TT KSampler (Context)

A normal KSampler asks you to hand it a model, a latent, conditioning, a seed, steps, CFG, a sampler name, and a scheduler - every single time, on every single node, until your graph looks like a plate of spaghetti. TT KSampler (Context) asks for exactly one thing: a TT_CONTEXT. Everything else it needs is already inside.

That's the whole trick. This node is the sampling step in TenserTensor's context pipeline. You feed it the context object (the pack's "one wire carries the whole pipeline" bag), and it reads model, latent, positive, and negative straight out of the context, pulls seed, steps, cfg, sampler_name, and scheduler from the embedded workflow_config - the thing the TT FLUX Workflow Settings or TT SDXL Workflow Settings nodes produce - and runs a completely standard ComfyUI sample. It even attaches the latent preview callback, so you get step-by-step previews in the UI like you would from the native sampler.

The mechanism under the hood is just comfy.sample.sample() wrapped up: it fixes latent channels for the model, prepares noise from the seed, runs the scheduler, and writes the result back into the context's latent slot. The two outputs tell the story: the updated CONTEXT (with latent now holding your samples, ready for the next context-aware node) and the plain LATENT output for when you want to break out of the context system and use a native VAEDecode instead. That second output is the smart escape hatch - you get the tidy context pipeline on the way in and the freedom to leave it on the way out.

The inputs and outputs are minimal enough to list whole:

• Input: context (TT_CONTEXT) - the only thing it takes.
• Outputs: CONTEXT (updated, with latent filled in) and LATENT (plain latent for native nodes).

Because it relies on context contents, it's strict about what must be present. Missing model, missing latent, or a context that never got a workflow_config and it raises a ValueError telling you exactly which one is required - the error messages are unusually readable ("Model is required for sampling", "Workflow Config is required for sampling"). That's the practical thing to remember: the TT_*Context family only works if the context actually carried the field you're asking for. If you hand it a context built by a node that never set workflow_config, sampling can't happen.

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.

Two honest gotchas. First, TT_CONTEXT is proprietary to this pack - you can't build a context in one pack and sample it with native nodes, so the context pipeline is all-or-nothing within TenserTensor. Second, this class is the V1 node, currently marked deprecated as the author migrates to the API V3 rewrite; the TT_KSamplerNode replacement is where new development goes. It still works, and if you're already running a TenserTensor context graph, it's the only sampler you need.