[Bug]: I made a dreambooth for the v2.1 512 and it's doesn't work on automatic1111

[alexfredo]

Is there an existing issue for this?

• [X] I have searched the existing issues and checked the recent builds/commits

What happened?

Hi I made a dreambooth for the v2.1 512 base model on hugging face and it's doesn't work on automatic1111 but it's work on hugging face though. I enter the "style" name in my prompt and I have something completely different not the style I have trainned, I use the 512 .yaml, please helps !!

[ClashSAN]

what is the error message?

[alexfredo]

There's no error message I just don't have the style I have train

[alexfredo]

For example my command style in the prompt is "skchan style" it's work on hugging face but on automatic1111 it's completely ignored

[ClashSAN]

is it 2.1 or 2.0? for me it says "v2-512" https://huggingface.co/spaces/MirageML/dreambooth

I don't know how the dreambooth on huggingface works, maybe test on 1.5 and see if its working first

[alexfredo]

it's 2.1 I used this for generate the checkpoint https://huggingface.co/spaces/multimodalart/dreambooth-training

[alexfredo]

I guess 2.1 dreambooth are not supported for the moment ?

[alexfredo]

I have noticed that the dreambooth 2.1 is highly superior to the dreambooth 1.5 so I really hope this will be fixed soon. I get medium result with 1.5 and I get jaw dropping result with 2.1 on hugging face

[alexfredo]

On the stable diffusion github there's write that "To enable fp16 (which can cause numerical instabilities with the vanilla attention module on the v2.1 model) , run your script with `ATTN_PRECISION=fp16 python <thescript.py>"

Can someone explain where I need to put ? : ATTN_PRECISION=fp16 python <thescript.py>

Thanks

[alexfredo]

I made another checkpoint with the 2.1 and this time I have a error message when I add --no-half :

Error completing request Arguments: ('tanaka2 style', '', 'None', 'None', 20, 0, False, False, 1, 1, 7, -1.0, -1.0, 0, 0, 0, False, 192, 192, False, 0.7, 0, 0, 0, 0.9, 5, '0.0001', False, 'None', '', 0.1, False, False, False, False, '', 1, '', 0, '', True, False, False, 0, 0, 384, 384, False, False, True, True, True, 1) {} Traceback (most recent call last): File "C:\AUTOMATIC1111\modules\call_queue.py", line 45, in f res = list(func(*args, **kwargs)) File "C:\AUTOMATIC1111\modules\call_queue.py", line 28, in f res = func(*args, **kwargs) File "C:\AUTOMATIC1111\modules\txt2img.py", line 49, in txt2img processed = process_images(p) File "C:\AUTOMATIC1111\modules\processing.py", line 430, in process_images res = process_images_inner(p) File "C:\AUTOMATIC1111\modules\processing.py", line 482, in process_images_inner processed = Processed(p, [], p.seed, "") File "C:\AUTOMATIC1111\modules\processing.py", line 222, in init self.sd_model_hash = shared.sd_model.sd_model_hash AttributeError: 'NoneType' object has no attribute 'sd_model_hash'

[alexfredo]

Do someone else have the same problem that me ? the dreambooth style of 2.1 is still ignored, please someone helps

[mkoryak]

I had this problem and fixed it by making sure to include the yaml file renamed correctly to match checkpoint name. The wiki talks about doing this for 2.x models.

Another thing I noticed is that loading a non 2.x model broke everything and nothing would load giving similar looking errors. I had to build a new instance.