FEAnyToString

Somewhere between "the graph works" and "I trust what's in it" there's a gap, and FEAnyToString is a two-second bridge. Give it any value and it hands you back a string. Dictionaries and lists get JSON-serialized (with ensure_ascii=False, so non-Latin text comes out readable rather than as \uXXXX escapes); everything else gets a plain str().

Its real job is debugging, though the node doesn't advertise it. ComfyUI hides the values flowing through *-typed wires unless you click around a node's display, and when something isn't behaving - a dict that came out empty, a list with the wrong length, a string with stray whitespace - this node is the fastest way to see it. Drop it on the wire, route its string output into a display text widget or a save node, run, and read.

The inputs and outputs

• input - any type. No other inputs.
• output - STRING. The JSON or stringified form of whatever came in.

It's also marked as an output node (OUTPUT_NODE), which in practice means it'll execute even if nothing consumes its output - convenient for a debug probe, slightly annoying if you wire it in for a one-off check and forget to delete it.

Where it's genuinely useful beyond debugging: feeding arbitrary data into text-based systems. Want to hand a dict's contents to a text encoder for logging, a filename builder, or an LLM prompt-gen node? Run it through FEAnyToString first and it becomes a plain string. Keep in mind JSON-serialized output can be long and ugly if your value is a big list of tensors - str() of a tensor includes the shape and device, which is usually what you actually wanted to know anyway.

The honest verdict: it's a one-function utility, about as thin as this pack gets, and you won't use it every day. But the day a mysterious * value refuses to cooperate, you'll be glad it's one node drag away. Nothing to install extra, nothing to configure - it's just there with the rest of the pack.