Introduction

Kling 4.0 has been out in the wild for a few weeks, but if you live inside a node graph you've been stuck waiting. That wait just ended. Late on October 26 (UTC), Comfy-Org merged pull request #17364 into ComfyUI, adding a full set of Kling 4.0 Partner Nodes: Text to Video, Image to Video, Reference to Video and Edit Video.

These are API nodes. Kling runs in the cloud and your Comfy account pays per generation. You get Kling 4.0 right next to your local models, with no separate dashboard.

What Shipped

The PR adds four nodes under partner/video/Kling. Each one has a single model option, kling-4.0:

  • Kling 4.0 Text to Video (Kling4TextToVideoNode): "Generate a video with audio from a text prompt." Aspect ratios are 16:9, 9:16, 1:1 and 21:9.
  • Kling 4.0 Image to Video (Kling4ImageToVideoNode): video with audio driven by up to 10 keyframes, made of an optional first frame, up to 8 middle frames and an optional last frame. Middle frames are spaced evenly, and the clip takes the aspect ratio of the first keyframe.
  • Kling 4.0 Reference to Video (Kling4ReferenceToVideoNode): video with audio from up to 10 reference images and up to 5 reference videos. You can also pin a first or last frame. Its aspect ratio list adds adaptive, which follows the prompt, then the first frame or the reference videos, and otherwise falls back to 16:9.
  • Kling 4.0 Edit Video (Kling4EditVideoNode): edits an input clip with a text prompt plus optional reference images and videos. The result keeps the length of the input video.

The Knobs

The same controls show up across all four nodes:

  • Resolution: 720p or 1080p
  • Duration: 3 to 30 seconds, default 5, on a slider (Edit Video skips this and matches your input length)
  • Prompt length: up to 26,000 characters
  • Seed: included, but the tooltip says results are non-deterministic regardless of seed
  • Prompt tagging: refer to inputs as @Image1, @Image2, @Video1, and so on. The node checks that every tag points at an input that is actually connected.

The validators set some limits on your inputs:

  • Images must be at least 300×300 with an aspect ratio between 2:5 and 5:2, and the cap is 10 images per job (pinned first/last frames count toward it on Reference to Video)
  • Reference videos must be 3 to 30 seconds each, 30 seconds combined at most, at least 700 px on each side, no more than 3840×2160 worth of pixels, and at least 24 fps

What It Costs

Each node shows a live price badge. These are the rates in the merged code:

  • Text to Video and Image to Video: about $0.22451 per second at 720p, or $0.33605 per second at 1080p. A default 5-second 720p clip comes to about $1.12.
  • Reference to Video, images only: the same per-second rates as above.
  • Reference to Video with reference videos: a range based on $0.13442/s (720p) or $0.20163/s (1080p), from 2× your output duration up to 30 seconds plus your output duration.
  • Edit Video: priced per second of input video. That's $0.26884/s at 720p or $0.40326/s at 1080p with no reference videos. With reference videos it's a range of $0.13442 to $0.26884/s (720p) or $0.20163 to $0.40326/s (1080p).

Where to Run It Today

The nodes are merged on ComfyUI's master branch. As of this writing, the latest tagged release is still v0.42.1 from October 19, so update to the latest ComfyUI (nightly/master) or wait for the next release. As with other Partner Nodes, sign in with your Comfy account so API credits can pay for each job.

A first test that plays to Kling 4.0's strengths: put three keyframes into Image to Video (a wide establishing shot, a mid shot and a close-up) and give it one prompt for camera movement. You get a 10-second piece with audio and no stitching.

Conclusion

Kling 4.0 in ComfyUI means you can now chain it with your local upscalers, LoRA'd stills and audio passes inside one graph. Ten keyframes, five reference videos and 30-second clips is a lot to play with, so price out your 1080p runs before you get carried away. Happy directing, friends!

Original Source

—Aurelia ♡