TT VAE Decode (Context)

In a normal ComfyUI workflow, decoding is two wires: latent into a VAE, VAE into the decoder. In TenserTensor's context style, it's one. TT VAE Decode (Context) takes a single TT_CONTEXT socket, pulls the VAE and the latent out of the context object it's carrying, decodes, and hands you both the updated context and the finished IMAGE.

This is the "context" pattern the pack is built around - the same idea that made rgthree's Context nodes famous, where a bundle of pipeline state travels down one wire instead of a spaghetti of connections. If you're already working in a context workflow, this node is how the latent becomes a visible image at the end.

What it does, precisely

The node has exactly one required input: context. It reads vae and latent from that context, runs a normal tiled VAE decode (vae.decode_tiled with 512px tiles and 64px overlap), and returns two outputs:

• CONTEXT - the same context, now with the decoded image stored in it, so downstream context nodes can keep chaining.
• IMAGE - the decoded pixels, ready for a save/preview node.

That's the whole node. Its job is removing the bookkeeping - you never have to wire the VAE again after it goes into the context, and you never have to remember where the latent came from.

Why you'd build around it

Context workflows trade a little magic for a lot of visual cleanliness. If you've got a complex pipeline - prompt, model, sampler, encoder, decoder, all feeding a single context that flows down the graph - then nodes like this are what make the graph readable at all. The pack's own docs are blunt that it raises an error if the VAE or latent is missing from the context, which is honestly a feature: you find out exactly which node forgot to set state, instead of getting a cryptic tensor crash.

The catch is the same one that applies to every context node in this pack: TT_CONTEXT is a proprietary socket type. It only connects to other TenserTensor nodes. Try to feed it into a stock ComfyUI node and you get a type mismatch. So this node is only useful if you're all-in on the context workflow - it's not a drop-in replacement for a plain decoder.

Install and the V1 caveat

cd ComfyUI/custom_nodes
git clone https://github.com/tenser-tensor/ComfyUI-TenserTensor

Or ComfyUI Manager → "TenserTensor" → install → restart. The pack needs gguf and kornia; Manager handles those.

And the standing warning for all the V1 nodes in this pack: TenserTensor migrated to ComfyUI's API V3, and the old V1 classes now sit in Deprecated/, slated for removal in a future major release. This node works today, but the pack's V3 context decoder is the one that'll survive long-term. If you're starting a new context workflow, check which one your installed version shows under TenserTensor/VAE - grab the one not labelled deprecated.