sd-webui-controlnet
sd-webui-controlnet copied to clipboard
Mikubill
[Bug]: Errors when generating with Style Adapter (and color Adapter)
Is there an existing issue for this?
- [X] I have searched the existing issues and checked the recent builds/commits of both this extension and the webui
What happened?
Trying to do a style transfer with Model checkpoint SD 1.5 They are both loading about 50% and then these two errors :/ Any help would be great as I would really like to try these style transfers
ControlNet 0: Preprocessor: Canny -- Model: control canny ControlNet 1: Preprocessor: Clip_vision -- Model: T2iadapter_style_sd14V1
RuntimeError: !(has_different_input_dtypes && !config.promote_inputs_to_common_dtype_ && (has_undefined_outputs || config.enforce_safe_casting_to_output_ || config.cast_common_dtype_to_outputs_)) INTERNAL ASSERT FAILED at "/Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/TensorIterator.cpp":407, please report a bug to PyTorch.
Error with Color Adapter
Controlnet 0: Preprocessor: Canny -- Model: control canny ControlNet 1: Preprocessor: Clip_vision -- Model: T2iadapter_color_sd14V1
RuntimeError: pixel_unshuffle expects height to be divisible by downscale_factor, but input.size(-2)=257 is not divisible by 8
Thank you!
Steps to reproduce the problem
ControlNet 0: Preprocessor: Canny -- Model: control canny ControlNet 1: Preprocessor: Clip_vision -- Model: T2iadapter_style_sd14V1
Start generating and then Error:
RuntimeError: !(has_different_input_dtypes && !config.promote_inputs_to_common_dtype_ && (has_undefined_outputs || config.enforce_safe_casting_to_output_ || config.cast_common_dtype_to_outputs_)) INTERNAL ASSERT FAILED at "/Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/TensorIterator.cpp":407, please report a bug to PyTorch.
What should have happened?
Applied style transfer into Canny Image
Commit where the problem happens
webui: commit: [0cc0ee1b] controlnet: fed7a529
What browsers do you use to access the UI ?
Brave
Command Line Arguments
COMMANDLINE_ARGS="--skip-torch-cuda-test --upcast-sampling --no-half-vae --use-cpu interrogate --api --disable-safe-unpickle"
(On a Mac)
Console logs
RuntimeError: !(has_different_input_dtypes && !config.promote_inputs_to_common_dtype_ && (has_undefined_outputs || config.enforce_safe_casting_to_output_ || config.cast_common_dtype_to_outputs_)) INTERNAL ASSERT FAILED at "/Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/TensorIterator.cpp":407, please report a bug to PyTorch.
Additional information
No response
unfortunately not related - Ive tried with Safari and it gives the same error for both Style Adapters
Also on Mac getting errors when trying to use the new color and style adapters. I wonder if this is because Apple silicon doesn't support xformers? Or perhaps it's related to pytorch?
Also on Mac getting errors when trying to use the new color and style adapters. I wonder if this is because Apple silicon doesn't support xformers? Or perhaps it's related to pytorch?
I dont know but I would really like to try these new styles - Ive tried to update to Pytorch 13 and it's all up to date on my machine but for some reasons SD is still telling me that I have the 12.1 version - Maybe related ?
let's just hope it gets fixed and we can use it soon
Also on Mac getting errors when trying to use the new color and style adapters. I wonder if this is because Apple silicon doesn't support xformers? Or perhaps it's related to pytorch?
I dont know but I would really like to try these new styles - Ive tried to update to Pytorch 13 and it's all up to date on my machine but for some reasons SD is still telling me that I have the 12.1 version - Maybe related ?
let's just hope it gets fixed and we can use it soon
I also cannot get SD to show 13, it always says I have 12.1
macos support is limited to controlnet.
cc: @brkirch (but he may be inactive)
you could make a feature request for cpu only option for clip_vision preprocessor
Same error here. Seems like this is quite common, possibly all M-series processors.
Same error here. Seems like this is quite common, possibly all M-series processors.
Not only M-series, I think here it is related to all Mac - Still having issues with Style & color adapters, it is just not working. And ive tried on multiple browsers, with multiple python versions and arguments...
I guess they will work on it, but it seems they don't have Mac available for testing... Hopefully it will be fixed soon as I would really like to use these with my controlnet settings.
Has anyone tried to use CPU instead of GPU as a param and see if it's working?
