TT SDXL Workflow Settings (Advanced)

The plain TT SDXL Workflow Settings holds your sampler knobs; this one holds your sampler knobs and your prompts. TT SDXL Workflow Settings (Advanced) bundles the sampling settings together with SDXL's two-encoder prompt split - clip_l and clip_g, positive and negative - plus aesthetic scores and the size pair, and hands it all over as a single WORKFLOW_CONFIG. It's the settings card for a graph that wants everything defined in one place.

SDXL doesn't take one prompt, it takes two: the short CLIP-L summary (the "what's in the picture" lane) and the longer CLIP-G description (the "how it looks" lane), conditioned at different resolutions. This node gives you four real text boxes - clip_l_positive, clip_g_positive, clip_l_negative, clip_g_negative - so you can keep those lanes separate instead of cramming everything into one string. Real SDXL prompters do exactly this: a terse CLIP-L like "portrait of a woman, studio light" and a detailed CLIP-G filling in pose, lens, and mood. Get the two lanes right and SDXL stops fighting you.

The extra inputs round out the standard SDXL conditioning bundle:

• ascore_positive / ascore_negative - the aesthetic score conditioning, defaulted to 9 and 6. That 6/9 pair is what most SDXL base-model pipelines run; leave it alone unless you're deliberately tuning aesthetics.
• width / height - the generation size (512×512 defaults).
• target_width / target_height - the target size for a refiner or hires pass. The context carries both so a later stage knows what the image was made at and what it's heading to.

Outputs mirror everything: WORKFLOW_CONFIG, SEED, STEPS, CFG, SAMPLER_NAME, SCHEDULER, the four prompt strings, ASCORE_POSITIVE/ASCORE_NEGATIVE, and the size fields. The context text encoders in the pack consume the prompt strings straight out of the config, so a full TenserTensor graph becomes: settings card → Large Context → context text encoder → KSampler (Context), with every value defined in exactly one place.

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.

Same walled-garden note as its sibling: TT_WORKFLOW_CONFIG only connects to other TenserTensor nodes. The individual outputs are standard types though, so you can use this purely as a prompt-and-settings panel feeding native CLIPTextEncodeSDXL nodes if you prefer - the four prompt strings and sizes are exactly what that node wants. And yes, this is the V1 class, deprecated in favor of the API V3 TT_SdxlWorkflowSettingsAdvancedNode. Works fine today; new graphs should start on the V3 version.