide or the comfyUI one

[pale steeple]

and no download

can you link me a good skyreel workflow? this might a shitty one, please?

[copper matrix]

https://www.reddit.com/r/comfyui/comments/1ist4g7/comfyui_native_support_of_skyreelsv1hunyuani2v/?rdt=41695

the added workflow is a video on the github linked in the top reddit comment

since its natively supported you should be able to use it from the get go

[pale steeple]

i am here

why does everything has to be so cryptic?

where is the workflow file?

i am trying to load the mp4 file already

does not work

[copper matrix]

the added workflow is a video

[pale steeple]

i am loading the video

[pale steeple]

oo so i got to do it myself than?

cool

[copper matrix]

look please don't take this the wrong way but using ComfyUI requires serious troubleshooting and research. since its just been a short while since its out theres probably a lack of documentation or good workflows. its a wildwest

so unless you got patience for someone to make it

you gotta drag them noodles your self

the video basically showed you a working workflow

[undone dock]

It also helps if you have the skills to make things yourself.

[pale steeple]

how do i add height width nodes?

[rotund veldt]

a tiger

[copper matrix]

Heres your AI generated image
🐅

Remix🔃 generate more ▶️ upscale🖼️
Just kidding, check #artisan-faq

[undone dock]

I'd like to generate more.

[copper matrix]

🚫 Error 🚫
you ran out of goodboy tokens to generate more images!
Click here to topUP on tokens ❌

but now fr time for bed. its almost 2am

[undone dock]

https://tenor.com/view/kanna-kamui-chloe-black-card-rich-hotdog-gif-23044743

[pale steeple]

which folder do i put the safetensor files for this node? pls help?

i tired undet

unet

[pale steeple]

where can i find the official skyreels gguf file any help pls?

[wispy nest]

i noticed in your workflow you have ipadapter and controlnet as shown in the video, but given your criteria of work i would also suggest incorporating florence2 for even more precise manipulation of the images you work with, have a look here: https://github.com/kijai/ComfyUI-Florence2 and basically you would incorporate florence2, ipadapter, controlnet with your SDXL workflow

[fleet goblet]

https://huggingface.co/Kijai/SkyReels-V1-Hunyuan_comfy/blob/a0387084af62b9fbcfc56bb9d92db1f35928c136/skyreels-hunyuan-I2V-Q8_0.gguf

[pale steeple]

thanks a lot

[pale steeple]

anyone knows why I am getting this error?

InstructPixToPixConditioning
'VAE' object has no attribute 'vae_dtype'

getting it with multiple workflows for skyreel

i am stuck 😦

[pale steeple]

i am loosing hope friends

this is the result i got with skyreels

this is fun

why do i got this shitty result?

[pale steeple]

can anyone give me the link for a good vae for skyreels please? thanks!

[pale steeple]

where can i find llava_llama_fp8_scaled.safetensors ?

[copper matrix]

Your trying to do something that 95% of people here cant run.

Cant help you there buddy. But i recommend looking on huggingface

[true frost]

@undone dock Hey, I changed the model to the one you recommended, but the effects are still the same... i.e. fairy tale style.

[undone dock]

If you're trying to do what I think you're trying to do, photoshop (or other image editor) might just be easier for you. Just use a global selector to grab the frame (non background), and plunk it into your background image

[sick remnant]

I have never done image editing, is there an easy way someone can give tips on how to change the word to be consortium at the bottom

[true frost]

@undone dock thanks for the information, I will try.

[undone dock]

I'm not sure what image editor you use but on Paint dot net, it's a Magic Wand. You'll just have to play with the percentage to grab only the frame.

[dry crow]

WQHD Pixelart Wallpaper

[wispy nest]

illustrious models can't do backgrounds as good as pony

that's my main reason not to use illustrious but illustrious is good for finer line detail

[woeful wind]

after ur training @dry crow

[tidal anvil]

illustrious models can't do backgrounds as good as pony

@crystal delta illustrious models can't do backgrounds as good as pony

[copper matrix]

As always don't engage with a scammer chat

[undone dock]

Sorry to bother you again @wheat girder but we have another one. Are you the only Mod we have?

[wheat girder]

currently yes

[copper matrix]

Are there any plans to change that in the future? Imo it feels overkill to ping the bosses for a spam/scam post

I think there would be plenty of volunteers considering there's still a solid group of people here

[copper matrix]

@wheat girder hate to ping but another one, sorry

[wispy nest]

Claude's cat prompt

Gemini's cat prompt

O3-mini's cat prompt

just the word "cat" on its own

[wispy nest]

nce

[clever oar]

like real movie

[still rose]

what can sd3.5L be used for? i would like to give it a go but many use flux and i cant find a answer to this

[royal monolith]

Everything.

[still rose]

It seems to struggle with hands for me ☹️

I’ve heard people use 3.5 for pixel art a lot

[royal monolith]

All ai's struggle with hands. Scott has a good tutorial on youtube for a simple comfy workflow to fix hands

[royal monolith]

Ill be happy to work with you and help you get a handle on it if you like

[still rose]

That would be helpful I’m kinda don’t use comfy all to much so idk how to use it properly and I’m also on MacBook so I use other UIs mainly for the initial generation then use comfy for I2I stuff but I don’t mind trying out T2I on comfy

[royal monolith]

Sd 3.5 large. Prompt: closeup of gloved hands holding roses

[royal monolith]

Comfy doesnt like to run on macs

Sd 3.5 large. Prompt: a sweet 16 smiling

[still rose]

Yh I’ve noticed which is mainly why I use it for I2I and not much for T2I due to Mac issues but I don’t necessarily know how to fix hands without comfy it seems and rn the GPU that I had that a fren let my borrow kinda just died so he’s getting it repaired while I search for a new GPU that isn’t to expensive in the UK as 40 series decided to increase in price massively

I originally had AMD GPU because at the time I was only gaming but now I do more AI and AMD isn’t gonna cut it

[royal monolith]

Sd 3.5 large. Prompt: a 3d cartoon chef rooster cooking dinner

[royal monolith]

You need nvidia gpu, and intel cpu

[still rose]

Specifically Intel CPU? how come?

I was thinking X3D CPUs as I still do more gaming than AI but I would get a ram and Nvidia GPU

Since the games I play don’t really require any high end hardware to run fully so they mainly love AMD 3D VCache

But also heard the new Intel core ultra CPUs are quite mid which is why I didn’t consider it

[royal monolith]

Because amd has a tendency to want to take over and do the gpus job, and you dont want to fight with it.

[still rose]

Oh I didn’t know that so would I need 14th gen intel? Or do I go with newer core ultra

I would assume I need minimum of 16gb vram on the GPU

[undone dock]

That sounds like a motherboard issue to be honest. On most motherboards you can choose integrated or discrete graphics. If you leave it on "Auto" I can see it being an issue.

[royal monolith]

Personally, id get one of the new 50 series machines from nvidia

[royal monolith]

It is. Easier to just get intel

[still rose]

For the GPU i shall try getting 50 series but i will have to do a whole platform swap on my pc for intel CPU and idk much about what the “best” bang for buck CPU

[undone dock]

8 cores / 16 threads is pretty much recommended

[royal monolith]

Just get the nvidia 50 series laptop

[undone dock]

a laptop.... right...

[still rose]

I used to use Intel a long time ago and used its IGPU for recording it was pretty cool that I figured out I can do that

[royal monolith]

They are designed to do what we are doing - generating with AI

[still rose]

I uhhhh would not prefer that due to having to keep it plugged in and would be way more expensive than upgrading my pc build now

[royal monolith]

Yup. Theyre designed for this work, unlike gaming laptops which are not

[royal monolith]

Ok. Well poke around on the nvidia website

[undone dock]

I'm laughing at the entire suggestion of getting a Laptop for any sort of AI work.

[royal monolith]

Keep laughing, but at the same time, go read the specs

[undone dock]

You want a real GPU, not the mobile version of the chipset.

[still rose]

When I did consider Intel I was looking at core ultra 7 and at the time 4070 ti super but that’s before 50 series came out then I switched to 9800x3D on the list when that came out but now during this conversation I’ve gotten very much confused on what to do

[royal monolith]

Once you read about their new machines, then we'll talk

[royal monolith]

Research is what you do

[sterile kiln]

I've rarely read such big bullshit.
No, the CPU doesn't matter. The only thing that matters is CUDA, tensor cores, fp16/bf16.

[undone dock]

I've used both Intel CPUs & AMD CPUs for AI generations, both are fine. GPUs on the other hand is definitely nVidia, hands down.

[royal monolith]

You can argue with @undone dock

[undone dock]

Stop trying to drag me into your idiocy crystalwizard

[still rose]

Rn I’m rocking a 5600 and was 6600 xt build i was looking at used 3090 at one point for £620 to £680 but not sure how good of a price that is

[sterile kiln]

Bah listen, invoke him, and ask him precisely for more information about this huge bullshit, because it's powerful anyway.

[undone dock]

crystal is just upset that I'm ignoring his ranting

[still rose]

Ik 5090 would be the ofc ultimate GPU but I don’t have £2k to 2.6k to spare XD

[undone dock]

That sounds like standard prices for a used 3090.

[sterile kiln]

as all of us

[undone dock]

5090s are also melting their power connectors, so I'd hold off on one of those anyways

[still rose]

That and I would need to replace my whole CPU mobo and PSU situation cuz it’s to weak for a 5090

[still rose]

Wasn’t this also 4090 situation also on launch?

[undone dock]

Not sure, I barely paid attention to the 4090 because I already had a 3090

[sterile kiln]

An isolated problem, in most cases due to the users.
It's an excellent card, with 5th gen tensor cores, overpowered, 32 gigs... And fp4, which will save a lot of time and Vram...

[still rose]

Ah I see u still have that 3090 or u switched if so how well does it perform?

[still rose]

And very card heavy 😔

[undone dock]

I'm not so sure it's isolated. That power draw is insane, and cards just keep drawing more.

[still rose]

But if I get a chance to ever try it somehow that would be cool asf

Same with trying the 9070 XT that would cool asf to try on a test game or sum

[undone dock]

I still have that 3090 but TBH a lot of crap happened to where I don't have the mobility I once had. It's still in the box until I can find someone to put my PC together or be in less pain and do it myself.

[still rose]

Ah ok that’s all good I guess used 3090 or maybe if 50 series has sum good to offer for £680 range

I can maybe buy that

If stock is lucky to me

[sterile kiln]

I've read several articles on the subject, with custommers admitting to having plugged in their connectors incorrectly and others admitting that “yes, it's not a certified connector”.

Of course, with 600 watts of TGP, the slightest mistake...

[undone dock]

Like I said, insane power draw. Eventually we'll have cables the size of ATX 12V cables sticking out of our GPUs.

We're almost there already with that 12pin cable

[sterile kiln]

I agree, but then you have to look at the bright side.
Is it better to have a 350 watt card that takes half an hour to do the job? Or a 600 watt card that only takes 10 minutes 🙂

[still rose]

£850 for 5070 TI 😟

Nvidia doesn’t really like giving us vram for the money ☹️

[undone dock]

Honestly, I miss the days of the GeForce 4 TI 4600, where it only drew power from the AGP slot. And didn't pump out massive amounts of heat, but that's just me.

Yes, I'm old.

[still rose]

@undone dock so would a build like this be like the build is should have for performance and AI performance? https://uk.pcpartpicker.com/list/6Q8FmC

Or would going Intel be “better” price value?

[undone dock]

First of all, what CPU do you currently have in your rig?

[still rose]

5600

[undone dock]

That's only 6 cores and 12 threads. That's odd, CPUs seem to be going backwards.

[still rose]

I’m pretty sure AMD 600 has always been 6 cores for longest time

Since 1600

1600, 1600af, 2600, 3600, 5600x, 7600 and now 9600

[undone dock]

Oh right, it has. I always thought that was a weird number so I always went 7 or 8.

[still rose]

Yh it’s all good but 5700x is AM4 8 core CPU

Idk if I should upgrade to that or still platform change

[undone dock]

I can't remember if the AMx are backwards compatible or not, but this is the list of CPUs I recommend sorted by lowest price.
https://uk.pcpartpicker.com/products/cpu/#C=8,64&th=16,128&sort=price&B=20000000000,65000000000&g=0&A=2400000000000,4700000000000

[still rose]

Only 1 CPU there for me and other than ryzen 7000 to 9000 is comparable with AM5 and for AM4 no one buys anything under 5000 since its so cheap and last set on that platform

[undone dock]

Yeah, their part filter is a bit annoying if you've already selected a motherboard. It's been so long since I've used PCPartPicker I forgot about it doing that.

[royal monolith]

No, i just figured you two wanted an argument, so you were perfect partners

[undone dock]

Anyways I have to head to bed now. A slight bump in CPU (to 5700) and an nVidia GPU, and you should be fine. AMD GPUs are such a pain in the butt to get working with AI generation.

[sterile kiln]

I had a Matrox MGA G200, so, i know

[still rose]

Got u

[wispy nest]

@still rose AMD CPUs are perfectly fine for image generation, comparing it with Intel doesn't make any differences, it's the GPU you need to focus on, nvidia is more ideal and also getting good support with linux too lately, however I would suggest when upgrading your system check with your AMD/Intel cpus if they are compatible with Windows 11, if that's the OS you plan to use

[still rose]

Ok thank you I figured out CPU wise I just need to find a GPU that’s decent in £600 range

[wispy nest]

Awwww poor little thing

"this author restricted video from being commented"

Clearly not knowing that it is disgustingly hard to make a whole solid style series using ai.

[copper matrix]

Or that he's simply not interested in being flamed/ your job is over vro

[wispy nest]

I was so invested in seeing war in comments

And this guy/girl just cut it

What a disappointment

[undone dock]

What's even worse is all the "News" sites pushing narratives (defending SBI as an example) and completely removing or disabling their comment sections.

[wispy nest]

for that price range you could look into NVIDIA GeForce RTX 4070

[still rose]

i just came in this dc to see whats going on and u reply at the exact time i click on the server XD

[wispy nest]

but @kat remember it will only get costlier to run AI tools locally, many people nowdays are renting clouds gpus

[still rose]

ooooo how much would that cost?

[wispy nest]

lots of different plans and range, will also depend on your needs

[royal monolith]

vast.ai seems to be the best option if you'e going to do that, but people use runpod as well

[wispy nest]

if you are going to use AI for hobby you could perhaps use on local setup, but with projects cloud gpus are more ideal

cause over a period your cloud gpu expense will surpass local set up cost

[still rose]

true i mainly use it for making cool images that all for now

but cuz pc gpu issue i have to run ai on macbook rn

[wispy nest]

rtx 4070 would be a good choice for now, that can handle sdxl, flux, sd3.5, but not sure what AI video will require in the next 6 months

[still rose]

i like video but i dont really want to do that

and for 4070 used or new

and just 4070 or 4070 super some people say super is best value but idk

[wispy nest]

word of caution, never buy used, or at least have 2 years warranty on the card

[still rose]

ah ok

[wispy nest]

s

so if things go wrong you could either replace or get refund

[still rose]

for flux on 12gb vram im using gguf model or fp8?

my fren has 3090

and use fp16 i think

[wispy nest]

i ran flux on my 12gb rtx 3060, i dont think i was happy with the speed, but most people use 4x series card for flux

you should get the GGUF version of the model if you plan to use flux on 3x series

for example gguf model that's q4_k would be around 6gb size, so that's sparing for 12gb card

i still love sdxl btw

in fact lately im playing with sdxl more than flux or sd3.5

have a look at some of the images if you search for my posts on top right search box

[still rose]

i see someone told me if i must have flux on mac on 16gb ram run q5 or sum around that but im a bit afraid of trying that on mac 😅

[wispy nest]

yes

q4 is good enough too

[still rose]

what sdxl models u recommend?

[wispy nest]

for sdxl i like pony a lot

[still rose]

ah i see

i like realism with sdxl

i use illustrious for anime and other art styles

[wispy nest]

@still rose one example of pony realism that i generated

i was playing with her eyes and such or i could make the image even more real

[still rose]

ur gens are quite nice how do u formulate prompts my brain struggles finding what to put

[wispy nest]

mix of ai and manul tweaking

[still rose]

i still need to know how to figure out how to fix eyes on comfy

i dont have auto1111 adetailer XD

[wispy nest]

you would find face detailer quite easy to use

face detailer is like adetailer but for comfyui

[still rose]

problem is what to connect to what thats what gets me lost

that and "pipe" versions and the other version

[wispy nest]

yeah it takes time to get used to it, but i would suggest always play with a pre made workflow to get an understanding what its doing and you could make minor custom edits w/o feeling overwhelmed, pick a workflow with the basic nodes

i dont have my comfyui running atm, but lemme share a few links with you with lots of pre made workflows

[still rose]

ok thank you

[wispy nest]

@still rose https://comfyanonymous.github.io/ComfyUI_examples/
https://comfyworkflows.com/
https://openart.ai/workflows/home
https://www.youtube.com/playlist?list=PLH1tkjphTlWUTApzX-Hmw_WykUpG13eza

watch them casually and at a slow pace for fun, don't make them a work so that you can enjoy learning

[still rose]

thank you my fren

[wispy nest]

yw!

@still rose another one with pony realism model that i use, look for CyberRealistic pony this model happens to be one of best on civitai for realism

[still rose]

oooo ok will do

i do like the new ui for comfy over the old makes some things easier

[wispy nest]

absolutely i love the new UI too

funny thing, it took me nearly 2 years to switch to comfyui, but once i did i never looked back, and its also funny how it happened, cause when flux came out w/ no support for it on a1111 (what i used back then) i had to look for alternative and ppl were talking about comfyui and how good it is with all the latest releases, so i dug in and in over a few weeks i started loving it 🙂

[still rose]

is using a lora trigger word as easy like auto1111 putting it in positive prompt? and any recommendations for loras for pony like detailer lora or anything like that?

[still rose]

ill try my best to stick to it

[wispy nest]

loras are easy pretty to use, and you won't need to actually mention the trigger words other than using the lora node, but if you wanted you could always mentione the trigger in positive prompt box

[wispy nest]

do you have comfyui up atm?

[still rose]

yes

[wispy nest]

ok give me 5-10 mins i'll whip up a workflow with lora, and face detailer for you

[still rose]

thank you ^^

[wispy nest]

are you using sdxl or flux atm?

[still rose]

SDXL illustrious

[wispy nest]

nice

ok gimme a few mins

[wispy nest]

@still rose done, all you need to do is open the image in new tab and save the files and load it up on your comfyui .. you will likely have some missing custom nodes but that's very easy to fix by going into comfyui manager and install missing nodes.

[still rose]

thank you possibly one more question whats the diff between gguf q4 and q4 ks and km?

[wispy nest]

q_k_s are standard quantized models, q_k_m are mixed .. get the M version if available, if not grab k_s

[still rose]

ah ok

[wispy nest]

if you are missing the upscale model that i'm using in that workflow, here is the link to it..
https://huggingface.co/uwg/upscaler/blob/main/ESRGAN/4x_foolhardy_Remacri.pth
the model goes into the following folder..
ComfyUI\models\upscale_models

logging out for now, if you have any question about the workflow feel free to ask, and i will follow up when i log in later

[still rose]

have a great day and thx again

[sterile kiln]

Flux Q8 works very well with 12 gigs of Vram, depends on what GPU you have.

obviously, a 3060 is a dogshit

[undone dock]

I for one hate the new UI, it breaks a lot of stuff. On the old one I can pin nodes to groups, on the new one nodes can only be pinned to the canvas.

Not to mention the massive Pin Icon that can't be turned off.

[undone dock]

Other good ones to look for are Crystal Clear XL / Crystal Clear One, and Everclear PNY by Zovya. There's also a good one called "mobius".

[still rose]

so ive tried it but after the gen on k sampler no matter the model its just solid black output

[undone dock]

You probably need the fp16 VAE fix located here.
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix

Filename is "sdxl.vae.safetensors"

[still rose]

Ah ok ty

[still rose]

its still black 😟

nvm

im silly let me try again

yh its still black

[undone dock]

It might just have to wait until you have that nVidia card, or follow CS1o's guide for Zluda.

[still rose]

i might have $9-10 left in my vast acc

u think that would be enough for some gens

but idk what system i could rent for that much

[undone dock]

I have no idea, I do everything local.

No mucking around trying to set up outside systems correctly (Jupyter Notebooks, etc), just download, move models, and run.

[wispy nest]

have you tried using this vae? https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

koitenshin updated you, if its still rendering black, try restarting comfyui

curious from the conversation above, aren't you using nvidia?

[wispy nest]

it's AMD gpu, can't say i know much about AMD GPUs 😐

[feral lagoon]

How can I generate more characters in this same style? Any good guides on this?

I am also interested in cleaning up the quality of the image and animating the image if possible.

[wispy nest]

imagine a society so twisted that a freak pic like this is accepted and pics of girls with little bit of skin are knee jerk resentment, i don't understand the freak mentality

and you'd wonder why criminals run free and society is so upside down

[lavish garden]

Prompt: https://civitai.com/images/59861874

---

"Profane Radiance" - OC

[still rose]

Well like mention before due to the GPU issue I’m using MacBook

[wispy nest]

i see, im not familiar with macbook either, but it's possily due to cuda configurations why you are seeing black outputs

[still rose]

Shoot ☹️ are some of the nodes cuda only? Or is a comfyUI issue

[undone dock]

macbook issue most likely

[still rose]

😔

I may have to use vast or runpod till I can get a Nvidia GPU

I’ve seen modded 22gb vram 2080 TIs but idk how good they are either may get 4070 or 4070 super like u guys mentioned

[undone dock]

I wouldn't trust anything modded. And anything 8GB + VRAM will let you run SDXL. You only need more than that if you're trying to run some absolutely insane models like video.

[still rose]

Would like to run flux and it’s fine tunes from civit

[undone dock]

my 8GB can run FLUX if I use the quantized Q4 (NF4) version. The models just have to be able to fit in your VRAM.

[still rose]

4060 ti with 16gb VRAM seemed interesting also

[undone dock]

SD3 2B Medium I can run, but I'm not sure about Illustrious (IIRC it's still SDXL based), and no idea at all about SD3.5

[still rose]

Man Nvidia 40 series increased in price in the UK

[undone dock]

Wow, looked it up on ebay (co.uk obviously) and even a 3060TI will run you over 300 GBP.

[copper matrix]

If you can run sdxl, you can run illustrious

Getting the same speeds

[still rose]

Yh 😭

Idk what I can do

I guess renting GPUs 😭

Until I get a good price and save up

[crimson pivot]

what card specifically is that , out of interest

[undone dock]

EVGA 3060 TI (the super small one that can fit in a NUC9i9QNX)

[undone dock]

https://www.techpowerup.com/gpu-specs/evga-rtx-3060-ti-xc.b8459

That's the card I have.

The only way I can upgrade from this GPU is with a Gigabyte WINDFORCE OC GeForce RTX 4060 Ti 16 GB Video Card, it's about 200mm long.

[crimson pivot]

thanks, pretty encouraging to see it run on one of those.

[sterile kiln]

Q4 & NF4_v2 are different.
Q4 is quantized version, with no text encoders.
NF4 is 4bit quantization (normal float), with a text encoder inside.
But NF4_v2 is poor, far away from Q8 one by example

[crimson pivot]

yeah thats why i asked what card. when i saw "NF4" i thought that was some new format the 50x0's supported

i'm running Q4 on a larger card and curious about options for increasing throughput (e.g. finding something for a spare slot in a spare machine )

[vivid widget]

hey guys I want to generate images like this but not getting any good results. Can someone suggest a model for this with an example prompt ? Thanks 🙂

[crimson pivot]

defnitely getting things like that out of flux with prompts like "classical pillars surrounding a pool with overgrown vegetation and flowers"

[vivid widget]

ok lemme try that,, thanks

[crimson pivot]

i'm generating images like these with randomized prompts , they're pretty straightforward, along the lines of the example i gave above

[vivid widget]

cool, i am generating an image with the example you gave me.

lets see what i get

i would say this is pretty close

but the one i sent got a lot of details

is there a way to achieve that ?

[crimson pivot]

ammend it iwth 'circular pool', "flowers in ornate plant pots", etc

[vivid widget]

okay

[crimson pivot]

i enjoy generating in batches with random variation, let it surprise me with combinations rather than me trying to get a specific result

[vivid widget]

ahh i see

lemme try that too

[crimson pivot]

http://104.238.170.161/textures/fluxai/mixed_objs_arch/20250225T020525Z_1.jpg this one has the curved bit on top .. let me grab it's prompt

[willow oriole]

Hi,
im new to sd and the ai world.
tried to create anime characters with Anything model but for some reason im getting really bad ones probably becuase of prompet right?
can anyone tell me how to make it better?

also its not taking into account the photo that i gave controlNet

[shell zinc]

i'm generating images like these with randomized prompts , they're pretty straightforward, along the lines of the example i gave above

[sterile kiln]

and, disable preprocessor, because this skeleton is already preprocessed.

AND, you need a controlnet openpose model, compatible with SDXL.
spoiler: there's nothing but crap models with Openpose on SDXL

[misty hollow]

Дюймовочка

Do her in red

[copper matrix]

🔴

[misty hollow]

Cat in black costume

[copper matrix]

🐈‍⬛

Alright enough joking, check #artisan-faq

[misty hollow]

😂😂😂😂 thank you! I am New , nothing understand

[tall haven]

never stop doing this bit

[copper matrix]

Its my sole purpose

💃

[tall haven]

as it should be 🙏

[copper matrix]

Finished my BEANS, time to get cooking. I saw theres a new video model out there requiring 6gb vram👀

480p or som tho

[tall haven]

deadass?

share

[copper matrix]

Sec im on da phone, their hub says 6, reddit said 8

Released like 2.5 hes ago ish

Theres probably no nodes for it though

[tall haven]

how big do you think it is

eh ill just look into it when you send me the link

[copper matrix]

Id say average but ive been told its big

[tall haven]

hear that often?

[copper matrix]

No

[tall haven]

youre so real for doubling down like that

[copper matrix]

Sometimes you just gotta hear it from a real one
🗣️ 🪞

[tall haven]

this is going on the wall of shame btw

https://tenor.com/view/jarvis-screenshot-jarvis-screenshot-and-crop-out-of-context-crop-cropped-gif-26975564

[dry blaze]

When you wonder why half your nodes break as newer cuda was faster, and thought installing pytorch 12.4 added all you need.. I got quite some time ahead of me, as this rubbish is slow!

[wispy nest]

pytorch 12.5 is the good one

they added a way to do torch.compile on only a part of the DiT model and its rly fast

like torch.compile in a few seconds

but uh

it only works for me sometimes. like most things

[dry blaze]

As in it will fail sometimes? Or only be faster sometimes?

And i'm still learning what does what. What do you mean by compiling a part of DiT model? And what's DiT model? :P

[crimson pivot]

DiT = "diffusion transformer"

older diffusion models use U-nets, i think a "DiT" is a more advanced replacement for the u-net idea (?) (transformers vs convolutional nets for the denoising part)

[dry blaze]

Gotcha. And it does that when you diffuse/process a image/video? Or in what situation is it torch.compiling?

[crimson pivot]

i get the impression "diffusion transformers" (being better at "long range dependencies") are more used in the video models (as well as some of the newer image generators). it's just a description of the engine doing the bulk of the work

there's several parts to these models... [1] CLIP encoder (understanding the text) .. [2] VAE (describing a compressed 'latent' image space) .. [3] and the actual diffusion U-Net or DiT (the lions share of it) doing the actual generation,. "Latent diffusion" = generating images in a compressed image space. The original diffusion models worked in pixel space and were much more expensive.

i'm seeing seperate AI files for each part described above, i think they are shared between models.

(like you can have multiple versions of the u-net or DiT using the same CLIP encoder & VAE)

someone feel free to correct me

[dry blaze]

[sterile kiln]

you're not supposed to install the CUDA Toolkit. There are dependency scripts for that.

[dry blaze]

Could you post one as an example? As i tried to install the pytorch cuda script, but all dependencies was already there. Yet the kijai video and 3d wrapper would not work until i added cuda toolkit for 12.4

[wispy nest]

if it works it will be faster

but it often doesn't work

DiT is the image model essentially

models have blocks so it does only some of the blocks

compile is just
way to make your image model go fast

that's pretty much all it does, for end user

DiT is like Flux or SD3

isn't everyone insecure? is that bad, or is it just how people are?

[dry blaze]

I really need to get a program that makes the nvme a cache for a hdd so i can store all the models and stuff on the hdd, and whenever using a new model and ssd is full, replace least used file(s) with the new file from the hdd lol

[undone dock]

You just need more storage in general. Those drives are very very low in total size.

[wispy nest]

i wonder cause im not familar with mac environment, but i have an idea, can you right click on facedetailer and bypass it, then try to gen a pic?

[still rose]

Ok

[wispy nest]

something i noticed about illustrious, style transfer doesnt work too well with it

or maybe i need to increase the weight more

but either way illustrious would work bit differently

two very different image outputs using illustrious and pony with style transfer

[wispy nest]

eat ze bugs

[rain gazelle]

nice consistency on this character... just prompt descriptors, or something else?

[wispy nest]

generic simple prompt with ipadapter

basically transfering style

[rain gazelle]

I never really learned how to use ipadapter 😐

[wispy nest]

its quite powerful for consistency

as well as creating consistent variations

[rain gazelle]

I switched to SwarmUI a while back, will have to see if it is available.

It's for certain models only, yes? Like, only SDXL based maybe?

[wispy nest]

you must know of mateo ... here's a video talking about ipadapter and ofc he's the creator of ipadapter https://youtu.be/gmwZGC8UVHE

[rain gazelle]

yeah familiar, I looked at it a while back from one of his videos, recollecting...

[wispy nest]

its easy to set it up, comfyui manager would let you install the ipadapter, not sure about other ui since i only use comfy since after flux release

[rain gazelle]

SwarmUI runs on top of comfy so, minimally, it'll have manual support.

[undone dock]

Sounds like a great way to break things with conflicting requirements.

[royal monolith]

this is why you create a private discord and store all your stuff you're not currenly using in it. unlimited storage, free, and if you have nitro, you can upload hugefiles

[copper matrix]

100% this lmao. works great for smaller stuff

[undone dock]

Imagine if the ComfyUI Loader nodes actually said when you last used that specific checkpoint / Lora / Embedding, etc.

[dry blaze]

Yep lol.

[dry blaze]

I know lol. It's why i got full nitro. And i don't use discord for that. Too messy. I use trilium for text based notes.

Way cleaner

[royal monolith]

you might try it. you can create channels for each theme for your images, if you make loras, create a channel for the loras, upload your data set, upload your lora, upload your caption file, etc.

[undone dock]

And don't forget to pin messages with important info to the channels.

[copper matrix]

Obsidian md for me

[dry blaze]

Also, what do you mean upload hung files?

[royal monolith]

huge

[undone dock]

https://tenor.com/view/thats-what-she-said-what-she-said-dirty-joke-joke-laugh-gif-15661968

[dry blaze]

Og, got my near 100TB server for that

[royal monolith]

you got a raid array?

[wispy nest]

discord don't let me upload my upscaled images lol

[royal monolith]

you need nitro for that

[wispy nest]

by the weirdest coincidence I got given a week of nitro

[royal monolith]

i was gonna gift you nitro, but you never accepted my friend request

[wispy nest]

if you release a cool AI thing I will add as friend 😄

[jagged halo]

Hey everyone!

I have a T-rex head sprite, but I need help using AI to generate the rest of its body while ensuring it matches the head’s style accurately. I want the AI to maintain consistency in proportions, shading, and details.

I’m considering tools like Stable Diffusion (ControlNet/Inpainting), RunwayML, or Adobe Firefly, but I’d appreciate recommendations on the best approach. If anyone has experience with AI-assisted sprite completion or knows a good workflow, I’d love your input!

I’ve attached my sprite sheet here. Let me know what you think or if you have any tips. Thanks in advance!

[wispy nest]

have you tried outpainting

[jagged halo]

Thanks for the suggestion! Which program or tool are you referring to for outpainting? I want to make sure I'm using the best method. Also, do you have any experience with this or know the success rate of getting a consistent result, especially for sprites? I'd appreciate any insights!

[west galleon]

can i ask for help here?

[wispy nest]

yeah sure

[wispy nest]

you could try comfyui

success rate is 100% at this point
if the tools are used properly

you can essentially make any image

[west galleon]

im trying to make a lora of markiplier and the latest one still is bad

it looks like him in an uncanny kinda way, like a twin

[wispy nest]

I'm in the UK so I can't give deepfake advice sorry
cos its not legal here anymore they passed a bill outlawing it

[west galleon]

oh sorry ig

[wispy nest]

US has not passed a similar bill
but surprisingly neither has the EU

you would have thought EU would regulate stricter on this

[west galleon]

the US is weird rn, coming from a citizen

[wispy nest]

[west galleon]

well, are you knowledgable abt loras period?

[wispy nest]

yeah I can give advice in general
the main thing is to use flux and settings that are somewhat like this:

alpha = grid search from alpha = 0.5*rank to 2*rank
learning rate = grid search from 1e-6 to 1e-3
weight decay = grid search from 0 to 1e-4, possibly up to 1e-2
gradient clipping = grid search from 1 - 2, possibly up to 5
optimiser = adamW
schedule = cosine
warmup = 2.5% to 10% of steps```

those are my notes

I basically took a bunch of Arxiv papers where they gave the method

and I combined the methods together

[west galleon]

ty, ill try tomorrow

[wispy nest]

ok nice

grid search just means "try some values" to be clear

it could be a different method to grid search

[royal monolith]

but YOU'RE the one that makes cool AI things

[wispy nest]

I actually had a look just now but could not work out how to add friend

[west galleon]

omg

k, so what trainers do yall use for loras? cuz im having hella problems finding one i want

cant install the pre-requirements for kohya gui, one trainer doos not export the finished file as a .safetensors files (meaning i cant use it), cant get easy scripts to train, and hollow strawberry's colab just failed on me today

[undone dock]

Hey @copper matrix, you can probably help this guy...

[copper matrix]

i mean yeah i could but i use koyhaSS and civitAI tbh lmao

[west galleon]

hi :3, sorry it took me forever to reply, power outage

[remote flame]

do yall know what this painting thingy is?

[royal monolith]

luca tacos on replicate

[undone dock]

You've never heard of Wolfram Alpha?

[west galleon]

this dude?
https://replicate.com/lucataco

[wispy nest]

yeah

replicate use this Cog thing they made

its rly good

[royal monolith]

Yup. He is beast programmer

[copper matrix]

No women today?

[wispy nest]

maybe later, im playing with this SDXL model

[terse glacier]

你好

Hello

Hello

[wispy nest]

hmmm

i clicked on his chineese words on top and chose to translate, and it translates inside discord

[copper matrix]

Hi

[wispy nest]

like i see 3 lines of hello

[copper matrix]

Oh same, maybe google translate

[wispy nest]

yeah it is but im surprised it didnt open any dialog box,

this discord platform is a hybrid of all chat apps

[terse glacier]

Hello everyone, learn from the boss

[wanton yoke]

sneak peak

[lavish garden]

👀 From: https://civitai.com/user/AXTAXALEN

---

🖼️ :https://civitai.com/images/59668186

[wispy nest]

trump spanked zelensky's butt red

[sterile kiln]

Custom LLM integrated in VSCode : ✅

[ripe scroll]

Hey guys I was wondering if anyone had a workflow for creating consistent character with different poses. I also want to integrate my logo in the image like on his shirts. Im using comfyui

For example this guy in different scenes different poses with my logo. A gme themed mascot

[placid notch]

11

[lofty meadow]

你好

[clever fog]

monochrome-candid-but-with-red-lips-and-blue-eyes, me

[sage ginkgo]

Can someone regen this image in a way it matches my pfp without changing the pose and his hair and looks

[clever fog]

candid, me-as-bipedal-bioluminescence

[clever fog]

candid, me-as-bipedal-holographic-bioluminescence, cutesie, open-air

[undone dock]

Trying to get that Cortana look huh?

[copper matrix]

@modern barn

Still get torch audio errors but it still boots and works fine

[plush seal]

??

[clever fog]

nope, making my own versions on holograms. cortana is your way of thinking on holograms, not mine.

[still rose]

Guys I’m considering getting a 5070 but it having 12gb vram would I have any major issues with performance with image gen for flux and stuff

[undone dock]

If you're only using it for AI Gens, the 5K series is good. But if you play any games that use PhysX, you're in for a bad time.

[wispy nest]

are you sure?

here is what im seeing about rtx 5070 and games requring physx

[undone dock]

https://www.pcguide.com/news/heres-every-game-affected-by-rtx-50-series-dropping-physx-32-bit-support/

[wispy nest]

since youre buying a new card, I would recommend looking into RTX 5070 Ti which may cost a bit more but not too much from 5070 and Ti version has 16gb vram

[still rose]

Oh ☹️ I do a lot of esports games and was considering playing some single player since with new GPU I’ll be able to for once

[undone dock]

The 4K series is still great, and will be even cheaper now that the 5K series is out.

[wispy nest]

didnt Nvidia drop dedicted physx support since gtx 900 series?

most of the new nvidia gpus should deliver good support for physx

[still rose]

What’s recommended from u to buy £660 is the highest I’ll go

[still rose]

Apparently games are struggling with performance I’ve read somewhere cuz of physx on the 50 series being mid support or sum

[wispy nest]

well don't suppose any of the new rtx now have any dedicated physx support either

[still rose]

Well I mean last gen 40 series was fine

50 series seems to struggle

So idk

What’s going on there

[wispy nest]

i still think if you can save up a bit more you should aim for 5070 Ti

take a bit more time to save up

[still rose]

That and if not I’ll guess 40 series

[undone dock]

Do you know the maximum length your case can handle?

[still rose]

I have Corsair 275r airflow

In white

[undone dock]

Says it supports 370mm length cards, so you shouldn't have any troubles with any recommendations, now to look.

[still rose]

Yeah I don’t know Nvidia GPUs so I’ll need some recommendation help

[undone dock]

It looks like the best card you can get for that pricepoint is going to be a 4060 TI with 16GB of VRAM.

[wispy nest]

in case you decided to buy 5070 Ti you would need more power for the PSU ... at least 750w

remember its not only about the vram .. the cuda core makes a huge difference

[still rose]

For £660 range it’s 4060 TI Nvidia pricing really is dodgy recently

[undone dock]

This is the best one I found, and it will match your theme. https://www.amazon.co.uk/dp/B0CBSM9KMY

[still rose]

I do plan to upgrade that

I was told I can get used 3090 for that price £660 ish @undone dock

[undone dock]

You probably can on Ebay, if you're lucky.

[still rose]

Definitely doable

[undone dock]

Definitely, just saw this one. https://www.ebay.co.uk/itm/235987314402

[still rose]

Yeah

[wispy nest]

dont forget gpu crisis few years ago, where ppl were buying out and farming bcoin with the cards, later selling their used cards when the bcoin trend popped, careful about used cards

[undone dock]

Apparently ebay uk works differently than normal ebay? Or they changed things... that Listing is out of stock, but it's a normal listing, odd

[wispy nest]

cat for as pricey item as gpu card i dont see the reason for used cards, you should look for a new one with warranty, if you cant buy a new one right away i suggest save up a bit and not rush

have seen cases where people bought rtx cards that went glitchy w/o warranty and stuck with a junk

[still rose]

I see thank you I’ve been saving for a while so the closer I get to the target price the more I get excited

I’ll try holding off

[wispy nest]

aim for 5070 Ti since you mentioned you were ready for 5070

the reason i mention the Ti it has more cuda power and vram

and Ti shouldn't cost too much from 5070

[undone dock]

If you can, set an alarm on your phone for a while from now telling you where you hid the money, and then hide it. It sounds like you have the "it's burning a hole in my pocket" feeling.

[still rose]

True

[hazy jasper]

where can i generate? is it free?

[still rose]

It’s either I can buy it and really excited and then after it’s micro depression cuz I spent so much or idc. Or it’s looking at the price then wondering if I should spend the money 😭 @undone dock

[undone dock]

I understand perfectly *stares at Brand New still in box 3090 on his shelf*

I want to use it, but we have very bad electrical wiring here. I'm worried I'll burn the place down.

[still rose]

It’s with everything I buy over £60 for some reason not matter how much I have it’s always that XD

[undone dock]

That's because 60 quid is a lot of money.

[still rose]

Indeed and my family quite a few years ago struggled I mean now it’s mainly fine my family but Yh effects always still stands

[undone dock]

This is true. I grew up so dirt poor I still have original Playstation (PSX) games without a single scratch or blemish on the discs.

Because I knew if anything happened to them, I couldn't replace them.

[still rose]

Same boat hugs

[undone dock]

But yeah, once you're actually able to use that new GPU you'll be wondering why it took you so long. "Why didn't I do this sooner?"

[still rose]

XD for sure many said once u get a GPU that can run single player games u will play them more hopefully thats also true

Because rn my 6600 xt struggles with games now

On launch I loved it

Now

Unfortunate

[undone dock]

I tried the AMD stuff once, I don't know if I got a bad manufacturer (SapphireTech) or what, but it worked until it didn't. So I sent it back. They refurbed it and sent it back. It lasted a week. =/

[still rose]

Also @wispy nest 16gb on 5070 TI shouldn’t be a issue for AI right I wanna run flux loras and fine tunes and the stuff with comfy

[still rose]

Unlucky card what card was it?

Sapphire also known to be one of the best AMD AIB cards

[undone dock]

TBH I can't remember right now, gimme a second.

[still rose]

Ok my friend

[wispy nest]

all rtx 5x series should be ok for flux, but i suggest rtx 5070 Ti for even better cuda and 16gb vram.

[undone dock]

It only took a few minutes, but it was this card. https://www.techpowerup.com/gpu-specs/sapphire-dual-x-r9-270-platinum-edition.b2663

[still rose]

Ah Yh older AMD GPUs are bad

I thought it was a more modern one

[undone dock]

It left such a bad taste in my mouth that I haven't tried a Radeon card since, and I first played games on a Radeon card in my Electronics / Engineering classes at school.

[still rose]

Ever since AMD did the bigger gen clutch vs Nvidia in RX 6000

They been cooking

[undone dock]

Trust me, it was an amazing upgrade from what I had. The Quality Control was horrible though, it died within 6 months.

[still rose]

Yh older GPUs from them is questionable

Glad they are doing decent now

I wonder if they hard cooking like Nvidia since the start

How the market share would look like

In 2025

[undone dock]

If AMD had actually kept pushing the envelope we'd all be using AMD cards for AI generation instead of nVidia.

Instead they floundered and are now playing catch-up

[still rose]

@undone dock https://uk.pcpartpicker.com/list/4yC7MC is this fine? I have a 5600 CPU I’m worried even I decide to go 1440p I’ll still need CPU upgrade

Keep in mind

I’ll keep on eye with pricing for GPU

That’s just there to show what the upgrade will look like

My case seems to be in trouble rn so I may have to change it

[undone dock]

I don't see a MB on that list so I can't tell you whether or not that RAM is good. For Ryzen CPUs you actually want all sockets populated.

[still rose]

I’m upgrading from current I have B450 board

So that ram is fine

I’m just doing a ram replacement from 16gb Corsair to that

[undone dock]

if it's an mATX board, you should be fine. but if it's a full size ATX board you'll want a 4 stick kit.

[still rose]

I have mATX it’s a ASU’s tuf B450m gaming plus

I think it’s II

Not sure

[wispy nest]

dark blue background using detail daemon ..

[plush seal]

@wispy nest please help

[plush seal]

@wispy nest how do you create images?

[wispy nest]

https://youtu.be/g8UlYE_HM2M

[plush seal]

Is there a phone version ?

[modern barn]

@modern barn big question, that message you sent me about stable not working on 50 series gpu can you ask the guy how do i install the pytorch cuda 12.8 for the stable diffusion forge web ui

Question i was asked

[wispy nest]

there are clouds services you could use where you don't have to host anything and just use your phone to access the website to generate images/videos

[plush seal]

favor ?

[wispy nest]

will cost you few cents for each images, do google search and check out the different services plans out there

[plush seal]

A few cents is lowballing for your work

[wispy nest]

if you want detailed control over your images then you have to host your own workflow in local pc

some people also use clouds services where they can get to choose which model they want to use and control over thier prompts

phone is not an option for that

you could however use phones to connect to your local installation within your network if you choose to

and if you dont want to deal with technical factors you could simply use services on the clouds but they are not usually uncensored, so there is a bit of restriction.

if you like nude art those are not generally possible from cloud services

@plush seal https://comfyui-wiki.com/en/install/install-comfyui/install-comfyui-on-windows

also for technical help with installation you can ask at #🤝｜tech-support

[rain gazelle]

I like this one

[rain gazelle]

And this one. Hair came out neat.

[wispy nest]

ty

was experimenting with image filter nodes

[rain gazelle]

what are image filter nodes? 🤔

[wispy nest]

basically image enhancer that allow tweaking sharpness, contrast etc

[wispy nest]

i use that as part of the workflow

much better control over outputs instead of prompting for those parameters

[plush seal]

This goes a lot deeper then what I thought 😬

[wispy nest]

you can start with the basics and keep learning and you will enjoy the ride

[plush seal]

I have nothing but my phone

[rain gazelle]

just use dalle-3 and call it a day. 😛

[undone dock]

https://tenor.com/view/sml-cody-shes-a-brick-house-brick-house-supermariologan-gif-27161129

[hushed mirage]

What do you use to make stuff like this?

[quiet current]

Not gonna discuss this here. no NSFW allowed, I m removing it.

[hushed mirage]

Where should i discuss this?

[quiet current]

Not in this discord ^^".

[cedar tiger]

Design a minimalist and clear logo for a ceramic tile brand named "TILE SUPPLIES." Use a letterform logo style incorporating the initials "TS" in a creative and modern way. The design should be sleek, professional, and visually appealing, reflecting the durability and elegance of ceramic tiles. Use a refined color palette that suits the industry, with clean typography and subtle tile-inspired elements to enhance the brand’s identity. The logo should be versatile for branding across packaging, signage, and digital platforms.

#artisan-1 Design a minimalist and clear logo for a ceramic tile brand named "TILE SUPPLIES." Use a letterform logo style incorporating the initials "TS" in a creative and modern way. The design should be sleek, professional, and visually appealing, reflecting the durability and elegance of ceramic tiles. Use a refined color palette that suits the industry, with clean typography and subtle tile-inspired elements to enhance the brand’s identity. The logo should be versatile for branding across packaging, signage, and digital platforms.

[copper matrix]

#artisan-faq

[warm umbra]

Accidentally discovered my NSFW model can generate AMAZING artwork

[wispy nest]

yeah is common cos they get trained on more photos

[buoyant fossil]

@warm umbra Loved that little mushroom house on the left, sooo... after some photoshop work...

[signal orchid]

现代都市公园，阳光柔和，绿地和树木，背景有摩天大楼。一个白人男性（30岁，浅棕色短发，浅蓝色衬衫，灰色休闲裤，微笑）和一个黑人男性（30岁，黑色卷发，深绿色针织衫，卡其色长裤，开朗笑容）站在桥上，影子交融，背景有鸽子和国际象棋棋盘，插画风格，低饱和色调。

/generate

现代都市公园，阳光柔和，绿地和树木，背景有摩天大楼。一个白人男性（30岁，浅棕色短发，浅蓝色衬衫，灰色休闲裤，微笑）和一个黑人男性（30岁，黑色卷发，深绿色针织衫，卡其色长裤，开朗笑容）站在桥上，影子交融，背景有鸽子和国际象棋棋盘，插画风格，低饱和色调。

[warm umbra]

How do you make a 3D model of stuff like this? That would be useful I think

[warm umbra]

生成器在此通道中不起作用。

Used Google Translate to tell him the generator doesn't work in this channel]

[buoyant fossil]

I can show you if you like. Just DM me. @warm umbra

Its kind of an in-depth workflow.

[warm vector]

pufferfish 3D

[south temple]

wan model is taking a while... woof

8 min render

testing the fp8

way faster

[hardy crest]

wan model is taking a while... woof

[fleet goblet]

Wan 2.1 1.3b makes images too

[pale steeple]

any worfklow for pony/illustrious that i can have that will generate the images really quickly? any helps pls?

[pale steeple]

wan 2.1 vae please? where can i find?

[pale steeple]

anyone can troubleshoot pls? i would really appreciate it

[wispy nest]

if you are using comfyui refer to this link i posted earlier #🏞｜general-with-images message

[ancient oyster]

hi! anyone knows what can I do to avoid these "dirty" colors in details and make my image cleaner? thx for any help!

[grim vortex]

I also have this question. Sometimes the screen is very dirty. Have you tried adding prompts and increasing weights to solve it?

[main venture]

#artisan-faq

[wise turret]

Anyone know how I can get my result looking more like the first image?

https://civitai.com/models/1262970?modelVersionId=1424253

Here's the link to the lora

[copper matrix]

lower lora weight a lil maybe

[quiet current]

You'd have to use the same workflow as they are (it's embedded in the picture, it's a comfyUI one) if you want to recreate the same picture.

[wispy nest]

codeformer for quick face fix is a much easier option since it takes lot less memory than facedetailer

a simple example of it using codeformer with flux/shuttle3

[high perch]

That is an amazing render

Just goes on to show the cross modal understanding of the model

What a flex

[fleet goblet]

happy cub e

[wispy nest]

YWNBAArtist

[sterile kiln]

John Lennon?

[lavish garden]

Prompt: https://civitai.com/images/61833626

---

"Sunlit Shores" - Nick Wilde (Zootopia) | Request

[warm umbra]

Here's one of my most hilarious fails.

[stone narwhal]

What can I do to keep my overall image in tact, but make it less blurry/higher res?

[quiet current]

img2img with low denoiser value