are illustrious models the same as sdxl models?
are illustrious models the same as sdxl models?
FHi everyone, and hello @Dr. Furkan Gözükara . Thank you so much for all the work and research you share, it's amazing!
Can I ask your advice on what's the best way to train a LORA for SDXL on cloud as of today (best quality possible)?
Can I ask your advice on what's the best way to train a LORA for SDXL on cloud as of today (best quality possible)?
Fwhich illustrious models?
Fyou can use both Kohya and OneTrainer
Fi recommend onetrainer on massed compute
Fthis is good
F
NI am curious, what exactly are you training here? Such results are already possible witrh Flux (de-distilled) out of the box, right?
.No version is fine tuned this way.
.YouTubePronto CG Tutor
A demo of an image generation of my custom Flux dev de-distilled fine tune trained on high resolution images.
.Ever had the issue when you upscale too high with sd ultimate upscale that your skin details get washed out and lose detail? With this way of training it doesn’t
Nyes, I see. It makes sense! May I ask how are you training it: Are you using custom prompts like ohwx? how does such fine-tuning work, like style training? With every database image captioned in details or not at all? I am trying to understand how the AI model "thinks" in such situations. With custom faces or objects it's straight forward, but with whole "concepts" like this?
MDoes anyone know this error when generating model in kohya
MDoes anyone know this error when generating model in kohya
raceback (most recent call last):
File "<frozen runpy>", line 198, in _run_module_as_main
File "<frozen runpy>", line 88, in _run_code
File "C:\Kohya\kohya_ss\venv\Scripts\accelerate.EXE__main.py", line 4, in <module>
File "C:\Kohya\kohya_ss\venv\Lib\site-packages\accelerate\commands\accelerate_cli.py", line 19, in <module>
from accelerate.commands.estimate import estimate_command_parser
File "C:\Kohya\kohya_ss\venv\Lib\site-packages\accelerate\commands\estimate.py", line 34, in <module>
import timm
File "C:\Kohya\kohya_ss\venv\Lib\site-packages\timm__init.py", line 2, in <module>
from .models import create_model, list_models, is_model, list_modules, model_entrypoint, \
File "C:\Kohya\kohya_ss\venv\Lib\site-packages\timm\models__init__.py", line 28, in <module>
from .maxxvit import *
File "C:\Kohya\kohya_ss\venv\Lib\site-packages\timm\models\maxxvit.py", line 225, in <module>
@dataclass
^^^^^^^^^
File "C:\Python311\Lib\dataclasses.py", line 1232, in dataclass
return wrap(cls)
^^^^^^^^^
File "C:\Python311\Lib\dataclasses.py", line 1222, in wrap
return _process_class(cls, init, repr, eq, order, unsafe_hash,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Python311\Lib\dataclasses.py", line 958, in _process_class
cls_fields.append(_get_field(cls, name, type, kw_only))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Python311\Lib\dataclasses.py", line 815, in _get_field
raise ValueError(f'mutable default {type(f.default)} for field '
ValueError: mutable default <class 'timm.models.maxxvit.MaxxVitConvCfg'> for field conv_cfg is not allowed: use default_factory
22:42:12-093630 INFO Training has ended.
raceback (most recent call last):
File "<frozen runpy>", line 198, in _run_module_as_main
File "<frozen runpy>", line 88, in _run_code
File "C:\Kohya\kohya_ss\venv\Scripts\accelerate.EXE__main.py", line 4, in <module>
File "C:\Kohya\kohya_ss\venv\Lib\site-packages\accelerate\commands\accelerate_cli.py", line 19, in <module>
from accelerate.commands.estimate import estimate_command_parser
File "C:\Kohya\kohya_ss\venv\Lib\site-packages\accelerate\commands\estimate.py", line 34, in <module>
import timm
File "C:\Kohya\kohya_ss\venv\Lib\site-packages\timm__init.py", line 2, in <module>
from .models import create_model, list_models, is_model, list_modules, model_entrypoint, \
File "C:\Kohya\kohya_ss\venv\Lib\site-packages\timm\models__init__.py", line 28, in <module>
from .maxxvit import *
File "C:\Kohya\kohya_ss\venv\Lib\site-packages\timm\models\maxxvit.py", line 225, in <module>
@dataclass
^^^^^^^^^
File "C:\Python311\Lib\dataclasses.py", line 1232, in dataclass
return wrap(cls)
^^^^^^^^^
File "C:\Python311\Lib\dataclasses.py", line 1222, in wrap
return _process_class(cls, init, repr, eq, order, unsafe_hash,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Python311\Lib\dataclasses.py", line 958, in _process_class
cls_fields.append(_get_field(cls, name, type, kw_only))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Python311\Lib\dataclasses.py", line 815, in _get_field
raise ValueError(f'mutable default {type(f.default)} for field '
ValueError: mutable default <class 'timm.models.maxxvit.MaxxVitConvCfg'> for field conv_cfg is not allowed: use default_factory
22:42:12-093630 INFO Training has ended.
Fuse Python 3.10 and reinstall
MI sent you a message with the printout I have Python 10..
but its automatic installer uses python 11 alone, which already comes with it
but its automatic installer uses python 11 alone, which already comes with it
Fautomatic installer the one i have
Fjust clones
Fand starts the installer of the kohya
Fyou can edit bat file to see
�Can you tell a bit more like - did you create the folders yourself and not use the dataset preparation tab at all, did you set the number of reapeats as 1 for both, did you set the number of repeats in the folder name only like this 1_tok man, and did you make any other changes to the configs?
�Bro what you are doing is impressive I didn't quite understand your pixel chunks etc. would love to learn more about how are you doing this.
SHello @Dr. Furkan Gözükara I am folowing your instructions for the windows version of rope and I'm getting and error for ffmpeg "PS C:\Users\Shadow> ffmpeg
ffmpeg : The term 'ffmpeg' is not recognized as the name of a cmdlet, function, script file, or operable program.
Check the spelling of the name, or if a path was included, verify that the path is correct and try again.
At line:1 char:1
ffmpeg : The term 'ffmpeg' is not recognized as the name of a cmdlet, function, script file, or operable program.
Check the spelling of the name, or if a path was included, verify that the path is correct and try again.
At line:1 char:1
- ffmpeg
~~- CategoryInfo : ObjectNotFound: (ffmpeg:String) [], CommandNotFoundException
- FullyQualifiedErrorId : CommandNotFoundException
Salso same error when I do nvcc --version "PS C:\Users\Shadow> nvcc --version
nvcc : The term 'nvcc' is not recognized as the name of a cmdlet, function, script file, or operable program. Check
the spelling of the name, or if a path was included, verify that the path is correct and try again.
At line:1 char:1
nvcc : The term 'nvcc' is not recognized as the name of a cmdlet, function, script file, or operable program. Check
the spelling of the name, or if a path was included, verify that the path is correct and try again.
At line:1 char:1
- nvcc --version
- ~~~~
- CategoryInfo : ObjectNotFound: (nvcc:String) [], CommandNotFoundException
- FullyQualifiedErrorId : CommandNotFoundException
Sis there a fix for this?
Fyes follow every step of this tutorial dont skip
FYouTubeSECourses
If you are interested in using AI, generative AI applications, open source applications in your computer, than this is the most fundamental and important tutorial that you need. In this tutorial I show and explain how to properly install appropriate Python versions accurately, how to switch between different Python versions, how to install diffe...
Fseperate folders under main parent folder
�example?
SThank you! Also, will you be doing a video of the recent VisoMaster soon?.
Fi dont have any
Fif gets published
Fjust messaged authors from twitter to make published i hope they listen
�I mean like this - tok man (parent folder) then in that folder have two other folders one 1_tok man and other 1_tok woman?
�and do you set the repeats to 1 like this 1_tok man?
F1_tok man and other 1_tok woman?
Fbut dont use tok for both
�okay so do we use dataset preparation tab to create the parent folder? and should it have any unique token as name and repeats or it can be anything?
SDr Furkan i am triyng to do onetrainer test, for flux, i put batch in 2 and resolution in 768, do you think this could dowungrade to much the final flux lora model? is a lora that im training
Sthis are the settings i using
4I send a message to their discord 
Dhttps://www.reddit.com/r/comfyui/s/lxJ2jKAiuG
So already beaten flux in terms of multiple characters without bleeding?
So already beaten flux in terms of multiple characters without bleeding?
Reddit
Explore this post and more from the comfyui community
Fyes it will reduce quality
Fi read how he did
Fhe is randomly assigning lora weights to lora layers and trying . but still surprisingly good worked
Sdr furkan in ai processes such as training flux models loras, etc. will the 5070 ti have more power for these types of tasks than a current 4090?
Fnope. flux requires more vram and you need to do more optimization on lower vram
Feven 5080 is inferior to 4090
F5090 or else 4090 is king
F3090 may even beat 5080 
due to lack of vram
due to lack of vram