Comfyui-Kling-Wrapper Effects

Kling's effects are its TikTok brain leaking into the API: pre-baked template videos where your photo gets dropped into a scenario - baseball swing, trampoline bounce, mooncake toss, "rampage_ape." The Effects node is the front door to all of them, and it's the most fun node in this pack. One image in, a choreographed template clip out, no prompting required.

If you're coming from Stable Diffusion, think of it as the opposite of a KSampler: there's no diffusion tuning, no seed search. You pick an effect and Kling handles the rest. That also means it's a one-way street - the effect dictates the motion, and you're mostly along for the ride.

The inputs that matter

• effect_scene - the star of the show. 104 options, from baseball and trampoline to memory_alive and pucker_up. The dropdown is the whole product; scroll it like a menu. There's no documentation of what each does beyond the name, so expect some trial and error on the obscure ones.
• image0 - required, your subject photo.
• image1 - optional second image. Some two-subject effects use it.
• model_name - the effect templates run on the older models: kling-v1, kling-v1-5, or kling-v1-6. No v2/v3 here - the effect API hasn't moved on, and the node shows what the endpoint actually accepts.
• mode - std or pro.
• duration - 5 or 10 seconds. Effects are short-form by design.

Outputs are url and video_id, wired the usual way into Preview Video.

Where people get burned

The biggest practical trap is subject fit. These effects are built for a centered, full or upper-body subject. A wide landscape shot or a tiny face in a crowd will produce a clip where the effect clearly doesn't know what to do with you. Crop to the subject before you feed it in - a quick Image Crop node upstream pays for itself.

Second, this is the "meme" tier of the API. Don't expect cinema. rampage_ape is exactly what it sounds like, and that's fine. Use it for social clips and light entertainment, not for client work you'd stake a reputation on.

Third, budget: each effect run is a full metered generation even though it's template-driven. The 10-second pro versions cost noticeably more, and since you'll want to test several effect_scene options to find the one that lands, the runs add up. Test on std/5s first, lock the winner, then consider upgrading.

Install

Standard pack install: ComfyUI Manager (search "Comfyui-Kling-Wrapper") or git clone https://github.com/magicwang1111/Comfyui-Kling-Wrapper.git into custom_nodes, pip install -r requirements.txt, restart, Kling API key in config.local.json. API node, no local models. (And per the README, effects aren't supported on the Vapeur provider - it fails explicitly rather than pretending.)

Honestly? This is the node you show friends who think ComfyUI is all technical drudgery. It's shallow, it's silly, and it does exactly one thing well - but that one thing is reliably entertaining.