Kling Video Effects
Turn two stills into a hugging, kissing, heart-floating video
- auth
- image_1
- image_2
- video
- video_file
- audio
- url
- task_id
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) andimage_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, defaulthug. 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-v1orkling-v1-5(effect templates live on the v1 ladder, not v3).duration- 5 or 10 seconds.mode-proorstd, defaulting tostd. 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.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| auth | KLING_AUTH | — | |
| image_1 | IMAGE | — | |
| effect_scene | STRING | hug | Effect type (e.g., hug, kiss, heart). Use Effect Templates node to see available options. |
| model_name | COMBO | kling-v1 | Model for video effects. |
| duration | COMBO | 5 | Effect video duration in seconds. |
| mode | COMBO | std | Generation mode. |
| image_2opt | IMAGE | — |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| video | IMAGE | — |
| video_file | STRING | — |
| audio | AUDIO | — |
| url | STRING | — |
| task_id | STRING | — |