Thanks for the help! Last time I got an error at 32 steps. I combined padding & cropping to bring al

DD4UNTLESS Thanks for the help! Last time I got an error at 32 steps. I combined padding & ...

F

Furkan Gözükara SECourses•12/23/24, 12:35 AM

great

F

Furkan Gözükara SECourses•12/23/24, 12:35 AM

250 steps = 250 *7 samples

F

Furkan Gözükara SECourses•12/23/24, 12:36 AM

1750 steps at batch size 1

C

ctz00•12/23/24, 6:26 AM

hi, i am doing character training in sdxl dreambooth in kohya but the results are like this:

C

ctz00•12/23/24, 6:26 AM

C

ctz00•12/23/24, 6:27 AM

--save_model_as="safetensors" \
--full_bf16 \
--output_name="momo_sdxl_dreambooth" \
--lr_scheduler_num_cycles=8 \
--max_data_loader_n_workers=0 \
--learning_rate=1e-04 \
--lr_scheduler="constant" \
--train_batch_size=1 \
--max_train_steps=10000 \
--save_every_n_steps=500 \
--mixed_precision="bf16" \
--save_precision="bf16" \
--cache_latents \
--cache_latents_to_disk  \
--optimizer_type="Adafactor" \
--optimizer_args \
--bucket_reso_steps=64 \
--noise_offset=0.0 \
--gradient_checkpointing \
--bucket_no_upscale \
--train_text_encoder \
--sample_every_n_epochs=1 \
--sample_every_n_steps=500 \
--sample_sampler="euler" \

--save_model_as="safetensors" \
--full_bf16 \
--output_name="momo_sdxl_dreambooth" \
--lr_scheduler_num_cycles=8 \
--max_data_loader_n_workers=0 \
--learning_rate=1e-04 \
--lr_scheduler="constant" \
--train_batch_size=1 \
--max_train_steps=10000 \
--save_every_n_steps=500 \
--mixed_precision="bf16" \
--save_precision="bf16" \
--cache_latents \
--cache_latents_to_disk  \
--optimizer_type="Adafactor" \
--optimizer_args \
--bucket_reso_steps=64 \
--noise_offset=0.0 \
--gradient_checkpointing \
--bucket_no_upscale \
--train_text_encoder \
--sample_every_n_epochs=1 \
--sample_every_n_steps=500 \
--sample_sampler="euler" \

--save_model_as="safetensors" \
--full_bf16 \
--output_name="momo_sdxl_dreambooth" \
--lr_scheduler_num_cycles=8 \
--max_data_loader_n_workers=0 \
--learning_rate=1e-04 \
--lr_scheduler="constant" \
--train_batch_size=1 \
--max_train_steps=10000 \
--save_every_n_steps=500 \
--mixed_precision="bf16" \
--save_precision="bf16" \
--cache_latents \
--cache_latents_to_disk  \
--optimizer_type="Adafactor" \
--optimizer_args \
--bucket_reso_steps=64 \
--noise_offset=0.0 \
--gradient_checkpointing \
--bucket_no_upscale \
--train_text_encoder \
--sample_every_n_epochs=1 \
--sample_every_n_steps=500 \
--sample_sampler="euler" \

--save_model_as="safetensors" \
--full_bf16 \
--output_name="momo_sdxl_dreambooth" \
--lr_scheduler_num_cycles=8 \
--max_data_loader_n_workers=0 \
--learning_rate=1e-04 \
--lr_scheduler="constant" \
--train_batch_size=1 \
--max_train_steps=10000 \
--save_every_n_steps=500 \
--mixed_precision="bf16" \
--save_precision="bf16" \
--cache_latents \
--cache_latents_to_disk  \
--optimizer_type="Adafactor" \
--optimizer_args \
--bucket_reso_steps=64 \
--noise_offset=0.0 \
--gradient_checkpointing \
--bucket_no_upscale \
--train_text_encoder \
--sample_every_n_epochs=1 \
--sample_every_n_steps=500 \
--sample_sampler="euler" \

`

C

ctz00•12/23/24, 6:27 AM

how to cure?

Cctz00 hi, i am doing character training in sdxl dreambooth in kohya but the results ar...

F

Furkan Gözükara SECourses•12/23/24, 8:49 AM

hi follow this video every step : https://youtu.be/0t5l6CP9eBg

YouTubeSECourses

Full Stable Diffusion SD & XL Fine Tuning Tutorial With OneTrainer ...

In this tutorial, I am going to show you how to install OneTrainer from scratch on your computer and do a Stable Diffusion SDXL (Full Fine-Tuning 10.3 GB VRAM) and SD 1.5 (Full Fine-Tuning 7GB VRAM) based models training on your computer and also do the same training on a very cheap cloud machine from MassedCompute if you don't have such compute...

F

Furkan Gözükara SECourses•12/23/24, 8:49 AM

it shows configs

T

Timson•12/23/24, 11:29 AM

@Furkan Gözükara SECourses how does new L40 compares to A6000 on training tasks?

TTimson <@205854764540362752> how does new L40 compares to A6000 on training tasks?

F

Furkan Gözükara SECourses•12/23/24, 12:04 PM

little bit faster

F

Furkan Gözükara SECourses•12/23/24, 12:04 PM

more disk space

F

Furkan Gözükara SECourses•12/23/24, 12:04 PM

disk space really worths it

Y

Yobin•12/23/24, 2:12 PM

@Furkan Gözükara SECourses so the dreambooth tab under the kohya gui for flux is not actually dreambooth but it is just native fine tune?

Y

Yobin•12/23/24, 2:17 PM

also it seems not using captioner (like joycaption) is ok with flux, any upside of using captioner still?

T

TheToday•12/23/24, 3:02 PM

Is there any news on Lora's training for Hunyuan?

YYobin <@205854764540362752> so the dreambooth tab under the kohya gui for flux is not ...

F

Furkan Gözükara SECourses•12/23/24, 3:14 PM

dreambooth and fine tune are using same script

F

Furkan Gözükara SECourses•12/23/24, 3:14 PM

only dataset preprationg slightly changes

F

Furkan Gözükara SECourses•12/23/24, 3:14 PM

dreambooth without reg images = fine tuning

TTheToday Is there any news on Lora's training for Hunyuan?

F

Furkan Gözükara SECourses•12/23/24, 3:14 PM

i didnt have chance to look at yet

T

Tonnu•12/24/24, 6:34 AM

Anyone have experience creating a finetuned model that is trained on a logo and can generate apparel / clothing with that logo?

T

Timson•12/24/24, 11:07 AM

L40 only 17% slower than L40S at fine tuning, really impressed

T

Timson•12/24/24, 11:26 AM

Looking at this: tips-info, I assumed it would be terrible

TTonnu Anyone have experience creating a finetuned model that is trained on a logo and ...

F

Furkan Gözükara SECourses•12/24/24, 11:46 AM

usually not working that way great. but if you train logo you can apply it on random stuff

F

Furkan Gözükara SECourses•12/24/24, 11:46 AM

or you can try in context lora of flux

TTimson L40 only 17% slower than L40S at fine tuning, really impressed

F

Furkan Gözükara SECourses•12/24/24, 11:46 AM

amazing

TTimson Looking at this: https://discord.com/channels/772774097734074388/106779985541570...

F

Furkan Gözükara SECourses•12/24/24, 11:48 AM

it could be either that guy didnt do accurate testing or it was runpod low quality pods :d

D

D4UNTLESSOP•12/24/24, 1:04 PM

For flux lora training on kohya, do I need to convert my webp images to png before starting the training? Or are webp images also supported?

DD4UNTLESS For flux lora training on kohya, do I need to convert my webp images to png befo...

F

Furkan Gözükara SECourses•12/24/24, 1:11 PM

i didnt test but i expect to be supported. still best is png

FFurkan Gözükara SECourses i didnt test but i expect to be supported. still best is png

D

D4UNTLESSOP•12/24/24, 1:14 PM

Thanks I'll try with webp. Also, my flux training was successful. Only problem is that I cannot extract lora from it for some reason. I tried it on my local kohya installation which could be the problem. So I'll try to extract it now on cloud

DD4UNTLESS Thanks I'll try with webp. Also, my flux training was successful. Only problem i...

F

Furkan Gözükara SECourses•12/24/24, 2:12 PM

could be

FFurkan Gözükara SECourses it could be either that guy didnt do accurate testing or it was runpod low quali...

T

Timson•12/24/24, 2:23 PM

nope, 31% slower at inference

strange

FFurkan Gözükara SECourses could be

D

D4UNTLESSOP•12/24/24, 2:45 PM