ComfyUI LORA training workflow
ฝัง
- เผยแพร่เมื่อ 4 ต.ค. 2024
- Comfy-UI Workflow for training LO
This workflow allows you to train a model on a certain style or on a certain object. which makes it possible to show consistent results
#comfyui #stablediffusion #ipadapter #mask #Relighting
follow me @ / pixeleasel
Mix Any Style - video and workflow
• ComfyUI Any Style comb...
workflow
drive.google.c...
Image-Captioning-in-ComfyUI
github.com/Lar...
Lora-Training-in-Comfy
github.com/Lar...
ComfyUI-WD14-Tagger
github.com/pyt...
Nice that you can explain that in 5 min while every other TH-camr babbles on and on.
yeah they try to milk their 30 fucking minutes. fuck that. this guy is a genius
FINALY A LOAR training ❤in comfyUi ! lov it ;)
THis doesn't work for a lot of people and its to do with some sort of python issues, about all i know... but it seems to be more that the person in the video knows about the issues people are having.
bro u help alot thank u hope u the best man
more than welcome 🙏
@@PixelEasel plz i have problem say:
Error occurred when executing LoRA Caption Load:
cannot access local variable 'image1' where it is not associated with a value
Been trying for days, doesnt work at all, error this and that, need more info about the error
Everything is working except the lora export, its nowhere to be found. Any help is appreciated!
the node needs an update, it seems its no longer working for anyone
agreed
broke my comfyui install. AMD users, you need to reinstall the proper pytorch rocm version after adding the extensions because one of them installs the cuda version.
I waited for this tutorial for a long time. Thank you 🙏. Can I train embeddings in the same way? If not can you please make tutorial for train embedding in comfyui?
thanks!, i'll make one on embedding
I installed it and comfyui doesn't work anymore,
did u try to update?
Same here. This workflow wrecked my ComfiUI installation.
Same. Now I need to reinstall Comfi and it's not the first time this node has given me this result.
Same
Thanks! 😁
I love your model! have you made it available for us? i would love to make a short film with it! 🙏🙏🙏
Can you show me the background change for portrait? Not product.
It’s a bit hard for me to imply nice light and shadow
i'm not sure what do you mean
@@PixelEasel Can you change the background of a portrait photo for me? Not for a product. When I do it, the shadows and lighting look awkward.
hey guys, i'd appreciate some help, i've repeated the steps, been able to caption etc, but when i queue the training, it executes very fast and i'm not able to find the lora file in the loras directory, even tho i specified it, any assistance is much appreciated!
Hi, this usually happens, the folder path is not set correctly, note that in the training node you need to set the folder that contains the data set folder
I selected the folder that contains the data set folder and still have the same issue. It runs in seconds and doesn’t produce a LoRA file.
@@MatthewHardyMusical I have the same problem. did you solve it?
@@fotszyrzk79 I was able to use Comfy to caption my photos but I couldn't get it to make the LoRA. Kohya doesn't work on my Mac so I tried Draw Things to create my LoRA but I'm not happy with the results.
@@MatthewHardyMusical Thanks, it is sad :(. For me it's been two full day struggle with this to work... I think I'll pass and check Kohya instead.
Another broken workflow. Do ANY of the SDXL trainers work anymore for comfy?
Lora Doesn't appear and train
not working
Great write-up, but something broke since you posted it. I keep getting the following on a fresh install of ComfyUI and all the nodes you suggested. Error appears in training, labelling works just fine:
ModuleNotFoundError: No module named 'accelerate'
not sure why, but pay attention to the folder name and the path of the folder
אני סובל מאותה שגיאה.. פתרת את זה?
Upgrade python to v3.12.x
it doesn't work, i got this error:
cannot access local variable 'image1' where it is not associated with a value
Same here, any solution?
I figured it out bro, you need only PNG format, you cant use JPEG or any other format, a 3 letter image format, preferably png.
@@MaghrabyANO thx a lot bro, now i have many other errors XD, gonna give up lora training on comfy
@@TheStoryteller-xyz same :D tell me what error maybe i can help
@@MaghrabyANO raise RuntimeError('Error(s) in loading state_dict for {}:
\t{}'.format(
RuntimeError: Error(s) in loading state_dict for CLIPTextModel:
Unexpected key(s) in state_dict: "text_model.embeddings.position_ids".
Hey~ Spelling mistake on the subject title~
thx!
Can this work as product placement using the same data training ?
yep
got the error "Error occurred when executing LoRA Caption Load". Any solutions?
the most important thing, is the folder path, and the name of the data folder, check it again
@@PixelEasel My problem with WD14 Tagger node, maybe the node I used was different, it only takes ONE image at a time... so when Lora Caption Load sends a list of file to it, it broke... I had to use another node in between to send each file to the WD14 Tagger
nice sir, resect you.
יפה מאוד, זה יעבוד לי גם ל FLUX ?
ModuleNotFoundError: No module named 'transformers.modeling_outputs'
Ensure you have the transformers library installed and updated to the latest version. You can do this by running: pip install --upgrade transformers
@@PixelEasel ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
datasets 2.13.1 requires dill=0.3.0, which is not installed.
yolov5 7.0.13 requires seaborn>=0.11.0, which is not installed.
optimum 1.18.0 requires transformers[sentencepiece]=4.26.0, but you have transformers 4.42.3 which is incompatible.
I have this error:
Lora-Training-in-Comfy/sd-scripts/train_network.py
The following values were not passed to `accelerate launch` and had defaults used instead:
`--num_processes` was set to a value of `1`
`--num_machines` was set to a value of `1`
`--mixed_precision` was set to a value of `'no'`
`--dynamo_backend` was set to a value of `'no'`
I got error message like this: The following values were not passed to `accelerate launch` and had defaults used instead:
`--num_processes` was set to a value of `1`
`--num_machines` was set to a value of `1`
`--dynamo_backend` was set to a value of `'no'`
To avoid this warning pass in values for each of the problematic parameters or run `accelerate config`.
E:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\diffusers\utils\outputs.py:63: UserWarning: torch.utils._pytree._register_pytree_node is deprecated. Please use torch.utils._pytree.register_pytree_node instead.
torch.utils._pytree._register_pytree_node(
Do you know how to fix? Thank you!
nop, ill try to search for an answer
@@PixelEasel same here. Any news?
I got an error saying: Error occurred when executing LoRA Caption Save:
list index out of range on the LoRA Caption Save node, any help? @pixeleasel
You have to pay close attention to the path of the folder, and the name given to the folder, it should be in this format 1_name
work with SDXL?
yep