LLS Simple VAE Decode

Let's be honest about what this node is: samples in, image out, exactly like ComfyUI's built-in VAE Decode. There's no novel algorithm here - the implementation literally calls ComfyUI's native VAE.decode under the hood. What LLS Simple VAE Decode adds is one output you didn't have before and a friendlier error when you did something wrong.

So why would you reach for it? Two reasons. First, consistency: if you're building LLS-pack workflows with the Simple VAE Encode or the repair chain, keeping decode in the same family means every step of the pipeline speaks the same metadata language. Second, the decode_info output - a small JSON string recording the resolved model family and the decode settings - is the kind of breadcrumb that makes a failing workflow diagnosable. And it gets the Flux case right with a clear error message instead of a confusing traceback.

Inputs and outputs

Two required inputs: samples (the LATENT from your sampler) and vae (the VAE - for Flux, that's the separate ae.safetensors, which the pack does not install for you). Two outputs: image (an IMAGE tensor for Preview/Save) and decode_info (a STRING with the metadata JSON).

The node infers the model family from the latent or the VAE and includes it in decode_info - mostly useful as a sanity check, and genuinely useful the day you wonder whether a weird result came from decoding with the wrong VAE.

Wiring it in

It slots anywhere you'd put a core VAE Decode:

LLS Simple KSampler → LLS Simple VAE Decode → Preview Image / Save Image

or the tail of the repair chain:

LLS Simple KSampler → LLS Simple VAE Decode → LLS Simple Repair Finish → Preview Image

Installing

Via ComfyUI Manager (search "LLS-node"), or:

cd ComfyUI/custom_nodes
git clone https://github.com/Gin3601/LLS-node

Restart, look under LLS/Image. The pack has no requirements.txt; this node uses only ComfyUI core.

Gotchas

• If you're on FLUX and the VAE is missing, the node tells you precisely: the source raises [LLS] Missing FLUX AE/VAE. Place ae.safetensors in ComfyUI/models/vae/. Do exactly that.
• This is a thin wrapper, so treat it as such - if you need tiled VAE decode for huge latents (VRAM constraints on 4K outputs), reach for a tiled decoder instead; this node is the simple path, on purpose.
• Remember that VAE decode is where a lot of "my image looks washed out / colors are off" reports actually originate - if that's your symptom, the decoder isn't at fault, the VAE file you fed it is. decode_info tells you which VAE was used, which is the first fact you want when debugging that.