Software Engineering Courses (SECourses)•14mo ago

Is it possible to train against the Mochi preview models in Kohya? https://huggingface.co/Comfy-Org/

Is it possible to train against the Mochi preview models in Kohya? https://huggingface.co/Comfy-Org/mochi_preview_repackaged/tree/main/split_files/diffusion_models

Comfy-Org/mochi_preview_repackaged at main

PPhando Is it possible to train against the Mochi preview models in Kohya? https://huggi...

Furkan Gözükara SECourses•11/8/24, 8:58 AM

i didnt see any work for it

Ddxqb <@205854764540362752> not sure if this is interesting for you, or you use Adafac...

dxqb•11/8/24, 2:53 PM

a part of this was merged into the Prodigy package and anyone can use it in other trainers

Zhigarev•11/9/24, 9:15 AM

Hi, could you tell me if I can train another character in DreamBooth using the same flux.dev model that was additionally trained on a different character? For example, I trained it on Arnold Schwarzenegger’s appearance, and now I want to create a dataset for Sylvester Stallone, or, for instance, train it in a specific style like Dark Fantasy.

ZZhigarev Hi, could you tell me if I can train another character in DreamBooth using the s...

Furkan Gözükara SECourses•11/9/24, 9:26 AM

flux gets bleed

Furkan Gözükara SECourses•11/9/24, 9:26 AM

not working as expected

Furkan Gözükara SECourses•11/9/24, 9:26 AM

but stable diffusion 3.5 large my next research hopefully

Archimiro•11/9/24, 11:16 PM

@Furkan Gözükara SECourses sorry, I'm a bit new, and I have a question about the finetunig, in XL model, could you help me?

AArchimiro <@205854764540362752> sorry, I'm a bit new, and I have a question about the fine...

Furkan Gözükara SECourses•11/10/24, 12:00 AM

hey sure

Furkan Gözükara SECourses•11/10/24, 12:01 AM

but i will be going sleep soon

FFurkan Gözükara SECourses hey sure

Archimiro•11/10/24, 12:03 AM

Umm, okay, I'll be quick. I'm trying to fine-tune an XL model, but the results from the epochs are coming out blank. My guess is that it's due to the low number of steps, which were set to 700, using a dataset of 1.5k images. I’d like to know what configuration you recommend for finetuning. If it’s too complex to address now, we can discuss it in more detail tomorrow.

AArchimiro Umm, okay, I'll be quick. I'm trying to fine-tune an XL model, but the results f...

Furkan Gözükara SECourses•11/10/24, 12:07 AM

blank sounds error

Furkan Gözükara SECourses•11/10/24, 12:07 AM

kohya best config : https://www.patreon.com/posts/very-best-for-of-89213064

Patreon

The Very Best Workflow For SDXL DreamBooth / Full Fine Tuning - Res...

Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon

Furkan Gözükara SECourses•11/10/24, 12:08 AM

onetrainer best config : https://www.patreon.com/posts/96028218

Patreon

OneTrainer Stable Diffusion XL (SDXL) Fine Tuning Best Presets | SE...

Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon

FFurkan Gözükara SECourses blank sounds error

Archimiro•11/10/24, 12:08 AM

black, error, this

FFurkan Gözükara SECourses onetrainer best config : https://www.patreon.com/posts/96028218

Archimiro•11/10/24, 12:08 AM

ooh perfect, ty

AArchimiro black, error, this

Furkan Gözükara SECourses•11/10/24, 12:08 AM

your checkpoints looks like broken

FFurkan Gözükara SECourses your checkpoints looks like broken

Archimiro•11/10/24, 12:08 AM

why?

AArchimiro why?

Furkan Gözükara SECourses•11/10/24, 12:09 AM

you got black output

Furkan Gözükara SECourses•11/10/24, 12:09 AM

which config did you use?

FFurkan Gözükara SECourses which config did you use?

Archimiro•11/10/24, 12:11 AM

use, LR 6e-3, Text encoder 3e-6, 32 Batch Size, 10 epoch, and Gradient accumulate steps 2

Archimiro•11/10/24, 12:11 AM

finetunig, illustr XL Model, in format safetensorts

AArchimiro use, LR 6e-3, Text encoder 3e-6, 32 Batch Size, 10 epoch, and Gradient accumulat...

Furkan Gözükara SECourses•11/10/24, 12:13 AM

each config unique . if you use my config

Furkan Gözükara SECourses•11/10/24, 12:13 AM

and if you want 32 batch size

Furkan Gözükara SECourses•11/10/24, 12:13 AM

multiply my learning rate size with 5.7

FFurkan Gözükara SECourses each config unique . if you use my config

Archimiro•11/10/24, 12:14 AM

the configuration of this video https://www.youtube.com/watch?v=EEV8RPohsbw, I was guided by the last tutorial I published about flux.

YouTubeSECourses

How To Do Stable Diffusion XL (SDXL) DreamBooth Training (Full Fine...

Patreon post for config files

https://www.patreon.com/posts/89213064

SECourses Discord To Get Full Support

https://discord.com/servers/software-engineering-courses-secourses-772774097734074388

My LinkedIn

https://www.linkedin.com/in/furkangozukara/

My Instagram

https://www.instagram.com/gozukarafurkan/

My Medium

@FurkanGozuka...

FFurkan Gözükara SECourses multiply my learning rate size with 5.7

Archimiro•11/10/24, 12:14 AM

AArchimiro the configuration of this video https://www.youtube.com/watch?v=EEV8RPohsbw, I w...

Furkan Gözükara SECourses•11/10/24, 12:23 AM

then checkpoints making black is the issue

Furkan Gözükara SECourses•11/10/24, 12:23 AM

did you try model without training?

Furkan Gözükara SECourses•11/10/24, 12:23 AM

that base model you are using?

FFurkan Gözükara SECourses then checkpoints making black is the issue

Archimiro•11/10/24, 12:24 AM

it is an XL model, I talked to the creator, and he recommended me not to use clip skip, and to add extra parameters to Adafactor.

J2_Nedo•11/10/24, 10:28 AM

the Mochi1 videos are really fun to create, are there any future plans for img2video?

JJ2_Nedo the Mochi1 videos are really fun to create, are there any future plans for img2v...

Furkan Gözükara SECourses•11/10/24, 12:31 PM

comfyui just told me they planning

Furkan Gözükara SECourses•11/10/24, 12:31 PM

great news

Leolis78•11/11/24, 12:06 AM

Hi, I want to train a rock band, it's a trio: vocals and guitar (leader), bass, drums. I plan to distribute the photos in folders: folder with photos of the singer, folder with photos of the bassist, folder with photos of the drummer, and folder with photos of the band complete. Has anyone trained Lora for more than one person? What photo distribution do you recommend?

Leolis78•11/11/24, 12:06 AM

dataset/
├── Band_complete/ # Group photos of the entire band
│ ├── Band_complete_001.jpg
│ ├── Band_complete_002.jpg
│ └── ... (X photos)
├── Leader/ # Individual photos of the leader
│ ├── Leader_001.jpg
│ ├── Leader_002.jpg
│ └── ... (Y photos)
├── Bassist/ # Individual photos of the bassist
│ ├── Bassist_001.jpg
│ ├── Bassist_002.jpg
│ └── ... (Z photos)
└── Drummer/ # Individual photos of the drummer
├── Drummer_001.jpg
├── Drummer_002.jpg
└── ... (W photos)

Leolis78•11/11/24, 12:08 AM

Folder Structure and Photo Distribution

dataset/
├── Band_complete/ # Group photos of the entire band
│ ├── Band_complete_001.jpg
│ ├── Band_complete_002.jpg
│ └── ... (X photos)
├── Leader/ # Individual photos of the leader
│ ├── Leader_001.jpg
│ ├── Leader_002.jpg
│ └── ... (Y photos)
├── Bassist/ # Individual photos of the bassist
│ ├── Bassist_001.jpg
│ ├── Bassist_002.jpg
│ └── ... (Z photos)
└── Drummer/ # Individual photos of the drummer
├── Drummer_001.jpg
├── Drummer_002.jpg
└── ... (W photos) Distribution of Photos

Band_complete: Approximately 30% of total photos.

Leader: 40% of total photos (assuming the leader is more prominent).

Bassist: 15% of total photos.

Drummer: 15% of total photos.

LLeolis78 Hi, I want to train a rock band, it's a trio: vocals and guitar (leader), bass, ...

Furkan Gözükara SECourses•11/11/24, 12:10 AM

you dont need

Furkan Gözükara SECourses•11/11/24, 12:10 AM

you can use .txt caption

Furkan Gözükara SECourses•11/11/24, 12:10 AM

and have all in same folder

Furkan Gözükara SECourses•11/11/24, 12:10 AM

if you gonna use repeat 1

Leolis78•11/11/24, 12:19 AM

great, so I only differentiate them with the captions? eg. leader(1).txt : nameofrockbandxyz, j0hnxyz leader vocalist, holding a microphone on a stage....

Leolis78•11/11/24, 12:24 AM

I would use a single folder, 1 nameofrock bandxyz people (1 repeat_triggerword class)

wardensc2•11/11/24, 4:32 AM

Hi @Furkan Gözükara SECourses have you tried training a style with dreambooth. I wonder what is different when a checkpoint learning a style compare with lora. Does it learn faster so we can train with lower epoch

harap.•11/11/24, 5:00 AM

hi, I trained this sdxl checkpoint in the past https://civitai.com/models/570998/cocaxl-01 but it was impossible to generate character sheets with it. So I trained another checkpoint based on https://civitai.com/models/293331/wildcardx-xl-turbo, the previous one had been trained on juggernaut xl. Not character sheets work better, but there is a problem.

harap.•11/11/24, 5:01 AM

for the same prompt the new model outputs this

harap.•11/11/24, 5:01 AM

the old one outputed this

harap.•11/11/24, 5:02 AM

I don't mind the output, but it doesn't render the clothing as well as the old model

harap.•11/11/24, 5:03 AM

I am talking about this

harap.•11/11/24, 5:04 AM

for example for the footwear, when I input "man wearing ((opinci-like moccasins)) " the old model creates pretty accurate opinci, the new one, not as much, also it creates 2 people where as I only asked for one. And in general it doesn't stick to the prompt as well.

Is it possible to train against the Mochi preview models in Kohya? https://huggingface.co/Comfy-Org/

Similar Threads

Similar Threads

Similar Threads