VAE Decode & VAE Encode (Tiled)
Translates compressed 8x latent tensors into RGB images and back, with tiled variants that eliminate out-of-memory spikes at 4K and in video runs. Package: Built-in Core.
Updated Oct 17, 2026 · 1 min read
- Node / Pack
- VAE Decode & VAE Encode (Tiled)
- Category
- latent
- Inputs
- samples (LATENT) or pixels (IMAGE), vae
- Outputs
- IMAGE or LATENT
- Source Pack
- Built-in Core
What VAE Decode & VAE Encode (Tiled) does in a ComfyUI graph
Translates compressed 8x latent tensors into RGB images and back, with tiled variants that eliminate out-of-memory spikes at 4K and in video runs.
On the canvas, this node accepts `samples (LATENT) or pixels (IMAGE), vae` and emits `IMAGE or LATENT`. Every pin is color-coded by data type so you cannot accidentally wire a pixel `IMAGE` tensor into a `LATENT` pin without an intermediate encode step.
Installation and missing-node recovery on Windows
VAE Decode & VAE Encode (Tiled) ships inside the default ComfyUI core on Windows. Double-click anywhere on the empty Studio canvas and type its name to drop it into your graph.
Download the full Windows desktop or portable package at /download to get started with a clean Python 3.12 environment.
FAQs
How do I fix "VAE Decode & VAE Encode (Tiled)" showing up as a red box in ComfyUI?
Update your ComfyUI installation to the latest build from /download so all recent core nodes are registered.
What are the main inputs and outputs of VAE Decode & VAE Encode (Tiled)?
Inputs: samples (LATENT) or pixels (IMAGE), vae. Outputs: IMAGE or LATENT.
Run on your hardware. Free forever.
The full ComfyUI engine: open source, fast, extensible, and yours to run however you want. Download the desktop app and open this workflow on your own machine.
