KIE Kling 2.6 (I2V) [SL]

If you've got a still image and you want it to move, this is the node you'll reach for first from this pack. Kling 2.6 image-to-video is the current mainstream pick in the closed-model space - take one frame, add a motion prompt, and get back a short clip that keeps your subject recognizable. It's the simplest of the Kling family here: one image in, one video out, no last-frame gymnastics, no motion-reference requirements.

That simplicity is the point. In the "character consistency → storyboard → rough motion pass" pipeline, this node is the workhorse between "I have a still" and "I have footage to composite." You generate a keyframe locally or via a KIE image node, drop it in, describe the motion, and let Kling 2.6 handle the physics. For most single-shot prompts - a subject turning, hair moving, a camera push-in - it's the lowest-friction way to get there.

The inputs that matter

• prompt - what motion you want, up to 1000 characters. This is the creative input; be specific about the movement, not the scene description.
• images - required, but only the first image in the batch is used. Feed it a single frame.
• duration - 5 or 10 seconds. Longer is pricier and riskier for consistency.
• sound - default off. Flip it on and Kling generates ambient audio to go with the clip. Handy for quick drafts, and it's the feature the T2V sibling shares.

Output is a single video output in ComfyUI's VIDEO format - the node downloads the finished MP4 and hands you a file reference that the official SaveVideo node accepts. Don't forget the SaveVideo; the reference is temporary.

How it works

The standard KIE job pipeline: upload the image as a PNG, create an async task for the kling-2.6/image-to-video model, poll until success, download, convert. Video jobs are slow enough that the node's timeout floor is 1000 seconds, and transient 429/5xx errors auto-retry with backoff up to twice. It all just runs in the background of your graph while you do something else.

Install

ComfyUI Manager, search "Kie-API-Serverless", or:

cd ComfyUI/custom_nodes
git clone https://github.com/v3devv/ComfyUI-Kie-API-Serverless

Restart. No models to download, no extra dependencies beyond what ComfyUI ships. The [SL] suffix means the API key is a node input (serverless-friendly) rather than a config file - and that it travels inside your saved workflow JSON, so scrub keys before sharing.

Where people get burned

The biggest disappointment is usually framing: Kling 2.6 I2V animates within your image, it doesn't reframe or invent new scene content - if you want big camera moves or a wide new shot, you want the motion-control variant or a fresh T2V pass. And every run is a metered video call, which adds up faster than image calls - Kling clips are among the pricier per-call items. For a cheap scout of whether a still is worth animating, consider the 5-second duration and accept that you may re-run.