Kling Video Effects

The gimmick node that's secretly the most fun: Kling's effect templates turn two still images into a short animated clip with a preset interaction. Feed it a portrait and get a video of the person in a heart-filled scene; feed it two portraits and Kling animates them into a hug. It's the "put your face on a dancing character" button of the Kling service - pure novelty, zero camera work, and a surprisingly good demonstration of what the model can do with motion that's specified by template rather than prompt.

The inputs are mostly self-explanatory:

• image_1 (required) and image_2 (optional) - your subject(s). One image gets the effect applied to it; two enables two-subject effects like hug and kiss.
• effect_scene - the effect name, default hug. This is a free-text string, and the tooltip explicitly points you at the pack's Effect Templates node to see what's actually available. Don't trust the default list; the API's catalog changes.
• model_name - kling-v1 or kling-v1-5 (effect templates live on the v1 ladder, not v3).
• duration - 5 or 10 seconds.
• mode - pro or std, defaulting to std. For effect videos, std is a reasonable start - the template does a lot of the quality work already.

Mechanically it's a submit-and-poll cloud task like every generation node in the pack. Outputs are the full video set: video, video_file, audio, url, task_id. Note the audio output - Kling's effect renders can come back with sound, so don't be surprised when the hug video has a soundtrack.

Installing it

Part of ComfyUI-API-Toolkit. Manager: search "API Toolkit". Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/IxMxAMAR/ComfyUI-API-Toolkit
pip install -r requirements.txt

Restart ComfyUI. Needs the Kling AI Authentication node and credits.

Gotchas

The effect_scene string is the whole game, and guessing it is how you waste credits: one queue = one billed generation, and a scene name Kling doesn't recognize is a wasted run. The Effect Templates node exists precisely so you don't guess - run it first, read the JSON, copy the exact scene key. Also keep expectations realistic on likeness: template effects prioritize the interaction over identity, so the hugging figures may look like loose interpretations of your inputs rather than faithful portraits. And the content-policy wall (error 1302) is strictest here - if a template touches flagged content, Kling refuses at the source and there's no bypass. For the novelty value alone it's worth having, just don't build your product on it.