☁️BizyAir SiliconCloud LLM API

SiliconCloud is SiliconFlow's own LLM hosting product - SiliconFlow being the company behind BizyAir itself, so this node is BizyAir reaching into its maker's other business rather than a third-party integration. What it does is simple and genuinely useful: you write a rough idea, it runs that through a language model with a prompt-engineering system message baked in, and you get back a properly structured Stable Diffusion prompt. It's a prompt-expansion node, not an image node - the output is text you then feed into your actual text encoder.

What it does

system_prompt ships with a default that instructs the model to act as a "Stable Diffusion prompt expert," expand whatever word or phrase you give it into a prompt suitable for image generation, output only in English, and print nothing except the prompt itself - you can leave it as-is or override it entirely if you want a different framing. user_prompt is your actual input; the shipped default demonstrates the idea with "a kitten, Van Gogh style," expanded into a fuller SD-ready description.

max_tokens (100–100,000, default 512) caps response length - 512 is plenty for a single-image prompt; push it higher if you're generating long, detailed multi-clause prompts. temperature (0–2, default 0.7) controls how much the model varies its phrasing between calls - lower for consistent, predictable expansions, higher if you want more creative divergence from run to run.

The one field that needs a heads-up: model shows up in the node's schema with zero available choices until you've authenticated. That's not a bug - the dropdown populates live from whatever models your BizyAir/SiliconCloud account actually has access to, so an empty list means "set your API key first," not "this node is broken."

Output

A single STRING - wire it straight into your positive prompt's text encode node, or into a "Show Text" node first if you want to see exactly what came back before committing to it downstream.

Installing it

ComfyUI Manager: search BizyAir, install, restart. Or by hand:

cd ComfyUI/custom_nodes && git clone https://github.com/siliconflow/BizyAir.git

Restart afterward. No model to download - this calls out to SiliconCloud's hosted LLMs rather than running anything locally. You need a BizyAir account with an API key configured before the model dropdown will populate at all, let alone before the node can run.

Where people get burned

Seeing an empty model dropdown and assuming the install is broken. It isn't - it's waiting on your API key. Set it through BizyAir's login flow and the list fills in.

Leaving the default system prompt in place for a different task. It's specifically tuned to expand a short phrase into a Stable Diffusion prompt - if you're trying to use this node for something else (captioning, general text generation), you'll want to rewrite system_prompt from scratch rather than relying on the shipped default.

Not checking what actually came back. LLM-generated prompt expansions can wander - drift into unrelated details, add style language you didn't ask for, or occasionally ignore an instruction in the system prompt. Route the STRING output through something you can read before it hits your sampler, at least the first few times.