☁️BizyAir Kling 2 Text To Image

What it is

Kuaishou's Kling is best known as a video model - it's a name that comes up constantly whenever people size up what the closed, paid services can do compared to Wan or LTX locally. This node is its plain text-to-image side, called through BizyAir so you're spending your existing account credits instead of setting up a separate Kling subscription. If you're already using this pack's Kling video nodes, this is the still-image counterpart that shares the same underlying account and billing.

The inputs that matter

• prompt - what you want generated.
• negative_prompt - what to steer away from.
• model_name - locked to kling-v2.
• aspect_ratio - eight options: 16:9, 9:16, 1:1, 4:3, 3:4, 3:2, 2:3, 21:9.
• resolution - 1K or 2K. This node offers the higher tier that the image-to-image Kling node in this pack doesn't.
• variants - 1 to 9 candidates per call.
• image_fidelity - 0 to 1, default 0.5. This field is inherited from Kling's shared image-generation schema and is more obviously meaningful on the image-to-image node (where it governs how closely the output tracks a source image); on a pure text-to-image call there's no source image for it to measure against, so don't expect it to change much here - leave it at default unless you notice it doing something specific in your testing.

Outputs: images and bizyair_model_name, a passthrough reporting which backend model version handled the request.

How to install it

ComfyUI Manager: search "BizyAir," install, restart. Manual:

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

You'll need a BizyAir API key configured - Kling is fully closed, so there's no local execution option regardless of how you install the node pack.

Common issues

Not sure why image_fidelity exists on a text-only node. It's a legitimately confusing carryover from the shared Kling image API schema rather than something specific to this endpoint - don't spend time tuning it here the way you would on the image-to-image node, where it has a clear, direct effect.

2K generations costing noticeably more than 1K. Higher resolution requests on hosted APIs almost always cost more per call; if you're iterating on a prompt, generate at 1K first and only bump to 2K once you've settled on the direction you want.

Burning through variants while still dialing in the prompt. Same advice as the image-to-image node: keep variants at 1 while you're still figuring out the prompt, and only raise it once you're confident and just want to pick the best roll.

Job fails with a vague error. Check API key and account credits first - that's the most common cause of an opaque failure on any of this pack's external API nodes, well ahead of anything wrong with your prompt or settings.