load available vae 载入可用的VAE

The VAE version of the pack's available loaders. Where AvailableCheckpointLoader (2lab) filters checkpoints to the 2lab platform's supported list, this one does the same for VAE files - the dropdown only contains VAEs from the pack's standardized/vae.json, and the output is a single VAE socket. Pick one, wire it into your VAE decode, done.

The supported VAE list is a short, sensible one: sdxl_vae.safetensors for SDXL, vae-ft-mse-840000-ema-pruned.safetensors for SD 1.5, and ClearVAE if you want a third-party option. Those three cover the overwhelming majority of real usage - they're the same VAEs the community actually reaches for. If you're an SD 1.5 user, the vae-ft-mse-840000 is the one you want (it's the standard recommendation for the grey/washed-out problem on SD 1.5 outputs); for SDXL, sdxl_vae is the baked-in default you're mostly confirming rather than choosing.

The inputs and outputs that matter:

• vae_name - the only input, a dropdown of the supported VAEs. That's the entire interaction.
• VAE out - feeds your VAE decode node.

The mechanism matches the checkpoint loader: it scans your local models/vae/ folder, keeps the files that are on the supported list, and shows those. It won't invent options, and it will refuse to build a workflow on a VAE the platform can't host.

The honest take is the same as its checkpoint sibling: if you're not publishing to the 2lab platform, this node is a restricted version of core "Load VAE" and you should probably just use the core one. The supported list is genuinely fine, but "fine and short" is not an improvement over "the full dropdown." Its real purpose is keeping published workflows honest - the 2lab server can only run what it hosts, and this node makes that constraint visible at build time instead of at publish time.

Installing it

Part of AI2lab/comfyUI-tool-2lab. ComfyUI Manager → search "comfyUI-tool-2lab" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/AI2lab/comfyUI-tool-2lab

Restart ComfyUI. No extra dependencies; the VAE files themselves you'd grab from the usual places (they're standard, widely mirrored models).

Common issues

An empty dropdown means no supported VAE is sitting in your models/vae/ folder - drop one of the listed files in and restart. If a workflow built with this node refuses to run on your machine, it's the same "not in available list" guardrail as the checkpoint loader: it's protecting the published-app path, not your local setup. And one genuinely useful tip: if you're on SD 1.5 and your outputs look grey or washed out, the fix everyone recommends is the vae-ft-mse-840000-ema-pruned that's sitting right in this list - swap it in and the color saturation problems usually vanish.