Pass Latent

Pass Latent is a pure passthrough for LATENT data - the compressed image representation that diffusion actually samples on before it's decoded back to pixels. One latent tensor goes in, the identical tensor comes out, and that's the whole feature list. If you've ever wanted to say "just move this thing along the wire without touching it," this is the node.

Latents are a little special to route because you can't glance at them and tell if they're right. An image wire, you can visually check. A latent is a black box of noise statistics, and by the time you realize a wire is pointing at the wrong node you've usually already burned the sampling pass. A passer gives you a named, stable junction in the graph - a single place where the latent from a sampler's output, or a latent about to be fed into the next KSampler, fans out to multiple consumers. If you later swap or bypass the upstream node, you reroute at the passer instead of chasing five dangling connections through a busy graph.

That's the exact problem the RvTools pack was built to solve. Per the README, passers were "created to be placed in front of nodes that had problems getting information from bypassed nodes," and they double as the second node you need to form a managed group, letting you hide the origin of a value and keep the group tidy. If you've ever run a multi-pass workflow - a base sample, a refine pass, maybe a hires pass - you've already got latent data flowing through several stages, and this is a clean place to control it.

Inputs and outputs

• latent (LATENT, required) - the latent batch to pass through.
• Output LATENT - the same latent, unmodified.

Note this is the plain v1 node: no Purge_VRAM toggle. If you want a latent passer that can also free VRAM while it passes, grab Pass Latent v2 from the same pack instead - this one stays deliberately dumb.

Install

From the ComfyUI-RvTools_v2 pack by r-vage. In ComfyUI Manager, search "ComfyUI-RvTools_v2" and install, or:

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

Restart ComfyUI afterward. No model downloads; the pack's dependencies (torch, numpy, Pillow, opencv-python, pilgram, etc.) are almost always already present.

Two things to keep in mind

The whole pack is marked deprecated in its README, superseded by ComfyUI_Eclipse - so this node works but won't gain features here. And if you're loading an old workflow that references RvTools v1, the original repo is gone from GitHub and those nodes won't resolve; v2 renamed the entire set, and installing this pack is how you get the graph working again. Use the passer when a latent fans out or a graph is getting unmanageable - for a straight one-to-one connection, a plain wire is all you need.