TT Base Context

Every context workflow has to start somewhere. TT Base Context is the somewhere: it takes the four things every generation pipeline needs - model, clip, vae, latent - plus an optional workflow_config, and bundles them into a single TT_CONTEXT object. From there, everything downstream travels on one wire.

It's the foundation node of TenserTensor's context system, which is the pack's version of the "carry a bundle of state down one wire" pattern that rgthree made famous in ComfyUI. Where rgthree's context stays interoperable with stock nodes, TenserTensor's is a proprietary socket type - which is both its superpower (one wire instead of four) and its cage (it only connects to other TenserTensor nodes).

What it does, precisely

Four required inputs in, one CONTEXT out:

• model - the diffusion model (UNet/DiT).
• clip - the CLIP text encoder.
• vae - the VAE for encoding/decoding.
• latent - the starting latent (usually from a latent factory or a VAE encode).
• workflow_config (optional) - the pack's TT_WORKFLOW_CONFIG object from a workflow-settings node, carrying your prompts, seed, steps, guidance, and so on.

Under the hood it just stuffs those into a context dict. The interesting part is what happens later: the pack's context-aware nodes - the CLIP text encoders, samplers, VAE decode - read from this same object instead of asking you to re-wire everything. So the graph reads: load models → TT Base Context → context encoder → context sampler → context decode, with each node quietly pulling its inputs out of the context.

The mental model

Think of the context as a backpack that walks down the graph, and TT Base Context as the moment you pack it. Its children are the setters (like TT Context Set Latent and TT Context Set Image) that add or replace items in the backpack, and the consumers (the context encoder and decoder) that take them out.

The optional workflow_config input is where the pack's whole workflow-settings idea clicks: plug a TT FLUX or SDXL Workflow Settings node in there and your prompts and parameters live in one place, in one object, instead of scattered across a dozen node fields.

Caveats and install

The proprietary-type warning applies at full strength here: the CONTEXT output only plugs into other TenserTensor context nodes. Mixing with stock ComfyUI nodes gives a type mismatch - so TT Base Context commits you to the pack's context style for the rest of that graph.

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

or ComfyUI Manager → "TenserTensor" → install → restart. Deps: gguf, kornia.

It's also a legacy V1 node. The pack migrated to ComfyUI's API V3 and moved V1 context nodes into Deprecated/, with removal planned for a future major release. It still works - and it's the pack's cleanest introduction to how its context flow works - but if you're starting fresh, prefer the pack's V3 context base node if your installed version exposes one. One beginner trap: because everything is required, a missing wire (say, no latent) errors at execution, and the fix is usually back at this node, not downstream.