TT CLIP Text Encode SDXL (Context)

Here's the version of the SDXL text encoder that doesn't have text boxes. TT CLIP Text Encode SDXL (Context) takes one TT_CONTEXT input and produces POSITIVE and NEGATIVE conditioning - the prompts themselves aren't on this node at all. They live in the workflow config that's traveling inside the context.

If that sounds backwards, it's because this node belongs to the pack's context style, where state gets bundled into a single object and passed down one wire. In that world, the SDXL workflow looks like: a workflow-settings node holds the prompts and conditioning params, TT Base Context bundles them with the model/CLIP/VAE, and this node quietly does the encoding step when the context reaches it.

What it actually reads

Under the hood it's the same encoding as the non-context TT_ClipTextEncodeSdxl - CLIP-G tokens with CLIP-L tokens overlaid, plus the aesthetic score and dimension conditioning baked in. It pulls all of those from the context's workflow config: clip_l_positive, clip_g_positive, the negative pair, ascore_positive/ascore_negative, and the four width/height/target_width/target_height values.

Outputs are CONTEXT (the context, now carrying the conditioning so downstream nodes can use it) plus the two CONDITIONING outputs, which you can wire anywhere a sampler expects them.

Why you'd use this over the plain encoder

Two reasons. First, cleanliness: one wire in, one wire out, no prompt fields cluttering the middle of your graph. Second, consistency: prompts live in exactly one place (the workflow settings node), so you never have three copies of your prompt drifting out of sync across a big graph. It's the same argument people make for rgthree's Context nodes - at scale, a bundle of state down one wire beats a spaghetti of connections you have to trace.

The cost is the same as every context node in this pack: TT_CONTEXT is a proprietary socket type that only connects to other TenserTensor nodes. You can't feed this node from a stock node's context, and its output conditioning is ordinary CONDITIONING so that part stays portable - but the context chain itself is locked to this pack.

Install and caveats

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

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

Two things to know before you build on it. First, it's a legacy V1 node - TenserTensor migrated to ComfyUI's API V3 and parked V1 nodes in Deprecated/, to be removed in a future major release. Second, and more importantly for beginners: if the context's workflow config is missing any of those prompt fields, this node will happily error out, and the error message tells you the encoder needs its config - which means you skipped or mis-wired the workflow-settings node upstream. Trace the context back to its source before you assume the node is broken.