ScuNET fails unless a txt2img or img2img inference is done before upscale

[the-dusky]

Auto1111 v1.7
Trying to run a serverless endpoint specifically for upscaling. Using the extra-single-image endpoint of the API.

When I try ScuNET I get the following error: {'output': {'detail': '', 'body': '', 'errors': "'Options' object has no attribute 'SCUNET_tile'"}, 'error': 'AttributeError'}

I can avoid this error, by first generating a txt2img output and then ScuNET works as expected. The workaround so far is to immediately generate a 1 step txt2img output to sort of "prime the pump". But this has issues within a serverless environment where containers are being created and destroyed all the time.

I am wondering if there is another way to make this work.

[Roadsfj]

--nowebui启动时有个回调没有执行，tile未初始化，没有赋值。
我这边测试下来这个变量的值没有变过，时间原因没有进一步梳理代码逻辑。
你可以测试下，当然如果你有新的发现欢迎讨论。
另外我这边把webui中extras里面的ScuNET超分独立出来了，有需要可以留言。

sd.webui\webui\extensions-builtin\ScuNET\scripts\scunet_model.py

def tiled_inference(img, model):

    # tile = opts.SCUNET_tile
    tile = 256

    # tile_overlap = opts.SCUNET_tile_overlap
    tile_overlap = 8

def do_upscale(self, img: PIL.Image.Image, selected_file):

    # tile = opts.SCUNET_tile
    tile = 256