VAE Loader 🗝️

Most checkpoints bundle a VAE, and that's usually the one you want. But "usually" isn't "always," and the VAE Loader from TensorVizion/OmniNodes exists for the exceptions: a standalone VAE you want regardless of what the checkpoint ships, a community-finetuned decoder for an SDXL model, or a tiny taesd preview VAE when you just want a fast thumbnail while you iterate.

Like the pack's other model utilities, it's a thin wrapper - this time around ComfyUI's own core VAELoader, delegated directly so behavior is identical to the stock node, including the special-casing ComfyUI gives the fast-preview taesd/taesdxl/taesd3 files. The only addition is the pack's signature summary output (here, a simple "VAE loaded: <name>" line).

Inputs and outputs

One input: vae_name, a dropdown populated from ComfyUI's vae/ model folder. One meaningful output: vae, which you wire into a VAE Decode, VAE Encode, or anything else that takes a decoder. Plus summary.

When it matters

The VAE lore is genuinely relevant here: the right VAE is the one the checkpoint was trained against, not the one with the best reputation. The grey-wash problem that made standalone VAEs famous was an SD 1.5 thing - modern models mostly bake in the correct decoder, and bolting on a "better" one can make things worse. So this node earns its keep in three scenarios:

1. Explicit override - you want a specific decoder (a community fine-tune, or the correct standalone for a checkpoint that ships none).
2. Fast previews - taesd/taesdxl load near-instantly and decode much faster, at the cost of quality. Perfect for browsing seeds before a final full-quality decode.
3. Pipeline control - you're swapping the VAE independent of the checkpoint, and you want the graph to show it.

That last one is the sleeper benefit. When the VAE comes from a named loader node instead of being hidden inside a checkpoint loader, the workflow documents itself - and debugging "why are my colors off" starts with seeing exactly which decoder is in play.

Install

Ships with OmniNodes:

cd ComfyUI/custom_nodes
git clone https://github.com/TensorVizion/OmniNodes

Or ComfyUI Manager → OmniNodes → restart. No extra dependencies. It's categorized under TensorVizion/Model Utilities.

Troubleshooting

• VAE not in the dropdown - it must live under ComfyUI's registered vae/ search path (usually ComfyUI/models/vae/). Drop it there and refresh.
• Preview VAE isn't listed - the taesd files are small downloads; grab them into vae/ and they'll appear like any other. Core special-cases them by name.
• Node missing from menu - restart ComfyUI fully and check the [OmniNodes] lines in the terminal log.

Reach for it when you want the decoder to be an explicit, visible, swappable part of the graph rather than a hidden assumption.