TT Vae Encode (Context)

In a TenserTensor context graph, the pipeline state lives in one bag - and that means even something as basic as "turn this image into a latent" can be a one-input operation. TT Vae Encode (Context) takes a single CONTEXT, pulls the vae and the image out of it, encodes the image to a latent, and stuffs that latent back into the context for the next stage. It's the context-aware version of VAEEncode, and it exists so you can do img2img or inpaint-style passes without re-wiring the whole graph.

It's a thin wrapper, mechanically. The node reads vae and image from the context, calls a tiled VAE encode (512×512 tiles with 64 overlap under the hood, so big reference images don't blow VRAM during encoding), wraps the result as a latent, and writes it to the context's latent slot. Two outputs: the updated CONTEXT (with latent populated) and the raw LATENT, so you can either keep flowing through the context system or hand the latent to a native node. The plain LATENT output is the escape hatch - if you don't want to stay in the walled garden, that's your way out.

Its strictness is its most useful feature. It validates the context before doing anything: no vae and it raises "VAE is required for encode"; no image and it raises "Pixel image is required for encode". Those errors are genuinely readable, and they make it obvious when you've wired a context that never got an image loaded. If you're chaining img2img stages, this is the node that quietly resets the latent between passes.

Inputs and outputs are the whole schema, so there's nothing to tune:

• Input: context (TT_CONTEXT).
• Outputs: CONTEXT (updated) and LATENT.

Install is the pack standard:

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

or search "TenserTensor" in ComfyUI Manager and restart.

The honest framing: this node is only useful inside the pack's context system, and that system has a real constraint - TT_CONTEXT is proprietary to TenserTensor. You can't take a context built here and feed it to a native VAEEncode; the context wire only connects within the pack. If you're not running a context-based graph, the native VAEEncode does the same job with fewer moving parts. Use this node because you've bought into the context pipeline and want the "one wire carries everything" workflow, not because it unlocks something the native node can't do.

Last caveat, same as the rest of the pack: this is the V1 class, marked deprecated as the author migrates to ComfyUI's API V3 (TT_VaeEncodeContextNode is the maintained successor). Works fine today; if you're building a context graph from scratch, start with the V3 version.