Nodes/Allma Nodes/Allma Stop
ComfyUI Node

Allma Stop

A stop button that reaches inside subgraphs

By OliveiraNickolas·Created 24 days ago·Updated 22 days ago· 0
Allma Stop

      Allma Generate has its own stop button - right there on the node - which is perfect until the node lives inside a subgraph. Then "stop the generation" means opening the subgraph mid-run, navigating to the node, and clicking, by which point the thing you wanted to stop has finished or you've lost your place. Allma Stop is a standalone stop button you can put anywhere, subgraphs included, and it cancels whatever Allma generation is in flight.

      Why it works from anywhere

      The interrupt is global rather than per-node. The button talks to the backend's interrupt endpoint from the browser, so a button sitting at the root of your workflow cancels a Generate buried three subgraphs deep exactly as well as the one on the node itself. That's the whole point: a stop button you can reach quickly, mid-run, without hunting.

      The node has no inputs and no outputs. It declares itself an output node so ComfyUI keeps it in the graph instead of pruning a node that feeds nothing - but nothing about it participates in execution. The button is pure frontend.

      One small nicety: the label reports what actually happened rather than pretending. ✅ stopped when something was genuinely streaming and got cancelled; · nothing running when you clicked it with no generation in flight. No fake success.

      When you'll actually want it

      Any time you have a Generate that's not on your main canvas. A long enhancement run on a video prompt, an image batch feeding the model, a workflow where the Generate is collateral in someone else's shared subgraph - all places where the built-in button is effectively out of reach. Drop an Allma Stop somewhere you can see it and keep working.

      Install

      cd ComfyUI/custom_nodes
      git clone https://github.com/OliveiraNickolas/AllmaNodes
      

      Restart, no pip install. It's in the Allma/llm group, so it assumes you have the LLM half running to be useful - but the node itself just talks to the backend's interrupt endpoint, so it needs no model of its own.

      Fair warning for a pack this new (no community footprint at all yet): the interrupt is global, which means the button cancels whatever Allma generation is in flight, not a specific one. In a workflow that runs Generate nodes in parallel, one click stops them all. That's almost certainly the intended behavior - a stop button that only stopped one of several streams would be useless - but it's the thing to know before you wire this into a multi-branch graph and wonder why everything died.

      CategoryAllma/llm

      Inputs (0)

      No inputs

      Outputs (0)

      No outputs