Slack Thread Start

Drop a batch of send nodes into one workflow and Slack stops being a delivery channel and starts being a mess: three messages scattered across the channel root, the image here, the caption ten minutes later. Slack Thread Start is the pack's tidy-up node. It posts a root message - by default the admirably on-the-nose "Your ComfyUI generations are here!" - and hands you back a thread_ts, which you wire into the thread_ts input of any Send Image/Video/Text/Audio node. Everything then lands as replies in that same thread, grouped the way you'd actually want to read it.

It's the one node in this pack that isn't a dead end: while every send node has no outputs, this one's whole job is to produce a STRING - the thread timestamp - that fans out to however many send nodes you have.

How it works

Mechanically it's a small, well-made piece of plumbing. The node posts the header message via the Slack API and outputs the returned thread timestamp. Wire that into a send node's thread_ts and two things happen: the send node replies in that thread instead of the channel root, and the dependency forces it to run after the thread exists - so the root is always there before the first upload lands.

Two details are worth knowing. First, the mode widget controls caching through ComfyUI's IS_CHANGED hook: in New thread each run mode the node returns NaN (which is never equal to itself), so ComfyUI re-executes it on every queue and posts a fresh root each time. In Reuse existing thread mode it returns a stable key, so re-running the workflow is cached and appends to the same thread. The reuse memory is per ComfyUI session - restart, or change the channel/header, and you start a new thread. Second, the thread carries its channel with it: once you connect thread_ts, the send node's own channel widget is disabled and the thread's channel wins, so you set the destination once, here.

The inputs that matter

• channel - where the thread lives, using the pack's usual syntax: #general, @alice, or a raw ID.
• mode - New thread each run (fresh thread every execution) or Reuse existing thread (keep appending within the session).
• header - the root message that opens the thread; Markdown is translated to Slack formatting like the text node. Leave it and you get the default string.

Install

Same pack, same steps as every other node here: ComfyUI Manager (search "ComfyUI-Slack"), or

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

Light dependencies, no models. And the same Slack-app setup: create the app from the README's manifest, install it, set SLACK_BOT_TOKEN in the environment that launches ComfyUI, and /invite @ComfyUI to the channel.

Troubleshooting

The failure modes are the pack's shared ones - missing token, not_in_channel (you skipped the invite), private channels not resolving until the bot is a member. The one this node adds is confusion about state: if you expect a new thread on every run and instead see results piling into one, you've got Reuse existing thread selected. And if you connect thread_ts but the messages still hit the channel root, check the wire is actually connected - an unconnected thread_ts posts to the channel root by design, which is also the pack's escape hatch when you want a mix of threaded and unthreaded output.