Griptape Display: Text

Every Griptape workflow ends the same way: with an agent that has produced a string, and you staring at a wire with nowhere to put it. That's the job of Griptape Display: Text. It's the text equivalent of Save Image - a terminal node that makes a string visible in the UI so you can actually read what your LLM came back with.

It's a deliberately boring node, and that's fine. Boring is what you want here.

How it works

The node takes a string in and sends the same string back out. Under the hood it's an OUTPUT_NODE: instead of producing something for the rest of the graph to consume, it packages the value into a UI payload that ComfyUI renders on screen. The pass-through output means you can drop it in the middle of a chain, inspect the text, and keep the flow going to something else - you're not forced to make it the dead end.

Two inputs, and they matter more than you'd think:

• INPUT - connect any string-producing node here. If it's connected, it wins.
• STRING - a manual fallback, for when you want to hardcode a label or a value to look at without building a node for it.

The single output is OUTPUT, typed STRING, carrying whatever text came in. Wire that into a Merge Text, a prompt builder, or the next Griptape step.

Where you reach for it

Anywhere you want to see intermediate text without saving anything. A few places people actually use it:

• At the end of a Prompt Task, so the agent's answer shows up on the canvas.
• After an Image Description node, to read what the model thinks it sees.
• Before a Merge Text, to sanity-check what each branch produced.

Installing

It ships in the ComfyUI Griptape Nodes pack, so install the pack, not the node:

• ComfyUI Manager: Manager → Install Custom Nodes → search "Griptape" → install ComfyUI-Griptape.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/griptape-ai/ComfyUI-Griptape, then restart ComfyUI.

The pack pulls in griptape[all] (plus openai, python-dotenv, and a couple of git-hosted driver extensions). Manager handles that automatically; if you install manually, let pip resolve requirements.txt. The one real install pain point is torch - Griptape declares it as a dependency and can nudge ComfyUI onto the wrong build. If sampling starts erroring after install, pip uninstall torch and reinstall the CUDA build (see the README's troubleshooting section) - the classic fix is pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121.

Gotchas

The INPUT-vs-STRING precedence trips people up: if you have a wire connected but the node shows blank, check whether the wire is actually feeding INPUT, because a connected INPUT silently overrides the manual STRING field. Also note this node needs no API key and calls nothing - the "Display" in the name is literal. If you expected an LLM node, you've got the wrong tab.

One thing the changelog confirms: the display used to get wiped between runs, which made it useless for comparing outputs. It's persistent now, so text you see stays until you change it.