Load VAE (Downloaded)

The VAE is the unglamorous part of every generation: it converts the latent noise your sampler works in back into actual pixels. Load VAE (Downloaded) is the piece of ComfyUI Model Batch Downloader that loads the vae category out of a batch result, and it's the simplest loader in the pack - one input pair, one output, no knobs. Which makes it sound pointless until you're assembling a split-model workflow and realize the decode half has to come from somewhere.

Where it fits

In an all-in-one checkpoint, the VAE is bundled inside the file, so the standard Load Checkpoint gives you a VAE output for free. Split-model setups don't work that way: the diffusion model loader and CLIP loader output MODEL and CLIP, and nothing hands you a VAE - you have to load it yourself. That's this node's job. The README's Anima and Krea 2 examples both list a separate VAE download (qwen-image-vae.safetensors in the Anima manifest), and this loader is what turns that download into pixels.

It's also the swap-friendly option. Different models sometimes want different VAEs - a finetune's recommended VAE can change the colors and fine detail noticeably - and having the VAE as its own file makes swapping it a manifest edit and a re-run instead of a file-folder hunt.

Inputs and outputs

• download_result - from the download node.
• id - the manifest ID of the VAE file (filename stem if unset).

Output: VAE, wired into VAE Decode at the end of the pipeline (or VAE Encode if you're doing img2img). That's genuinely the whole thing.

The one thing to get right is the manifest category: the file must be filed under vae, or this loader refuses it with the pack's standard category-mismatch error ("requires vae"). It's the same strict guard every loader in the pack has, and it catches the "I filed my VAE under checkpoints because I was tired" mistake before it becomes a confusing load failure.

Install and troubleshooting

Shared pack install: ComfyUI Manager search for "ComfyUI Model Batch Downloader", or clone into custom_nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/watarika/ComfyUI-Model-Batch-Downloader.git

then restart. No Python dependencies; aria2c on PATH is the one external requirement, and only the download half needs it.

Failure modes are the shared trio: an id that doesn't exist in the manifest (typo, or the download failed - interrupted downloads resume next run), a category mismatch, and the one people report as "my image looks washed out" - which usually isn't this node at all, but a VAE mismatch that shows up in the decode. If the colors are off and the VAE is the recommended one for your model, check that the manifest id actually points at the right file, then look downstream.