Kling Video Extend

Every hosted video model has a hard ceiling on clip length, and Kling's is no exception. The answer isn't a longer generation - it's an extension pass, and this node is that pass. It's a partner/video API node built into ComfyUI core that takes a previously generated Kling video, by its ID, and continues it. No install, nothing runs locally, billed per render through your Comfy account.

The input that defines this node is video_id - a plain STRING input, not a video file. That's the part that confuses people on first contact: you don't drag your finished clip into this node. You wire in the video_id string output from whichever Kling node created it - text-to-video, image-to-video, or a previous extension. The tooltip spells out the whole contract: it supports videos generated by T2V, I2V, and prior extension operations, and the total can't exceed 3 minutes after extension. So the node is meant to sit downstream of other Kling nodes in the same graph, chaining their ID outputs - not to digest arbitrary uploaded footage.

The creative inputs are what you'd expect. prompt guides what happens in the extended segment ("she turns and walks toward the door"), negative_prompt keeps unwanted elements out, and cfg_scale (0–1, default 0.5) tunes how hard the model follows those prompts - lower values let the model stay more consistent with the original footage, which for extensions is usually what you want. Kling's motion-control sibling nodes use the same convention, so if you've touched those, this feels familiar.

Outputs are where Kling shows off: VIDEO (the extended clip), plus video_id and duration strings - and that new video_id is the hook for the next extension pass. You can chain this node to itself: extend, grab the new ID, extend again, until you hit the 3-minute ceiling. That's the intended workflow for building longer pieces out of a 5-second base, one deliberate beat at a time.

The honest take: extension is a slow, iterative way to build a long clip, and each pass costs money, so the economics argue for thinking hard about what you want the continuation to do before you pay for it. The prompts between extensions matter - each one sets up the next segment's motion - and a vague prompt yields a wandering middle. Draft the extension's prompt at low cost per pass rather than assuming one expensive long render will fix everything.

Standard partner-node rules apply: this is a Kling-side API, so moderation and vendor-side failures are possible, and the price badge on the node is your cost indicator. Kling's extend node has been part of the core lineup since the Kling API nodes arrived in 2025, and it remains the most direct way to get a Kling clip past its native length limit.