I hope I can get this Runpod error fixed today, though.
I hope I can get this Runpod error fixed today, though.
ESomeone have any idea what prompts I can use to have my subject be generated whole on one picture? My training pictures are fine and presents whole object. But when I'm trying to generate it is just zoom in or in a different perspective
IIn my experience "full body shot" or "full body portrait" but you also need to change ratio, like 512x768. In general it is not easy for me either, most images are closeups.
Fhello welcome
Fyes i am
Fstart collecting training images
Fhopefully my style teaching video will be published this week
Fhi can you show the runpod error you are getting
IFirst of all thank you for all the great work, but especially for being available to give support to your viewers.
IMany others don't even bother replying to comments on YouTube, so we appreciate that.
IBut just 5 minutes ago I think I found what I did wrong. I was updating requirements file, not requirements versions.
II will do another session again, and let you know...
Zif i do lower learning rate for longer, will i get better details on my training
Fbut it will take longer time
Zlike i got really good results, but she's like, "The eyes are a little off" and what not
Zit's really close tho
Zmay train with 0.0005 next time or somethin
IFor the sake of reporting, this was the error I was getting. Exception initializing accelerator: Accelerator.init() got an unexpected keyword argument 'project_dir'
IMaybe it can be useful. In my case after many hours online, I got easily distracted and didn't see that there are two files: requirements & requirements versions
IIf the training starts and finishes correctly, I will share my experience with you Dr Furkan.
IFirst update: In Dreambooth optimizer 8bit Adam is not there anymore. I have 1. Torch AdamW 2. Lion
Fyes this requires experimentation
Fyep this is version error
Fi spent like 5 hours first time i was trying to fix it 
IOK, the training started, so I can confirm that error above was my mistake.
INow with the exact same settings, as in your video, I'm getting this error:
IException training model: ''NoneType' object is not subscriptable'.
II will try to figure out what to change
Fcan you show entire message
IDo you want me to copy/paste message from terminal here? Or in private?
IIt's long.
IPlease check DM
Fyou can paste here
I Running training
Num batches each epoch = 0
Num Epochs = 200
Batch Size Per Device = 1
Gradient Accumulation steps = 1
Total train batch size (w. parallel, distributed & accumulation) = 1
Text Encoder Epochs: 150
Total optimization steps = 0
Total training steps = 0
Resuming from checkpoint: False
First resume epoch: 0
First resume step: 0
Lora: False, Optimizer: Torch AdamW, Prec: fp16
Gradient Checkpointing: False
EMA: True
UNET: True
Freeze CLIP Normalization Layers: False
LR: 1e-06
V2: False
Steps: : 0it [00:00, ?it/s]Traceback (most recent call last):
File "/workspace/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/ui_functions.py", line 727, in start_training
result = main(class_gen_method=class_gen_method)
File "/workspace/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/train_dreambooth.py", line 1448, in main
return inner_loop()
File "/workspace/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/memory.py", line 119, in decorator
return function(batch_size, grad_size, prof, *args, **kwargs)
File "/workspace/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/train_dreambooth.py", line 1200, in inner_loop
latents = batch["images"].to(accelerator.device)
TypeError: 'NoneType' object is not subscriptable
Steps: : 0it [00:00, ?it/s]
Restored system models.
Duration: 00:00:08
Num batches each epoch = 0
Num Epochs = 200
Batch Size Per Device = 1
Gradient Accumulation steps = 1
Total train batch size (w. parallel, distributed & accumulation) = 1
Text Encoder Epochs: 150
Total optimization steps = 0
Total training steps = 0
Resuming from checkpoint: False
First resume epoch: 0
First resume step: 0
Lora: False, Optimizer: Torch AdamW, Prec: fp16
Gradient Checkpointing: False
EMA: True
UNET: True
Freeze CLIP Normalization Layers: False
LR: 1e-06
V2: False
Steps: : 0it [00:00, ?it/s]Traceback (most recent call last):
File "/workspace/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/ui_functions.py", line 727, in start_training
result = main(class_gen_method=class_gen_method)
File "/workspace/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/train_dreambooth.py", line 1448, in main
return inner_loop()
File "/workspace/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/memory.py", line 119, in decorator
return function(batch_size, grad_size, prof, *args, **kwargs)
File "/workspace/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/train_dreambooth.py", line 1200, in inner_loop
latents = batch["images"].to(accelerator.device)
TypeError: 'NoneType' object is not subscriptable
Steps: : 0it [00:00, ?it/s]
Restored system models.
Duration: 00:00:08
Fdo pip reinstall -r requirements.txt again?
Falso you fixed versions right?
Fyou are using filewords?
IYes, versions fixed. Accelerate updated to 0.17 this morning was 0.16
F"I confirm, there is issue. I had the same error when some of "filewords" had the UTF16 LE encoding. With UTF8 it works properly."
FGitHub
File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/extensions/sd_dreambooth_extension/scripts/dreambooth.py", line 455, in start_training result = main(config, use_txt2img=us...
II replicated your tutorial, just using my own token.
II will try the reinstall and update you
Fdo you use file captions? filewords?
Fi asked to extension developer
INo I don't think so. I'm only using images, renamed with a little description.
INow after the I reinstalled the requirements, I got a new error, so I can't launch web ui:
II gues the GPU is not avilable on Runpod, after I stopped it. GPUs have high availability, but my pod is set to 0 GPU.
IThank you!
