TT FLUX Workflow Settings

If you've ever stared at a Flux workflow with three different KSamplers each holding their own copy of seed, steps, and guidance, this node is the answer to a question you didn't know you were asking. TT Flux Workflow Settings is a single card where you set your sampling parameters once, and it hands them to the rest of the pipeline bundled into one WORKFLOW_CONFIG object - or as six separate typed outputs, whichever you prefer.

It's pure data. No model loading, no sampling, no math on its own. Under the hood the node stuffs seed, steps, cfg, sampler_name, scheduler, and guidance into a dict, and that dict travels inside the pack's context system. The TT KSampler (Context) node reads workflow_config out of the context and runs the sampler without you ever touching a second settings panel. The whole design goal of the pack is fewer wires and less clutter, and this is the knob panel that makes that work.

The defaults are worth a second look because they encode how Flux actually behaves. Flux is guidance-distilled: it has no true CFG and no real negative prompt, so the community runs it with a guidance scale instead. The node defaults to guidance 3.5, which is the stock setting - but note the KB's advice that 3.5 is widely blamed for the same-face "Fluxmaxxed" look. Drop guidance toward 1.8–2.0 if everyone in your output looks related. The cfg field defaults to 1.5 and behaves like a secondary scale; you generally leave it near 1 and do the tuning on guidance. steps defaults to 25, which sits right in the comfortable 20–30 band for Flux Dev.

Inputs that matter, realistically:

• seed - your reproducibility knob; wire the SEED output back if you want to track it.
• guidance - the actual quality dial for Flux.
• sampler_name / scheduler - euler + simple is the safe default; wrong scheduler choices give muddy output.

The outputs are the whole point: WORKFLOW_CONFIG feeds the context nodes, and each individual value (SEED, STEPS, CFG, SAMPLER_NAME, SCHEDULER, GUIDANCE) is also exposed separately so you can use them in native ComfyUI nodes too - that's the escape hatch when you don't want to buy into the context system.

Install is the pack standard:

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

or grab "TenserTensor" from ComfyUI Manager and restart. Only real dependencies are gguf and kornia.

The one trap to know: TT_WORKFLOW_CONFIG is proprietary to this pack. You can only plug it into other TenserTensor nodes. Try to run it into a native ComfyUI node and you'll get a type mismatch. That's not a bug - it's the wall the pack puts around its context pipeline. If you don't want the whole context system, skip the WORKFLOW_CONFIG output entirely and just use the individual outputs, which are plain INT/FLOAT/STRING types that work anywhere.

Also note: this is the V1 node and the pack labels it deprecated as it migrates to the API V3 rewrite. It works fine today. If you're starting a fresh graph, the V3 TT_FluxWorkflowSettingsNode is the maintained path - but for existing workflows this node keeps chugging along.