LC Scoreboard π
Settle the A/B test with a number instead of your memory
- wins_1
- ties
- wins_2
- summary
What it is
A scoreboard. It keeps count while you compare two models, two LoRAs, two sampler settings side by side, and it puts the running total on the node so you stop doing arithmetic in your head at image 30.
Three buttons sit on it: #1 Win, Tie, #2 Win. One press scores both sides, because a win for one is a loss for the other. Reset all clears the board. Under the buttons you get the leader and the margin (#1 leads (60%)), a share bar, and each side's win-loss-tie record. Equal wins shows Even.
Inputs and outputs
The three counts are ordinary integer widgets - wins_1, ties, wins_2 - each 0 to 100,000. They're editable by hand, which is there so you can fix a misclick without resetting the whole board. The outputs mirror them (wins_1, ties, wins_2) plus summary, a string like #1 leads (60%): #1 12-4-4 (60%), #2 4-12-4 (20%), ties 20% of 20 - ready to pipe into a Show Text node or a logger if you care about that sort of thing.
The score saves with the workflow, and you do not need to run the graph. Pressing a button updates the counts; the outputs are just there if you want them.
Install
cd ComfyUI/custom_nodes
git clone https://github.com/lonecatone23/ComfyUI_LC_ModelBuilder_nodes
Restart, then hard-refresh your browser (Ctrl+Shift+R). That last step matters more than usual: the buttons are a JavaScript extension, and if your browser is holding a cached copy of the old JS you'll get a node that renders fine and does nothing when clicked. If the buttons don't respond, that's the cause nine times out of ten.
The honest take
This is not a metric, and it doesn't pretend to be. It doesn't score an image or know what "better" means - you're the judge, and the node just refuses to let you forget the score. Which is exactly the part people get wrong: you run twenty comparisons, like #2 more by the end, and conclude #2 won, when the first fifteen went the other way.
Pair it with a fixed seed and an actual side-by-side view. The pack ships a LoRA Test and a couple of Model test workflows that use it for exactly that, and the author's A/B material leans on LC Image Batch From Folder from the sibling ComfyUI_LC123_nodes pack, so if you load those workflows you'll want that installed too.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| wins_1 | INT | 00β100000 | #1's wins (= #2's losses). Use the buttons; type here to fix a mistake. |
| ties | INT | 00β100000 | Ties. Use the buttons; type here to fix a mistake. |
| wins_2 | INT | 00β100000 | #2's wins (= #1's losses). Use the buttons; type here to fix a mistake. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| wins_1 | INT | β |
| ties | INT | β |
| wins_2 | INT | β |
| summary | STRING | β |