Help, trying to create extension for Latent Modifier, but parameters become NaN when imported into Comfy Workflow tab

[Dwedit]

I'm trying to create a Swarm UI extension for the Latent Diffusion Mega Modifier node.

The node itself has about 22 parameters, but I'm only exposing one of the parameters in the UI, which would be the "contrast_multiplier" parameter.

When I add the extension into the chain, it seems to work okay, I'm seeing images getting generated, and the Contrast Multiplier setting seems to apply okay. However, when I go to the Comfy Workflow tab, then click "Import from Generate Tab", I see that many of the arguments for the Latent Diffusion Mega Modifier node suddenly have values of NaN.

In particular, any parameter where I'm assigning 0 instead appears to have a NaN as its value, this affects 9 different parameters. The full list is: sharpness_multiplier, tonemap_multiplier, tonemap_percentile, combat_cfg_drift, rescale_cfg_phi, extra_noise_multiplier, divisive_norm_multiplier, spectral_mod_percentile, spectral_mod_multiplier. Other numbers like 100, 127, or 5 are unaffected, and are not turning into NaN.

I can't tell if this is an issue with my code, or with Swarm's import feature itself.

You need to install the Comfy node for "Latent Diffusion Mega Modifier" to use the extension.

The code for the extension is tiny, so I'll paste it here:

File is named "LatentModifierContrastMultiplierExtension.cs"

using SwarmUI.Core;
using SwarmUI.Utils;
using SwarmUI.Text2Image;
using SwarmUI.Builtin_ComfyUIBackend;
using Newtonsoft.Json.Linq;

// NOTE: Namespace must NOT contain "SwarmUI" (this is reserved for built-ins)
namespace MonkeysDocs.CoolExtensions.LatentModifierContrastMultiplierExtension;

// NOTE: Classname must match filename
public class LatentModifierContrastMultiplierExtension : Extension // extend the "Extension" class in Swarm Core
{
    // Generally define parameters as "public static" to make them easy to access in other code, actual registration is done in OnInit
    public static T2IRegisteredParam<int> MyParameter;

    public static T2IParamGroup MyParameterGroup;

    // OnInit is called when the extension is loaded, and is the general place to register most things
    public override void OnInit()
    {
        Logs.Init("Latent Modifier Contrast Extension initializing"); // Use "Logs" for any/all logging.
        MyParameterGroup = new("Latent Modifier", Toggles: true, Open: true, IsAdvanced: false);
        // Note that parameter name in code and registration should generally match (for simple clarity).
        MyParameter = T2IParamTypes.Register<int>(new("Contrast Multiplier", "Latent Modifier Contrast Multiplier - When negative, reduces frying effects of high CFG scale.",
            "0", Min: -100, Max:100, ViewType: ParamViewType.SLIDER, Toggleable: false, Group: MyParameterGroup, FeatureFlag: "comfyui" // "comfyui" feature flag for parameters that require ComfyUI
            // Check your IDE's completions here, there's tons of additional options. Look inside the T2IParamTypes to see how other params are registered.
            ));
        // AddStep for custom Comfy steps. Can also use AddModelGenStep for custom model configuration steps.
        WorkflowGenerator.AddModelGenStep(g =>
        {
            //Logs.Info("WorkflowGenerator.AddModelGenStep");
            // Generally always check that your parameter exists before doing anything (so you don't infect unrelated generations unless the user wants your feature running)
            if (g.UserInput.TryGet(MyParameter, out int myParamNumber))
            {
                //Logs.Info("myParamemterNumber = " + myParamNumber.ToString());
                // Create the node we want...
                string shiftNode = g.CreateNode("Latent Diffusion Mega Modifier", new JObject()
                {
                    // And configure all the inputs to that node...
                    ["model"] = g.LoadingModel,
                    ["contrast_multiplier"] = myParamNumber,
                    //default options repeated below
                    ["sharpness_multiplier"] = "0.0",
                    ["sharpness_method"] = "anisotropic",
                    ["tonemap_multiplier"] = "0.00",
                    ["tonemap_method"] = "reinhard",
                    ["tonemap_percentile"] = "100.000",
                    ["combat_method"] = "subtract",
                    ["combat_cfg_drift"] = "0.00",
                    ["rescale_cfg_phi"] = "0.00",
                    ["extra_noise_type"] = "gaussian",
                    ["extra_noise_method"] = "add",
                    ["extra_noise_multiplier"] = "0.0",
                    ["extra_noise_lowpass"] = "100",
                    ["divisive_norm_size"] = "127",
                    ["divisive_norm_multiplier"] = "0.00",
                    ["spectral_mod_mode"] = "hard_clamp",
                    ["spectral_mod_percentile"] = "5.00",
                    ["spectral_mod_multiplier"] = "0.00",
                    ["affect_uncond"] = "None",
                    ["dyn_cfg_augmentation"] = "None",
                    ["seed"] = 0,
                    ["control_after_generate"] = "randomize"
                });
                // And then make sure its result actually gets used. The final save image uses 'FinalImageOut' to identify what to save, so just update that.
                g.LoadingModel = [shiftNode, 0]; // (Note the 0 is because some nodes have multiple outputs, so 0 means use the first output)
            }
            // The priority value determines where in the workflow this will process.
            // You can technically just use a late priority and then just modify the workflow at will, but it's best to run at the appropriate time.
            // Check the source of WorkflowGenerator to see what priorities are what.
            // In this case, the final save image step is at priority of "10", so we run at "9", ie just before that.
            // (You can use eg 9.5 or 9.999 if you think something else is running at 9 and you need to be after it).
        }, 9);
    }
}

[mcmonkey4eva]

"0.0" is a string (plaintext), 0.0 is a zero-double, 0 is a zero.
So you want eg ["rescale_cfg_phi"] = 0,

"NaN" means "Not a Number", and indeed you are feeding it not-a-number! You're feeding it a string that happens to look like a number.

ps please be sure to change the placeholder names to real ones before publishing your extension

[Dwedit]

Is there a way for an extension to install a Comfy node? It would be annoying if someone had to go to the Comfy Workflow tab and install the node from there before using the extension. But it would also make sense if this was not allowed for security reasons.

[mcmonkey4eva]

Look at the making extensions doc https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Making%20Extensions.md and referenced dynamic thresholding builtin extension, which has an installable node