Output2String

NodeGPT runs on its own custom TEXT type - the pack's way of passing LLM output around its own nodes. That's fine inside the pack, and useless almost everywhere else in ComfyUI, because most nodes speak STRING. Output2String is the translator: TEXT in, plain STRING out, and suddenly your agent's reply can feed into anything in the wider ecosystem - a prompt box, a text node, a filename, a metadata field.

That's the whole node, and it's genuinely the kind of glue a graph needs. When you're mixing NodeGPT agents with the rest of your workflow - which is the point of the pack, really, getting LLM output back into normal ComfyUI land - you'll hit the type mismatch almost immediately. This is the fix, and there's nothing clever about it: one input, one output, no settings.

• text - the pack's TEXT value, typically from a Chat or llava node.
• Output: STRING, for anything that wants a string.

Why a whole node for this?

It looks trivial, and it is - but in ComfyUI, type compatibility is how the graph communicates, and a custom type with no adapter strands your output at the edge of the pack. The author shipping this node is a sign they understood the ecosystem's rules: every custom pack eventually needs a "back to normal types" exit ramp. If you combine the LLM's output with the Conditioning node, you don't need this (that takes TEXT directly); you need it when the destination is a plain string input. Think of it as the adapter plug you reach for when the wire colors don't match.

The pack's standing applies as ever: unfinished, author-flagged as "not everything is functioning correctly yet," and now a fossil - xXAdonesXx/NodeGPT returns 404 as of this writing and the account is gone, so the README's install below is dead and ComfyUI Manager can't help. The surviving fork only has the pack's earliest build.

cd ComfyUI/custom_nodes
git clone https://github.com/xXAdonesXx/NodeGPT

Restart to auto-install requirements. If you're building a modern workflow with an LLM node, check whether it already outputs STRING - most maintained ones do - and you won't need a translator at all. If you're decoding a 2023 workflow that has one, now you know what the little box between the chat and your prompt box is doing.