Get Parameter From Table

Same idea as Get Parameter From List - pull a value out of <params:name:value> tags in your prompt - but built for when you're pulling a lot of them. Instead of every reader node re-scanning your whole prompt text from scratch, you parse the text once into a table and then each lookup just reads from that table. The node's own description says it plainly: the table has a higher upfront cost, but with many parameters it can be "much, much faster." If you've got two or three parameters, don't bother. If you've got fifteen, this is the one you want.

Think of it like the difference between searching a document every time you need a fact versus reading it once into an index. The searching-every-time approach is fine until it isn't, and a big parameter-driven prompt is exactly where it isn't.

How it works

There are two halves. A companion node (MakeParameterTable, referenced in this node's description) does the parsing work once - it reads your prompt and turns all the <params:...> tags into a JNODES_PARAMETER_TABLE object. Then you drop as many Get Parameter From Table nodes as you have values to read, feed each one that same table, and name the parameter you want. The heavy lifting happened once; the reads are cheap.

Everything else behaves like the list version. Values come back as their natural type, the wildcard output plugs into anything, and the value still has to be valid for its destination.

The inputs and outputs that matter

• parameter_table - the JNODES_PARAMETER_TABLE from the MakeParameterTable node. This is the input that replaces "scan the raw text every time."
• parameter_name - which value to pull.
• return_type - auto for natural-type conversion, string to force text when a downstream node expects it.
• add_to_png_info - on by default; writes the name/value into image metadata. Turn off to keep your PNG info clean.

Optional parameter_default (any type) is the fallback when the name isn't in the table. The output is a single wildcard pin with the resolved value - wire it wherever the value belongs.

How to install it

ComfyUI Manager, search JNodes, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/JaredTherriault/ComfyUI-JNodes
pip install -r ComfyUI-JNodes/requirements.txt

No model downloads.

Common issues & troubleshooting

"Where's the table coming from?" This node doesn't parse text itself - it needs the table object from MakeParameterTable feeding its parameter_table input. If you plugged your raw prompt in expecting it to work like the list node, that's the mismatch. Build the table first.

No speedup. With only a handful of parameters the table's upfront cost eats the win. This paradigm pays off at scale; below roughly a dozen lookups, Get Parameter From List is simpler and just as quick.

Tags leaking into the image. As with the list node, reading a parameter doesn't remove the <params:...> text from your prompt. Strip it with the pack's RemoveParseableDataForInference node before the text reaches your encoder.

Default keeps firing. The parameter_name doesn't match a tag that actually made it into the table. Check spelling and confirm the tag was present in the text you fed MakeParameterTable.