BigPlayer LLM Provider

BigPlayer LLM Provider is the entry point of every BigPlayer workflow, and the name is doing more work than it deserves. It doesn't run a model, hold any weights, or call anything by itself. What it actually does is bundle your xAI (Grok) API key, model choice, and a caching flag into one provider_config value that the rest of the pack consumes. It's a config card you wire in once and reuse.

BigPlayer's whole pitch is "LLM-assisted, honestly." The transformation of your natural-language prose into prompts, sampler settings, and checkpoint picks is delegated to a cloud model; the local code just prepares the input, enforces a schema, caches, and validates. This node is the front door for that delegation. It's also the only place in the pack that touches a secret.

The inputs you actually set

• api_key - your xAI key from console.x.ai. A plain string field. The node never echoes it back, and the code redacts it from cache hashes and error messages, so it doesn't leak into debug output either.
• provider - a dropdown with xAI and No Provider. The second is a local passthrough that "goes nowhere, does nothing" - useful for testing your wiring and schema plumbing without spending a cent.
• provider_model - grok-4-1-fast-non-reasoning (default), grok-4-1-fast-reasoning, grok-4-latest, plus No Provider's Positive/Negative options. The web front-end swaps the model list when you change provider.
• provider_base_url - leave empty; it defaults to https://api.x.ai/v1. The override is there for proxies.
• assume_determinism - default on. This is the caching switch: identical requests reuse the previous result instead of hitting the API again.

The single output, provider_config, feeds into BigPlayer Natural Language Root. That's the whole graph contract.

Install

ComfyUI Manager, search "BigPlayer Prompting", or:

cd ComfyUI/custom_nodes
git clone https://github.com/josh-ent/comfyui-bigplayer-llm-nodes

Then restart ComfyUI. The dependencies are light - httpx and pydantic - and there are no model downloads. The only "model" is your API key. It wants ComfyUI >= 0.16.4.

Where people get burned

The classic mistakes are all on this node. Forgetting the API key gives you a hard validation error at graph time, not a mysterious failure mid-run. Leaving assume_determinism on while you iterate on your prose means you'll keep getting the cached answer - flip it off when you want a fresh call every run. That's also the switch that makes it cost you money, because with caching on, unchanged runs are free. And if you're coming from the local-LLM ComfyUI world (the many local prompt nodes that run a small model on your GPU), this is the opposite trade: no VRAM cost, but your prompts leave your machine and go to xAI. Worth knowing before you build a workflow on it.