🐯 YFG 20 Image Switcher

Twenty image inputs on a single node. That's either a power move or a workflow you'll never actually finish wiring - and this is the honest review: it's mostly the latter, with one genuinely good trick up its sleeve.

The YFG 20 Image Switcher is the biggest member of the switcher family in the gonzalu/ComfyUI_YFG_Comical pack. Same design as its 3/5/10/15 siblings: selected_image dropdown, tolerant optional inputs, and two IMAGE outputs - the Selected Image tensor and a Connection Diagram graphic. The whole family shares one implementation, so the behavior is identical across sizes, just with more slots.

What it's actually good at

Two things. First, preview_type set to Connection Diagram draws a routing graphic where a line from the active input runs down to the output - at twenty rows, that's the only sane way to tell what's selected. Second, the graceful missing-input handling means you can use it as an oversized-but-friendly router in a workflow other people load: wire up whatever subset you have, and unconnected slots don't throw errors.

Honestly, though, if you need to switch among twenty images, the better move is usually a directory-based picker with an index (the same pack's Random Image From Directory in by_index mode) or an integer-driven switch. A dropdown with twenty entries is a lot of scrolling to change one thing. Use the 20-slot node when you genuinely have twenty sources that must all stay wired in the graph at once - for a big LoRA comparison grid, say.

Inputs, outputs, install

selected_image is the only input that changes behavior; the font/label settings are cosmetic, preview toggles the node preview. Outputs: Selected Image and Connection Diagram, both IMAGE.

Install is the pack install, once for the whole family: ComfyUI Manager → search "YFG Comical", or

cd ComfyUI/custom_nodes
git clone https://github.com/gonzalu/ComfyUI_YFG_Comical

Restart ComfyUI. Switchers are PIL-only; the pack-wide gotcha is the undeclared cv2/matplotlib in the histogram and mask files (pip install opencv-python matplotlib if the whole pack fails to import).

Bottom line

The diagram view is the family's selling point, and this is the size where it's most needed. But twenty wired inputs is a workflow smell - if you're actually running twenty comparisons, you'd probably rather be using a batch/plot node. Buy this one for completeness; reach for the smaller sizes in practice.