☁️BizyAir KSamplerSelect

This one is refreshingly small. BizyAir_KSamplerSelect takes exactly one input, sampler_name, and gives you back a SAMPLER object. That's it. No model, no conditioning, no cloud round-trip that you'd notice, it's just packaging a name into a typed object that other nodes can consume.

The reason it exists at all is the split between the monolithic KSampler and ComfyUI's "custom sampling" family. The regular KSampler bundles model, conditioning, sampler, scheduler and steps into one big node that does everything in a single call. Custom sampling breaks that apart: a node picks the sampler, a separate node builds the sigma schedule, another applies conditioning, and a final SamplerCustom-style node stitches them together and actually runs the denoise loop. That's more wiring, but it's the only way to do things a plain KSampler can't, like feeding your chosen sampler into a sigma-manipulation node (BizyAir ships one, Multiply Sigmas) or a sampler-wrapper like BizyAir Lying Sigma Sampler before it ever touches a model.

Where it plugs in

The SAMPLER output has no BIZYAIR_ prefix, unlike model or conditioning sockets elsewhere in the pack, and that's a real signal, not just naming noise: this node isn't calling out to SiliconFlow's cloud at all, it's a pure local lookup of a sampler function by name. You'll typically wire its output straight into BizyAir_LyingSigmaSampler's sampler input (which wraps it with a sigma-scaling trick) or into a custom-sampling KSampler node further down the chain that does need a BIZYAIR_MODEL.

The 44 choices on offer are the full stock ComfyUI sampler list, euler, euler_ancestral, heun, dpm_2, lms, dpm_fast and on through the rest, nothing BizyAir-specific added or removed. Which one to actually pick still comes down to what kind of model you're running: DPM++-family samplers on a Karras schedule are the safe call for SD 1.5/SDXL-derived checkpoints, while flow-matching architectures like Flux and Z-Image want Euler-family samplers and specifically do not want Karras or exponential scheduling, that combination is a documented failure mode on those architectures rather than merely a worse pick. If you're not sure which family your model belongs to, check the model card before you check this dropdown.

Installing it

Through ComfyUI Manager, search "BizyAir" and install. Or manually:

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

Restart ComfyUI afterward. Because this specific node is a local lookup with no API call, it's one of the few in the pack you could technically use without a BizyAir API key set up, but the moment its SAMPLER output feeds into anything that touches a BIZYAIR_MODEL, you're back to needing the account and key from the README's "click to login" step. Signup runs through SiliconFlow and involves phone verification, which some users outside China have reported trouble completing.

Common issues

Because this node does so little, there's not much to go wrong on its own. If you build a whole custom-sampling chain around it and the run fails or hangs, the API key on the downstream cloud nodes is the first thing to check, not this selector. If the sampler you picked here behaves worse than expected on a flow-matching checkpoint, it's very likely a scheduler mismatch further down the chain (Karras where you needed beta or simple), not a problem with the selection itself.

Also worth knowing if you're troubleshooting an install rather than a workflow: BizyAir occasionally gets pulled into a ComfyUI install automatically when Manager misresolves an unrelated missing-node error on an older shared workflow. It's legitimate software, not malware, and removable through the Manager if you didn't want it.