Add Parameters🧸

Every LLM API has a pile of knobs the node UI doesn't expose. top_p, frequency_penalty, response_format, stop sequences, and so on. Add Parameters is the escape hatch that lets you set those anyway and feed them into the Advanced Prompt Enhancer node. If you've hit the ceiling of what APE's built-in fields give you, this is how you go deeper without waiting for the author to add a widget.

It's a power-user node. Most people never need it. But when you need response_format to force JSON output, or top_p to tune sampling, this is the only way to get there in Plush.

How it works

You type parameters into a text box as name/value pairs, and the node packages them into a LIST that plugs into APE's Add_Parameter input. At inference time APE merges those into the request it sends to the model. The clever bit: these nodes daisy-chain. You can string several Add Parameters nodes together to stack multiple settings in one run, each feeding the next.

Because these are raw API parameters, whether they actually do anything depends entirely on the service and model you're using. OpenAI, Anthropic, and the various local front-ends all accept different sets. Add Parameters doesn't validate them; it just passes them along, so check the docs for whatever you're talking to.

The inputs

• Parameters (multiline text): where you write the parameter name and value. This is the whole point of the node. Follow the format the pack's example workflow shows.
• Save_to_file (boolean) and File_name: optionally save a parameter setup to disk so you can reload it later with the Load Saved Files node. Handy if you've dialed in a good config for a specific model and want to reuse it instead of retyping.

The output Add_Parameter is a LIST, and it goes into the matching input on Advanced Prompt Enhancer (or another Add Parameters node to chain them). There's also Help and Troubleshooting text outputs.

A concrete use

The common combo: use Add Parameters to set response_format to JSON, force your model to reply in structured JSON, then pull fields back out with the Extract JSON data (ParseJSON) node. That turns a chatty LLM into a reliable data source you can branch on inside a workflow. The pack ships an example workflow (How_to_use_additionalParameters.png) that walks through exactly this.

Installing it

It comes with Plush-for-ComfyUI. Install the pack via ComfyUI Manager (search "Plush", install Plush-for-ComfyUI, restart), or clone it: cd ComfyUI/custom_nodes && git clone https://github.com/glibsonoran/Plush-for-ComfyUI.git, then pip install -r requirements.txt and restart.

Common issues

The most common disappointment is a parameter that seems to do nothing. That's almost always because the model or service you're using doesn't support it, not because the node failed. OpenAI's parameter list is not Anthropic's, and local front-ends often ignore anything they don't recognize. Check the API reference for your target model before assuming the node is broken. If you're saving parameter files, remember they land in Plush's own directory and are loaded back through the Load Saved Files node, not through a normal file picker. As always with Plush nodes, the Troubleshooting output is where a real error message will show up.