Tensorboard Access

This is the smallest node in the pack, and it does exactly one thing: it launches TensorBoard and hands you the URL to open it. That's the entire feature. No inputs, no outputs, no widgets to configure - just drop it in your graph, hit Queue Prompt, and check your ComfyUI console for the link it prints there.

TensorBoard itself isn't specific to this pack - it's the standard training-log viewer that's been bolted onto more or less every diffusion trainer since kohya's scripts made it common practice (OneTrainer bundles it too, for the same reason). It shows you loss curves and other metrics over the course of a run, which is genuinely useful for LoRA training: a loss graph that's still dropping steadily says keep going, one that's flattened or bouncing around says you've probably hit diminishing returns, and it's a faster read than eyeballing sample images every few epochs.

How it works

The node has an empty schema by design - no required, no optional, no outputs. It's a trigger, not a data node. Clicking it once starts a TensorBoard server pointed at the logs folder your training run writes to (that folder lives alongside your launcher file, in the same place the Lora Training in ComfyUI / Advanced nodes write their run data), and the URL to view it gets printed to the command prompt running ComfyUI - not anywhere in the ComfyUI web UI itself, so you do need to go look at your terminal.

Per the README, you can queue this node before, during, or after a training run - it just opens the viewer, it doesn't care whether there's anything to show yet. Queue it before you start training if you want the tab open and ready; queue it mid-run to check progress without interrupting anything.

Using it

There's nothing to set. The only "workflow" is: add the node, queue it, copy the URL from your console into a browser. If you're actively training, refresh the TensorBoard page periodically to see the curve update - it doesn't push live, so you're pulling.

How to install it

Same pack as the training nodes, so if you've already installed either of those, you already have this one:

• ComfyUI Manager - search "Lora-Training-in-Comfy", install, restart.
• Manual:

  cd ComfyUI/custom_nodes
  git clone https://github.com/LarryJane491/Lora-Training-in-Comfy
  

  then install the requirements file for your platform (requirements_win.txt on Windows) and restart ComfyUI.

Common issues & troubleshooting

No URL printed, or the node errors immediately. This node depends on the same Python environment as the training nodes in this pack - if requirements didn't install cleanly (the usual culprit: installed outside an active venv, or a version conflict with another custom node's dependencies), this is often where it surfaces, even though the node itself has nothing to configure.

Nothing shows up in TensorBoard, or the graph is empty. Normal if you haven't actually run a training job yet, or if you queued this node against a fresh install with no logs folder created. Run the basic or Advanced training node first, then come back to this one - there has to be log data for TensorBoard to display.

Can't find the URL. It's written to the terminal/command prompt window ComfyUI is running in, not to the browser tab or the node itself. If you're running ComfyUI as a background service without a visible console, you'll need to check wherever that process's stdout is going.