Image Multi-Switch

The Image Multi-Switch takes up to five IMAGE inputs and passes through the first one that isn't None. That's the entire logic, and it's the same "first live input wins" pattern as the pack's other multi-switches. For images specifically, this is the node you reach for when you have several candidate images - a load image branch, a VAE-decoded result, an upscale pass - and you want one of them to win based on which parts of the graph are active, without a menu or a manual picker.

Why slot order is the whole game

There's no Input widget here, so the node can't be switched by hand - selection happens entirely through what's connected and bypassed. That's both the appeal and the trap.

The appeal: if you're comparing the output of two img2img paths, you can wire both into the switch and toggle by bypassing one branch. The winner is decided upstream. It composes beautifully with group bypassers - flip a whole branch off with one click and the switch silently reroutes.

The trap is the author's rule from the README, which applies harder here than anywhere: a node that can't be bypassed is never None, so an always-live connection in slot 1 means slots 2–5 are dead weight. The switch stops at the first non-None and never looks further. If you want slot 3 to win, slots 1 and 2 must be genuinely empty (bypassed upstream). Plan the order carefully: things you'll turn off early, things that might win, always-live things last.

Also worth knowing: all inputs are forceInput (wire-only - no typing images in, obviously), and if every slot is empty the node raises a hard "Missing Input" error rather than passing nothing. Loud and debuggable, not silent and broken.

When you'd reach for it

• Path comparison. Two candidate images from different branches; bypass decides.
• Conditional composition. A "use this image only if that branch ran" pattern, built from bypass.
• Passing with a fallback chain. The same "rescue data from bypassed nodes" job the pack's passers exist for - a multi-switch in front of a stubborn consumer keeps the image flowing.

What's on the node

• input1–input5 - optional IMAGE inputs, wire-only.
• image output - one IMAGE, ready for a preview, a VAE encode, or the next stage.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/r-vage/ComfyUI-RvTools_v2
# restart ComfyUI

Or search "ComfyUI-RvTools_v2" in ComfyUI Manager. No models; deps are the pack's standard torch/numpy/Pillow/opencv-python/pilgram.

The pack caveat

The README banners the pack as unmaintained and points to ComfyUI_Eclipse as the successor. The RvTools name also has history: the original repo was deleted from GitHub in early 2025, breaking shared workflows and generating the usual "how do I install RvTools now?" thread on r/comfyui. v2 is the renamed, still-installable version - fine for legacy workflows, and Eclipse is where the author moved these tools for new builds.