A few questions about using the API

[dayunbao]

I'm almost done switching over from Stability AI's API to Automatic1111/SD WebuiControlNet's API, and I have a few questions.

1. Do we need to pass enabled every time we call one of the /sdapi/v1/ endpoints if we plan on using a ControlNet? This example has it, but other examples like this one don't.
2. I've enabled multi ControlNet in the settings, but am not seeing how to add more ControlNets in the client. I don't have any extra tabs appearing, and I don't see any way to add them. I'm guessing that the API equivalent of multi ControlNets is by having multiple items in the args list? Like this:

    "alwayson_scripts": {
        "controlnet": {
          "args": [
            {
              control_net_1
            },
            {
              control_net_2
            },
           {
              control_net_3
            },
          ]
        }
      }

And the number of items in args can't exceed the value of Multi ControlNet: Max models amount?
3. I'm still a little unsure about the pre-processor resolution and thresholds. It seems like the thresholds mean different things to different models, when they are used at all. Should I just leave them at their defaults? In what situation would I need to send something other than the default values?
4. I noticed that enabling pixel perfect in the client automatically sets the pre-processor resolution, but the /controlnet/detect endpoint doesn't accept pixel_perfect. Is there any plan to add that? I was thinking I could just always send pixel_perfect=True and not worry about setting the resolution if calling /controlnet/detect. However, since the endpoint doesn't accept that parameter, I guess I have the same question as I do for thresholds. When to change the default value?

[dayunbao]

Not sure if I did something wrong, but I feel like something isn't working correctly. Using the API from the app I'm building, I was able to successfully use a ControlNet pre-processor directly (openpose), and then used the returned image as input for a text-to-image generation. However, the returned images didn't seem related to the OpenPose input at all. See below.

Openpose:

Sample response images:

Request body:
`{

"prompt": "Samurai cat standing in a post-apocalyptic wasteland.",
"negative_prompt": "",
"cfg_scale": 7,
"sampler_name": "Euler",
"batch_size": 6,
"seed": 0,
"steps": 100,
"width": 512,
"height": 512,
"alwayson_scripts": {
    "controlnet": {
        "args": [
            {
                "input_image": very_long_string,
                "enabled": true,
                "lowvram": false,
                "pixel_perfect": true,
                "module": "openpose",
                "model": "control_v11p_sd15_openpose [cab727d4]",
                "weight": 1,
                "guidance_start": 0,
                "guidance_end": 1,
                "control_mode": 0,
                "resize_mode": 1
            }
        ]
    }
}

}`

Did I do something wrong?

[huchenlei]

I think the script name ControlNet is case sensitive. See here:
https://github.com/Mikubill/sd-webui-controlnet/blob/efda6ddfd82ebafc6e1150fbb7e1f27163482a82/scripts/controlnet.py#L220C1-L221C28

[dayunbao]

OK. I'll try that. All of the example requests in the repo use controlnet, though.

Thanks!

[huchenlei]

Sorry about that, I was just suspecting. Another thing I noticed is that if you are passing an openpose skeleton to ControlNet, you should select preprocessor none, intead of openpose.

[dayunbao]

I'll give that a shot. Thanks again!

[taii241]

Sorry, I would like to ask you how to implement the change of predictor to none. Do you mean changing model: "openpose" to model: "none"?

[dayunbao]

Sorry about that, I was just suspecting. Another thing I noticed is that if you are passing an openpose skeleton to ControlNet, you should select preprocessor none, intead of openpose.

That was the problem. Thanks!