BigPlayer KSampler Config

BigPlayer KSampler Config folds the fiddliest part of a workflow - picking steps, CFG, sampler, scheduler, denoise - into the same LLM call that writes your prompt. It reads the shared session from BigPlayer Natural Language Root and outputs steps (INT), cfg (FLOAT), sampler_name, scheduler, denoise (FLOAT), and comments. Wire them straight into a KSampler and the whole sampling config is prose-driven.

The mechanism is where this node earns its keep. When the root builds the request, this module resolves the live sampler and scheduler lists from your actual ComfyUI install (comfy.samplers.KSampler.SAMPLERS / SCHEDULERS) and bakes them into the JSON schema as enums. The LLM can only choose a sampler you could genuinely pick in the dropdown - then local validation checks it again. It literally cannot hallucinate res_2m or a sampler you don't have installed. That's also why the list shows the newer *_cfg_pp and seeds_* variants: recent ComfyUI ships them, and this node just passes your environment through.

The settings context

The LLM is told to "choose practical settings that match the user's request, and the chosen model and LoRA preferences where known" and that it "may need to do research" - which, with Grok, means it can actually look things up. The broader knowledge here: sampler/scheduler pairs are architecture-specific. On SDXL-lineage models, DPM++ 2M Karras remains the comfortable default. On flow-matching models, aggressive schedules like Karras and exponential can distort rather than correct, and guidance-distilled models want CFG 1 by default, not 5–7. A good LLM knows the model family and picks accordingly. You still get a veto: don't wire its outputs if you disagree, or edit the prose to say "SDXL, quality-focused, 30 steps."

Troubleshooting

• "Unsupported sampler" validation error - the provider returned a sampler that isn't in your list. Usually means your ComfyUI changed between the cached response and now. Flip assume_determinism off on the provider node to force a fresh call, and update ComfyUI.
• Denoise out of range - the schema constrains denoise to 0.0–1.0, steps to 1–10000, and CFG to 0–100, and rejects anything outside. If you see a range error from this node, the provider broke the contract.

Install is the pack install: ComfyUI Manager, search "BigPlayer Prompting", or git clone https://github.com/josh-ent/comfyui-bigplayer-llm-nodes into ComfyUI/custom_nodes, then restart. Deps are httpx and pydantic; no model downloads, but the root needs a working Grok key.