Nodes/ComfyUI-Kie-API/KIE Kling 2.6 (T2V)
ComfyUI Node

KIE Kling 2.6 (T2V)

Kling 2.6 text-to-video without leaving ComfyUI

By gateway·Created 9 months ago·Updated 4 months ago· 39
KIE Kling 2.6 (T2V)
    • video
    prompt
    soundfalse
    aspect_ratio9:16
    duration5
    logtrue

    Most video nodes are image-to-video: they need a still to animate. Kling 2.6 (T2V) is the exception in this pack - you give it nothing but text and it gives you a short clip. It's Kuaishou's Kling 2.6 running through the Kie.ai API, so there's no local model, no GPU load, just a prompt, your API key, and some credits.

    When would you actually use this? When you don't have a first frame worth keeping. Maybe you're sketching a mood, testing a setting before committing to an I2V pipeline, or you want a throwaway clip in a storyboard. If you already have a good image, the Kling 2.6 I2V node in the same pack will serve you better - text-to-video has less to hold onto, and it shows.

    How it works

    The mechanism is about as simple as API video gets: the node posts your prompt to the KIE createTask endpoint, polls the job until the mp4 is ready, and returns it as a ComfyUI VIDEO. The default async timeout is 2000 seconds because Kling 2.6 jobs can legitimately sit in the queue for a couple of minutes, especially at 10 seconds with sound on.

    Inputs worth touching

    • prompt - the only required input, and it's multiline. Kling rewards reasonably detailed scene description, same as its I2V sibling.
    • sound - default off. Flip it on if you want audio baked into the clip; it generally costs more and can push render time up.
    • aspect_ratio - defaults to 9:16 (portrait, the social-first default), with 1:1 and 16:9 available.
    • duration - 5s or 10s. Kling's T2V isn't a 15-second beast; keep expectations modest.
    • log - console logging, on by default. Leave it until something breaks.

    There's no image input on this node at all. If you wire an image into it expecting I2V behavior, it won't do what you want - you've got the wrong node for that.

    Output: a single video (VIDEO type) that plugs straight into SaveVideo or any ComfyUI video consumer.

    Install and key setup

    Same story as every node in this pack: it's an API client, so there are no checkpoints to download and no GPU requirements. Install from ComfyUI Manager (search "ComfyUI-Kie-API") or:

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

    Restart, then create config/kie_key.txt from the example file and paste in your Kie.ai key. A quick sanity run of the pack's "Get Remaining Credits" node will tell you the key is valid before you spend anything.

    Common issues

    • "Task completed without resultJson." The job finished on KIE's side but returned no usable result - usually a prompt the model refused or a credit/balance hiccup. Check https://kie.ai/logs for the actual record.
    • Silent hangs. If the node sits there past a few minutes, it's not necessarily broken - 2000s timeouts exist for a reason. Watch the console log (with log on) to see the poll progress.
    • Audio is slower/absent. With sound off you get silent clips by design; with it on, expect a longer wait. If the output has no audio despite sound=true, check the job at kie.ai/logs - it can be a KIE-side quirk rather than something your workflow did.

    For a beginner this is the friendliest video node in the pack: one prompt, one wire out, done.

    Categorykie/api

    Inputs (5)

    NameTypeDefaultDescription
    promptSTRING
    soundoptBOOLEANfalse
    aspect_ratiooptCOMBO9:163 options: 1:1, 16:9, 9:16
    durationoptCOMBO52 options: 5, 10
    logoptBOOLEANtrue

    Outputs (1)

    NameTypeDescription
    videoVIDEO