☁️BizyAir Flux 2 API

This node calls Black Forest Labs' Flux 2 API through SiliconFlow - no local weights, no VRAM budget to plan around, just a prompt and a bill. There are two tiers on offer, and which one you pick changes which parameters on this node actually do anything.

How it works

flux-2-pro is the streamlined tier - you get a prompt, dimensions, and a seed, and BFL's backend handles the rest. flux-2-flex is the tier that exposes real dials: per BFL's own naming, Flex is the API tier built specifically to expose steps and guidance scale so you can trade speed against quality per request. That's exactly why guidance and steps are both marked "flex only" in this node's tooltips - they're inert unless you've set model to flux-2-flex.

The inputs and outputs that matter

• prompt - required, multiline.
• model (default flux-2-pro, or flux-2-flex) - picks the tier, and gates whether guidance/steps do anything.
• seed (0 to 2,147,483,647).
• width / height (default 1024, step 16) - must be a multiple of 16 per the node's own tooltip. Setting either to 0 tells it to match your input image's corresponding dimension instead of a fixed size - only meaningful if you've plugged something into images.
• safety_tolerance (default 2, 0-5) - the node's tooltip is direct: a larger value is looser content filtering.
• guidance (default 4.5, 1.5-10) - flex only, per the tooltip. Standard Flux guidance behavior: higher pushes stricter prompt adherence at the cost of a more "processed" look.
• steps (default 50, 1-50) - flex only, per the tooltip.
• images (optional, IMAGE) - feed a reference image for image-conditioned generation; leave it empty for pure text-to-image.

Outputs: image, plus bizyair_model_name - a small mapping (flux-2-pro → flux-2, flux-2-flex → flux-2) identifying which billing bucket the call landed in.

How to install it

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

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

then restart. This is a genuine paid cloud call, not a local model - set your API key on first use as the README instructs, and expect it to draw from your BizyAir credit balance per generation.

Common issues & troubleshooting

guidance/steps changes have no visible effect. You're almost certainly on flux-2-pro. Switch model to flux-2-flex to make those two parameters live.

Odd output dimensions with an image plugged in. If width or height is left at 0, the node uses the input image's matching dimension instead of a fixed value - that's expected behavior, not a bug, but it's easy to forget you set it that way.

Straight text-to-image runs behaving unpredictably at width/height 0. With no images input there's nothing for a 0 to match against - set explicit dimensions for pure text-to-image rather than relying on the fallback.