Closed MaestroFury closed 7 months ago
Which of those folders did you set as "Image folder" in the gui?
Suppose you had a directory structure like /somewhere/dresslora/IMG/20_MorroccanDress
and within _20MorroccanDress will be the images.
With that given directory structure, the path to the IMG folder would be the one you'd set as the Image folder.
And what file formats are your images in? Are they png, jpg/jpeg, or webp?
What do you mean Kohya doesn't see it? post a screenshot so we get what you mean. To me the GUI shows the paths gradually (and not all of the available paths in my directory tree), so I just copy and paste the correct absolute path in the GUI
Well I managed to finally get Kohya to see the images, but now it won't see the caption files, which are named exactly the same as the image files, but with a .txt
and then it just dumps this at the end and never starts.
Traceback (most recent call last):
File "C:\Users\Administrator\Desktop\Kohya\kohya_ss\sd-scripts\sdxl_train.py", line 792, in
What an incredibly frustrating experience this has been so far. Can someone PLEASE just share their setup json file with me? Do I have some setting on I shouldn't? I'm not getting any errors, just a bunch of warnings and no clear signal what I'm doing wrong.
Here is the full dump, from start to finish
`16:03:33-720094 INFO accelerate launch --num_cpu_threads_per_process=2 "C:\Users\Administrator\Desktop\Kohya\kohya_ss/sd-scripts/sdxl_train.py" --bucket_no_upscale --bucket_reso_steps=64 --cache_latents --cache_latents_to_disk --enable_bucket --min_bucket_reso=256 --max_bucket_reso=2048 --gradient_checkpointing --learning_rate="0.0" --learning_rate_te1="1e-05" --learning_rate_te2="1e-05" --logging_dir="C:/Users/Administrator/Desktop/Facial_Sheet_Masks/log" --lr_scheduler="constant" --lr_scheduler_num_cycles="10" --max_data_loader_n_workers="0" --resolution="1024,1024" --max_train_steps="9000" --mixed_precision="bf16" --optimizer_args scale_parameter=False relative_step=False warmup_init=False --optimizer_type="Adafactor" --output_dir="C:/Users/Administrator/Desktop/Facial_Sheet_Masks/model" --output_name="Facial_Sheet_Mask" --pretrained_model_name_or_path="C:/Users/Administrator/Desktop/Webui - Forge/webui/models/Stable-diffusion/realvisxlV40_v40Bakedvae.safetensors" --save_every_n_epochs="1" --save_model_as=safetensors --save_precision="bf16" --train_batch_size="1" --train_data_dir="C:/Users/Administrator/Desktop/Facial_Sheet_Masks/img" --xformers 2024-03-24 16:03:43 INFO prepare tokenizers sdxl_train_util.py:135 INFO Using DreamBooth method. sdxl_train.py:140 INFO prepare images. train_util.py:1469 INFO found directory train_util.py:1432 C:\Users\Administrator\Desktop\Facial_Sheet_Masks\img\20_Facial_Sheet_Ma sks contains 45 image files WARNING No caption file found for 45 images. Training will continue without train_util.py:1459 captions for these images. If class token exists, it will be used. / 45枚の画像にキャプションファイルが見つかりませんでした。これらの画像につ いてはキャプションなしで学習を続行します。class tokenが存在する場合はそれを使います。 WARNING C:\Users\Administrator\Desktop\Facial_Sheet_Masks\img\20_Facial_Sheet_Ma train_util.py:1466 sks\1.jpg WARNING C:\Users\Administrator\Desktop\Facial_Sheet_Masks\img\20_Facial_Sheet_Ma train_util.py:1466 sks\10.jpg WARNING C:\Users\Administrator\Desktop\Facial_Sheet_Masks\img\20_Facial_Sheet_Ma train_util.py:1466 sks\11.jpg WARNING C:\Users\Administrator\Desktop\Facial_Sheet_Masks\img\20_Facial_Sheet_Ma train_util.py:1466 sks\12.jpg WARNING C:\Users\Administrator\Desktop\Facial_Sheet_Masks\img\20_Facial_Sheet_Ma train_util.py:1466 sks\13.jpg WARNING C:\Users\Administrator\Desktop\Facial_Sheet_Masks\img\20_Facial_Sheet_Ma train_util.py:1464 sks\14.jpg... and 40 more INFO 900 train images with repeating. train_util.py:1508 INFO 0 reg images. train_util.py:1511 WARNING no regularization images / 正則化画像が見つかりませんでした train_util.py:1516 INFO [Dataset 0] config_util.py:544 batch_size: 1 resolution: (1024, 1024) enable_bucket: True network_multiplier: 1.0 min_bucket_reso: 256 max_bucket_reso: 2048 bucket_reso_steps: 64 bucket_no_upscale: True
[Subset 0 of Dataset 0]
image_dir:
"C:\Users\Administrator\Desktop\Facial_Sheet_Masks\img\20_Facial_Sheet_M
asks"
image_count: 45
num_repeats: 20
shuffle_caption: False
keep_tokens: 0
keep_tokens_separator:
caption_dropout_rate: 0.0
caption_dropout_every_n_epoches: 0
caption_tag_dropout_rate: 0.0
caption_prefix: None
caption_suffix: None
color_aug: False
flip_aug: False
face_crop_aug_range: None
random_crop: False
token_warmup_min: 1,
token_warmup_step: 0,
is_reg: False
class_tokens: Facial_Sheet_Masks
caption_extension: .caption
Forge/webui/models/Stable-diffusion/realvisxlV40_v40Bakedvae.safetens
ors
INFO building U-Net sdxl_model_util.py:192
INFO loading U-Net from checkpoint sdxl_model_util.py:196
2024-03-24 16:03:49 INFO U-Net:
all my settings as instructed by some online tutorials..
Please ignore me, for anyone who was struggling, please try and follow this tutorial TO THE LETTER.
https://techtactician.com/how-to-train-stable-diffusion-lora-models/#step-1
I am successfully training SD1.5 right now, and I hope now that I have everything working, the leap from 1.5 to SDXL will not be too far.
I am super excited to be on my way finally. And I hope to help other struggling people reach this point!
Well I managed to finally get Kohya to see the images, but now it won't see the caption files, which are named exactly the same as the image files, but with a .txt
In case you haven't set that one yet, there's a setting called "Caption Extension" which defaults to .caption
as the file extension. You'll want to change it to .txt
Sorry for the flood of tickets lately...
So now my final problem is, I can't get Kohya to actually see the images to train. I followed all kinds of youtube tutorials, I read the EDG guide on training LORAs for SDXL, the naming structure in the EDG tutorial was simple
MoroccanDress -> IMG -> 5_MorroccanDress
Where 5 is the number of times to train an image. I'm trying to create a lora with 20 steps, so I imagine mine would be 20_MorroccanDress
So I created that structure, Kohya doesn't see it.
I found another thread here which says I should use __ before the folder name, so I tried that too, and still nothing..
So i'm completely lost. What is the current expectation folder naming wise for Kohya to see my images. Both training and Regulisation?