☁️BizyAir Kling 2.1 Image To Video

Kling is Kuaishou's commercial video-generation model - one of the handful of closed, API-only video models (alongside things like Runway and Sora) that regularly show up at the top of image-to-video quality comparisons. There's no local checkpoint to download; it only exists as a paid API. This node is BizyAir's wrapper around Kling 2.1 specifically, and it's the kind of thing you reach for when a local open-weight video model (Wan, LTX, whatever you've got running) isn't hitting the motion quality or coherence you need and you're willing to pay per generation for something sharper.

What it does

You give it a still image and a prompt describing the motion or scene you want, and it comes back with a short video clip animating that image. This is image-to-video specifically - it's not text-to-video, the starting frame is always your input image, and the model's job is filling in plausible motion from there.

Inputs that matter

• first_frame_image - required. The node's own tooltip is explicit that this is literally the first frame of the output video, not just a loose style reference.
• prompt / negative_prompt - standard text guidance for what should and shouldn't happen in the motion.
• model_name - three tiers: kling-v2-1-std, kling-v2-1-pro, or kling-v2-1-master, default std. This is the usual budget-vs-quality ladder you see across commercial video APIs - std is cheapest and fastest, pro and master cost more per generation for (presumably) better motion coherence and detail. Start on std to dial in your prompt, then bump the tier once you're happy with the direction.
• duration - 5 or 10 seconds, default 5.
• aspect_ratio - 16:9, 9:16, or 1:1, default 16:9.
• last_frame_image - optional, and gated: the tooltip states it's only supported on the pro tier. If you're on std or master and wire an image into this input, expect it to be ignored or the run to fail - this is a pro-only feature for controlling both the start and end frame of the clip.

Outputs are video (your generated clip) and bizyair_model_name - an internal identifier mapping each of the three model tiers back to the underlying kling-v2-1 model family, mostly plumbing you won't need to touch.

How to install it

Standard BizyAir install - this node ships with the full pack, no separate download.

ComfyUI Manager: search "BizyAir," Install, restart.

git clone:

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

You'll need a BizyAir API key set before this runs, same as every cloud node in the pack - and specifically for API video nodes like this one, make sure your BizyAir account has enough credits before kicking off a batch. Kling generations are not cheap compute, and that cost gets passed through to your BizyAir balance.

Common issues

last_frame_image silently doesn't apply on std/master. This is the single easiest mistake to make with this node - if you're trying to control both ends of the clip and it's not working, check your model_name first. It's a pro-only input per the tooltip.

Generation time. API video models like Kling aren't instant - expect the node to sit for a while (often a minute or more, longer for the pro/master tiers or 10-second duration) while the request processes server-side. That's normal, not a stuck node - don't kill the run prematurely.

Prompt adherence on motion is looser than on image models. Video-generation prompting is a different skill than image prompting: describing camera movement, pacing, and what happens over time matters more than a dense list of visual adjectives. If motion looks generic, try describing the action more concretely rather than adding more descriptive words.

Cost adds up fast in a batch. Because this is metered API usage rather than local GPU time, looping this node over many images (especially at pro/master and 10-second duration) can burn through BizyAir credits quickly - test on std at 5 seconds before scaling up a batch job.