Send Text

Most nodes in a ComfyUI graph pass data forward to the next node. Send Text goes the other way - it pushes a value back out of the workflow, to the tool that launched it. In Acly's pack this is the return channel for the Krita AI Diffusion plugin: when the generation finishes and the workflow has computed something the plugin's UI wants to display - a piece of text, a value, a status - this node is how it gets there. It's an output node, meaning it's a terminal endpoint of the graph, not a step in the middle.

The reason this exists is the same reason the whole "krita" category exists. Krita AI Diffusion, built by the same author, runs Stable Diffusion inside Krita by driving a ComfyUI backend with API-format workflows. Data flows into that backend through nodes like Krita Image Layer, and it flows back out through this one. Without a return path, the plugin could only ever get images back; Send Text lets the workflow hand structured values back to the application so the UI can react - surface a translated prompt, report a computed number, whatever the tool arranged to receive.

The mechanism is a WebSocket message. When the graph runs, this node serializes its value and emits it over the connection back to the client that submitted the prompt, tagged with the name you gave it so the receiving tool knows which output is which.

Its inputs, from the schema:

• value (*) - a wildcard input, so you can pipe in essentially anything the workflow produced. In practice that's usually a string, but the * type means it isn't fussy.
• name (STRING, default "Output") - the label this value travels under, so the receiving tool can tell one Send Text from another when a workflow has several.
• type (COMBO, default "text") - declares how the value should be interpreted on the way out, so the client handles it appropriately.

There are no outputs - it's an output node, the end of a branch. It does its work as a side effect (sending the message) rather than by producing something for a downstream node.

Installation is the pack standard: "ComfyUI Nodes for External Tooling" in ComfyUI Manager, or cd custom_nodes && git clone https://github.com/Acly/comfyui-tooling-nodes.git from your ComfyUI directory, then restart. No dependencies, no models. As with the other krita nodes, the thing that actually uses this is the Krita AI Diffusion plugin, which installs and manages the pack itself - so most Krita users get it without a manual install.

The predictable trap: dropping Send Text into a browser ComfyUI graph and looking for its effect on screen. There isn't one you'll see in the normal editor - the message goes to whatever client opened the WebSocket connection, and if that client isn't a tool listening for it, the value effectively goes nowhere visible. That's not a malfunction; it's a node designed to talk to an external listener. This is squarely a build-your-own-tooling node - genuinely useful if you're writing software that submits prompts to ComfyUI and needs computed values back over the socket, and mostly invisible if you're not.

If you got here from the plugin side and something isn't showing up in Krita, the usual suspects are a name that doesn't match what the plugin expects, or a type that doesn't line up with how the client parses the message. Both are contract details between the workflow and the tool, so they're worth checking against whatever the plugin (or your own code) is set to receive.