Unload Controlnet model from buffer?

[DragonHawkAlpha]

This code disables controlnet once used. But I don't see a controlling API endpoint to disable. Anyone know how to disable once used via txt2img API in Automatic1111 without server restart?

Code from Wiki does not have an endpoint in the API to run it. I do not understand how this would be called either, where do you find StableDiffusionProcessing object? Can someone add a disable endpoint? Leaving out of txt2img request is not enough because it does not unload from buffer as far as I can workout.

import importlib
external_code = importlib.import_module('extensions.sd-webui-controlnet.scripts.external_code', 'external_code')

def disable_controlnet(p: StableDiffusionProcessing):
external_code.update_cn_script_in_processing(p, [])

[ljleb]

This is code that should be put in a webui extension.

There is no endpoint to disable controlnet at the moment, but as long as you don't ask for controlnet to run through the API, it shouldn't do any processing. Just remove "alwayson_scripts": {"controlnet": {"args": [...]}} from the request and it should not use controlnet.

Does your use case require you to permanently disable controlnet for some reason?

[SethRobinson]

Yes, seeing this behavior too with my AI Tools front end for auto1111.

After using ControlNet once (via api), can no longer do a normal Text to Image generation at all (via api) due to it somehow still trying to use ControlNet despite not sending any controlnet parms via "alwayson_scripts", and gives an error that breaks all future renders. (unless those renders are setup to use controlnet, of course)

File "/home/glados/aitools_server/extensions/sd-webui-controlnet/scripts/batch_hijack.py", line 42, in processing_process_images_hijack
      return getattr(processing, '__controlnet_original_process_images_inner')(p, *args, **kwargs)
    File "/home/glados/aitools_server/modules/processing.py", line 743, in process_images_inner
      devices.test_for_nans(x, "vae")
    File "/home/glados/aitools_server/modules/devices.py", line 158, in test_for_nans
      raise NansException(message)
  modules.devices.NansException: A tensor with all NaNs was produced in VAE. This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

If the auto1111 server is reloaded or a new SD model is loaded (and not internally cached probably), it fixed it.

It seems to me there is probably a work around if we send controlnet API info and instruct it to turn off all controlnet units but I'm not sure how to do that, sending "none" or blank strings didn't work for me.

[ljleb]

I see, this is pretty odd. Passing the appropriate number of units with "enabled": false in each unit should be a workaround for now.

[DragonHawkAlpha]

I don't really understand the "enabled": false and where that would need to go?

[SethRobinson]

After updating to latest versions of auto1111 and ControlNet today, issue seems fixed completely, thanks!

[DragonHawkAlpha]

@ljleb I came back to this issue and your suggestion worked great! Thanks.