TT FLUX Workflow Settings (Advanced)

FLUX is a two-encoder model: a small CLIP-L for the tag-like structure and a huge T5-XXL that actually understands English sentences. It also runs on a guidance scale, not old-school CFG, and it doesn't really do negative prompts. None of that is obvious when you're staring at a blank graph. TT FLUX Workflow Settings (Advanced) collects all of it - sampler choices, step count, both prompts, and guidance - into one node, then hands the whole bundle to the rest of the pack as a single config object.

It's the "advanced" version of the pack's FLUX settings node: same sampling controls, plus four prompt fields for the two encoders (positive and negative each).

The inputs

• seed, steps (25), cfg (1.5), sampler_name, scheduler - the usual sampling stack.
• clip_l_positive and t5xxl_positive - two separate positive prompts. This is the feature: write short tags for CLIP-L and natural language for T5. FLUX's dual-encoder setup is exactly why people combine "LLM-style sentences + keywords" - the KB's notes on FLUX prompting describe this middle ground.
• clip_l_negative and t5xxl_negative - the negative pair. Be honest with yourself here: FLUX is guidance-distilled, and negative prompts are mostly inert on it. The boxes exist; just don't expect them to do what they did on SDXL.
• guidance (default 3.5) - FLUX's guidance scale. The pack defaults it to 3.5, which matches FLUX's documented setting. Community experience says 3.5 gives everything the same face; for people specifically, drop it to 1.8–2.0.

The outputs - this is the part to understand

Every input comes back out as an individual socket (SEED, STEPS, CFG, SAMPLER_NAME, SCHEDULER, CLIP_L_POSITIVE, T5XXL_POSITIVE, CLIP_L_NEGATIVE, T5XXL_NEGATIVE, GUIDANCE), so you can wire any single value wherever you like. Plus one WORKFLOW_CONFIG output - a TT_WORKFLOW_CONFIG object that bundles everything.

Here's the gotcha, straight from the pack's README: TT_WORKFLOW_CONFIG is proprietary to TenserTensor. It only plugs into other TenserTensor nodes - the context nodes, the context FLUX encoder. Feed it to a stock node and you get a type mismatch. So this node is the front door to a TenserTensor context workflow: settings in, then context nodes carry the config forward.

Install and the standing caveat

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

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

Like every node in this pack's V1 set, it's marked deprecated - the pack moved to ComfyUI's API V3 and plans to drop V1 in a future major release. It works today, and honestly the workflow-settings pattern is the pack's most distinctive idea: one node holding all global settings, instead of values scattered across a dozen samplers. The trap is mixing it with non-TenserTensor nodes - the config object won't cross the pack boundary, so decide up front whether you're building a TenserTensor-native workflow or a stock one.