Software Engineering Courses (SECourses)•2y ago

Does anyone else ever run into an issue where upon generating an image the facial similarity is pret

Does anyone else ever run into an issue where upon generating an image the facial similarity is pretty good but after using adetailer its not even close? And if so how did you fix it?

AAIGambino How many steps do you guys do for supir?

Furkan Gözükara SECourses•3/4/24, 2:03 AM

Sshortdwarf Does anyone else ever run into an issue where upon generating an image the facia...

Furkan Gözükara SECourses•3/4/24, 2:03 AM

do you set custom prompt?

Furkan Gözükara SECourses•3/4/24, 2:03 AM

to adetailer

shortdwarfOP•3/4/24, 2:03 AM

yes, ive tried a few things but the one i would expect to work (since again, its pretty accurate upon ggenerating the image) is ohwx woman

Sshortdwarf yes, ive tried a few things but the one i would expect to work (since again, its...

Fabricatedgirls•3/4/24, 2:09 AM

if you are using SDXL are you setting the resolution to 1024x1024

shortdwarfOP•3/4/24, 2:23 AM

im using a custom resolution of 1024x1536 however ive had success with it with other dreambooth models ive made. and ive set the adetailer resolution to match at 1024x1536, yes

AIGambino•3/4/24, 2:23 AM

Anything for linear stage 2?

Sshortdwarf im using a custom resolution of 1024x1536 however ive had success with it with o...

Fabricatedgirls•3/4/24, 3:40 AM

does it make a difference if you set the adetailer resolution to 1024x1024, i've read that departing from the default res causing image degradation

shortdwarfOP•3/4/24, 4:04 AM

I trained the DB model on 1024x1536 so arguably 1024x1024 would be departing from the default res, but ill do some testing now on that resolution and edit this with results
edit: same result, adetailer comes back looking nothing like the initial results, much further from what the DB model was trained on

Fabricatedgirls•3/4/24, 4:30 AM

can i see the difference?

Fabricatedgirls•3/4/24, 4:30 AM

why train so high res?

Fabricatedgirls•3/4/24, 4:31 AM

are you generating the image at 1024x1536 and using adetailer at 1024x1536 too?

shortdwarfOP•3/4/24, 4:33 AM

I want to produce images at 1024x1536 so best to train for the resolution im producing. and yes i tried that and it actually worked out well, however i noticed that separate CFG scale was checked and although it was set to the same as the original generation i wonder if that was causing the error, experimenting with 1024x1024 for adetailer now

shortdwarfOP•3/4/24, 4:39 AM

Based on new tests, it seems to have been a bug with CFG scale even though it was set to the same amount as the initial generation

Fabricatedgirls•3/4/24, 4:40 AM

what denoising strength are you using

shortdwarfOP•3/4/24, 4:40 AM

i jump around from 0.4 to 0.6 but usually stay on the lower side

Fabricatedgirls•3/4/24, 4:51 AM

0.6 is definitely going to change it a lot

Fabricatedgirls•3/4/24, 4:51 AM

even 0.4

Fabricatedgirls•3/4/24, 4:52 AM

set it down to like 0.25 and see how it looks

AIGambino•3/4/24, 5:04 AM

What's the best ai tool for colorization of black and white photos? Does anyone know

shortdwarfOP•3/4/24, 5:24 AM

Maybe someone has a better recommendation, but if you use canny with controlnet and describe what you want, should do a good job adding color

Fabricatedgirls•3/4/24, 6:12 AM

https://www.youtube.com/watch?v=5W2_avVdMHM

YouTubeHow Do?

Colorize and Restore Old Images

This video provides a guide for colorizing and restoring old images using Unsampling and ControlNets in ComfyUI with Stable Diffusion. This is a follow-up of my video on "Reimagining" images and uses the same workflow with a few tweaks. For a more in-depth look at how that workflow works, you can watch that video here: https://www.youtube.com/wa...

Fabricatedgirls•3/4/24, 6:12 AM

https://www.youtube.com/watch?v=dHQ4rkAGghs

YouTubeAlbert Bozesan

How to Colorize Photos with AI - Stable Diffusion + ControlNet Tuto...

To try everything Brilliant has to offer—free—for a full 30 days, visit http://brilliant.org/AlbertBozesan/ . The first 200 of you will get 20% off Brilliant’s annual premium subscription.

Links:
How to Install and Use Stable Diffusion (June 2023) - Basic Tutorial
https://youtu.be/nBpD-RbglPw

ControlNet: https://github.com/Mikubill/sd-webui-co...

Fabricatedgirls•3/4/24, 6:12 AM

there's some youtube videos about it (using controlnet)

FFurkan Gözükara SECourses do you set custom prompt?

Dreamer@TDFilmStudio•3/4/24, 9:42 AM

Hi, please check out this workflow using CCSR as the first step, and then SUPIR as the second step to achieve superior results. Please see if such a workflow can be integrated into your gradio - https://www.reddit.com/r/StableDiffusion/comments/1b50sp0/ccsr_vs_supir_upscale_comparison_portrait/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

From the StableDiffusion community on Reddit

Explore this post and more from the StableDiffusion community

Dreamer@TDFilmStudio•3/4/24, 9:49 AM

Excerpt: "Conclusion

CCSR = high fidelity, but low quality (no fine details, washed out, softens image)

SUPIR = low fidelity (hallucinates too much), but very high quality (reintroduce fine details/texture)

CCSR ➜ SUPIR combo is simply mind blowing as you can see in example k, l, m. This combo gave the best fidelity and quality balance. CCSR is able to reconstruct as faithfully as possible even a destroyed jpg while SUPIR can fill in all the lost details. Prompting is not necessary but recommended for further accuracy (or to sway specific direction.) If I do not care about fidelity, then SUPIR is much better than CCSR."

DDreamer@TDFilmStudio Hi, please check out this workflow using CCSR as the first step, and then SUPIR ...

Furkan Gözükara SECourses•3/4/24, 11:10 AM

ye looking pretty nice

protector131090•3/4/24, 11:11 AM

Hello! i have a problem with SD img2img batch speed. I launch sd 1.5 Img2img with controlnet tile 200 images. at first it is super fast but the longer it goes the slower it becomes... the render itself is same speed but swithcing from 1 img to next is beocming slower over time. What can be the reason?

protector131090•3/4/24, 11:11 AM

i got 4090 and SD installed on fast SSD

DDreamer@TDFilmStudio Hi, please check out this workflow using CCSR as the first step, and then SUPIR ...

Furkan Gözükara SECourses•3/4/24, 11:13 AM

i think i can make a standalone installer for CCSR so that can be used to generate initial image sounding good?

Pprotector131090 Hello! i have a problem with SD img2img batch speed. I launch sd 1.5 Img2img wit...

Furkan Gözükara SECourses•3/4/24, 11:14 AM

very possibly a bug

FFurkan Gözükara SECourses i think i can make a standalone installer for CCSR so that can be used to genera...

Dreamer@TDFilmStudio•3/4/24, 11:14 AM

Yeah that would be great to try out

Furkan Gözükara SECourses•3/4/24, 11:14 AM

in a1111

protector131090•3/4/24, 11:14 AM

yes

Pprotector131090 yes

Furkan Gözükara SECourses•3/4/24, 11:14 AM

i would report this. i think there can be some memory leak

Furkan Gözükara SECourses•3/4/24, 11:14 AM

right now i am trying to make kaggle notebook of SUPIR

FFurkan Gözükara SECourses right now i am trying to make kaggle notebook of SUPIR

Deleted User•3/4/24, 11:31 AM

Nice

Furkan Gözükara SECourses•3/4/24, 12:03 PM

ok it uses too much ram

Furkan Gözükara SECourses•3/4/24, 12:03 PM

so crashes notebook

DDreamer@TDFilmStudio Hi, please check out this workflow using CCSR as the first step, and then SUPIR ...

Pelolisu•3/4/24, 12:17 PM

i read this yesterday! the things move fast... haha

Furkan Gözükara SECourses•3/4/24, 12:25 PM

i think i found a good way to improve model loading of SUPIR

Furkan Gözükara SECourses•3/4/24, 12:25 PM

it was using 30 gb ram

Original message was deleted

Furkan Gözükara SECourses•3/4/24, 12:25 PM

nope

FFurkan Gözükara SECourses nope

VAMP•3/4/24, 12:26 PM

sorry it was a clickbait video..

VVAMP sorry it was a clickbait video..

Furkan Gözükara SECourses•3/4/24, 12:29 PM

yep

Furkan Gözükara SECourses•3/4/24, 12:36 PM

Please leave a comment : https://github.com/Fanghua-Yu/SUPIR/issues/60

GitHub

create_SUPIR_model uses more than 29 GB RAM - Thus free Kaggle note...

Currently create_SUPIR_model inside util.py loads everything into RAM first with full precision Thus it is using more than 29 GB RAM I think model half param can be given to it as well Or perhaps s...

Pelolisu•3/4/24, 2:03 PM

i think you can use fp8

Pelolisu•3/4/24, 2:04 PM

its use same stable diffusion engine that a1111

Pelolisu•3/4/24, 2:06 PM

i am busy with work jobs i cant try it yet

Pelolisu•3/4/24, 2:11 PM

you can send that A1111 PR to they or we can be the first to do this

Does anyone else ever run into an issue where upon generating an image the facial similarity is pret

Similar Threads