VAE Encode Pipe

The core VAE Encode node does one job: IMAGE in, LATENT out. Fine - until the graph needs that same VAE again on the decode side, which it almost always does, and the wire from the VAE loader to both encode and decode is exactly the kind of line that gets lost in a big canvas. VAE Encode Pipe is the same encode with the VAE passed through: IMAGE in, LATENT and VAE out, so encode, sampler and decode can all hang off one VAE supply. It's the "pipe" pattern - bundle a value with the thing that produced it.

How it works

Thin on purpose. It crops the image to a multiple of 8 (a centered crop, so you don't shave one edge off), runs the standard vae.encode on the RGB channels, and returns the latent plus the original VAE untouched.

Inputs: pixels (IMAGE), vae (VAE). Outputs: LATENT, VAE. That's the whole contract.

The honest problem

This node left the pack. It shipped in the braintacles pack's early releases, then was removed in May 2024 when the author slimmed the pack down to its current five nodes - the version in GitHub and ComfyUI Manager doesn't include it. If a workflow references it and ComfyUI reports it missing, that's why.

You can get the exact behavior with the core VAE Encode node and a second wire from your VAE loader to the decode. One extra line, identical result. Or, if you want the pipe for real, install the pack pinned to a pre-May-2024 commit:

git clone https://github.com/braintacles/braintacles-comfyui-nodes
cd braintacles-comfyui-nodes
git checkout 26732fd

Honestly? Not worth changing your install over. The pipe is a nicety, not a capability. If the missing-node message is what brought you here, the answer is: use core VAE Encode, run the VAE wire twice, move on.