Effects

If you've seen those one-click AI video trends - someone's photo suddenly doing a backflip on a trampoline, morphing into a mooncake, getting launched into a baseball swing - this node is the ComfyUI door into that exact category, built on KLing's own effect templates. The README describes it plainly: "you can achieve different special effects based on the effect_scene." You're not directing a custom animation here; you're picking one of KLing's pre-built scenes and feeding it your photo(s).

How it works

You supply one or two images and pick an effect_scene from a long preset list, and KLing's servers apply that specific, pre-trained effect to your subject - placing them into the scripted motion, environment, or transformation the template defines. It's the same idea as Image2Video, except instead of writing a prompt describing motion, you're selecting from a menu of effects KLing has already built and tuned. That trade-off is the whole point of this node: much less creative control than a prompt-driven video node, much more reliable, consistent output for the specific effect you picked.

The inputs and outputs that matter

• effect_scene - the big one: an enum with over 100 choices (baseball, trampoline, trampoline_night, pucker_up, feed_mooncake, rampage_ape, flyer, dishwasher, and dozens more). Some are clearly single-subject effects, others - like anything implying two people interacting - expect the second image1 slot to be filled.
• image0 (required) - your primary subject photo.
• image1 (optional) - a second subject, for effects built around two people.
• model_name - kling-v1, kling-v1-5, or kling-v1-6. Note this is a narrower, older set of versions than Image2Video's model list - effects are tied to whichever model generation they were originally built for.
• mode (std/pro) and duration (5/10) - same meaning as elsewhere in the pack: pro for higher fidelity at higher cost, duration in seconds.

Outputs are url (the finished clip) and video_id, which chains onward into Video-Extender, Video2Audio, or Lip-Sync just like any other KLing video output in this pack.

Installing it

ComfyUI Manager: search ComfyUI-KLingAI-API, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/KwaiVGI/ComfyUI-KLingAI-API
cd ComfyUI-KLingAI-API && pip install -r requirements.txt

No local model weights to fetch - a Client node with a valid KLing AI API key upstream is what actually makes this run.

Common issues

The most common trip-up is a mismatch between the effect_scene you picked and how many images you've supplied - a two-person effect fed only image0 is likely to either fail outright or produce something that doesn't match what you expected, since the template assumes a second subject exists. Beyond that, subject readability matters more here than in freeform generation: these templates were built and tuned against clear, well-composed reference photos, so a cropped, low-light, or heavily obscured subject photo tends to produce worse results than it would with a prompt-driven node that can compensate with text. And as with the rest of the pack, the usual Client failure modes apply underneath all of this - a bad key, a region mismatch on area, or an exhausted KLing account balance will fail this node the same way they'd fail any other.