CLIP Multi-Switch

The CLIP Multi-Switch is the laziest kind of smart: it takes up to five CLIP inputs, checks them in order, and hands back the first one that isn't None. No menu, no widget, no logic to configure. If slot 1 is empty it tries slot 2, and so on down the line. That one behavior turns out to be exactly what you want when a workflow has several possible text encoders and you can't know in advance which one will be live.

Why you'd actually use it

Here's the scenario the pack author built this for, straight from the README: you have a Flux workflow that can use different CLIP setups - a dual CLIP loader (Flux's T5 + CLIP-L), a triple CLIP loader (SD3), or just the CLIP that came out of a checkpoint loader. Wire all of them into the multi-switch and bypass the ones you're not using. Whatever's left live flows through. It's an "options menu" for model choices, built entirely out of ComfyUI's own bypass mechanics.

The trick that makes it work is where you put things. A bypassed node upstream produces None, which the multi-switch skips. But a node that can't be bypassed - a checkpoint loader, which is always enabled - never produces None. So the author's rule, verbatim in spirit: connect the bypassable clip loaders to the first slots, and put the always-on checkpoint CLIP in the last slot. If you connect an always-enabled node to slot 1, it's never None, so slots 2–5 are dead weight and the switch silently ignores your careful setup.

It's also a quietly useful passer. Bypass a CLIP that's being read through a stubborn node and the multi-switch will still deliver it downstream - the same "fix nodes that can't see bypassed inputs" job the pack's passers do, which is what the author originally built this whole set for.

What's on the node

• input1–input5 - all optional CLIP inputs. All of them are forceInput, meaning they can only be wired from another node; you can't type a value into them. That's fine - you don't want to.
• clip output - a single CLIP, wired into whatever text encoder you use (CLIP Text Encode, or an RvTools pipe if you're in that style of graph).

One catch to know: if nothing is connected to any slot, the node doesn't quietly return nothing - it raises a hard "Missing Input" error. So it's not a passthrough with a fallback; it needs at least one live input.

Install

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

Or via ComfyUI Manager, searching "ComfyUI-RvTools_v2". No models to download; the pack itself is pure Python plus the usual torch/numpy/Pillow/opencv-python/pilgram deps.

One honest warning

The README banners this pack as no longer maintained and points to ComfyUI_Eclipse as the successor - the same author's newer pack, which folds most of these nodes in. And the whole RvTools lineage has had a rocky install history: the original ComfyUI-RvTools repo was pulled from GitHub in early 2025 and people were swapping folders around between installs to keep workflows alive. If a workflow you downloaded demands this node, v2 is the version to grab - it's frozen rather than dead, and it still works. If you're building fresh, peek at Eclipse first.