Kling 3.0 Omni Video to Video
Restyle footage and keep the sound
- reference_video
- reference_images
- VIDEO
The Kling 3.0 Omni Video to Video node is what you reach for when you already have footage and want it reimagined. It takes an existing video, optionally a few reference images to lock in style or subjects, and regenerates it under a new prompt - new look, new vibe, same source material. For restyling a clip to match a brand, swapping a scene's mood, or iterating on a shot you like but don't want to re-shoot, this is the Kling node that does it.
Like every node in this family, the work happens on Kling's servers via your Comfy account. Your reference video and images get uploaded, the generation runs in the cloud, and a fresh VIDEO comes back when the task completes. It's billed per generation to Comfy account credits, with the price badge on the node scaling with duration and resolution.
The inputs that matter:
- reference_video - the footage to transform. Kling validates it hard: it must be 3 to 10 seconds, between 720×720 and 2160×2160. Short and square-ish works best.
- prompt - describes what the output should be. Like the other Omni nodes, you can fold negative descriptions straight in. Up to 2,500 characters.
- keep_original_sound - default on. When it's on, the original audio rides along; turn it off to get a silent re-render.
- reference_images - optional, and capped at 4. These add character or style anchors on top of the video. Each one needs to be at least 300×300px with an aspect ratio between 1:2.5 and 2.5:1. Connect a batch of images if you want several.
- model_name -
kling-v3-omniorkling-video-o1. - aspect_ratio - 16:9, 9:16, or 1:1.
- duration - 3 to 10 seconds.
- resolution - 1080p (maps to pro mode) or 720p (std mode).
- seed - the usual stub; it just controls whether the node re-runs, output stays non-deterministic.
Single VIDEO output into Save Video / Preview Video.
The trap here is the reference image cap plus the validation. It's easy to wire in a fifth image and get an immediate error, and easy to feed a 20-second clip and hit the 10-second ceiling. Also remember that the video and images are uploaded, so larger sources mean slower starts - keep clips lean. Oh, and keep_original_sound is a bill multiplier in its own way: on is the default, and if you're re-rolling a lot of restyle attempts, that's credits adding up.
Partner-node house rules apply: be signed in to your Comfy account or it errors out, and if the API-node family is cluttering your node search, --disable-api-nodes gets rid of them.
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| model_name | COMBO | 2 options: kling-v3-omni, kling-video-o1 | |
| prompt | STRING | A text prompt describing the video content. This can include both positive and negative descriptions. | |
| aspect_ratio | COMBO | 3 options: 16:9, 9:16, 1:1 | |
| duration | INT | 33–10 | — |
| reference_video | VIDEO | Video to use as a reference. | |
| keep_original_sound | BOOLEAN | true | — |
| reference_imagesopt | IMAGE | Up to 4 additional reference images. | |
| resolutionopt | COMBO | 2 options: 1080p, 720p | |
| seedopt | INT | 00–2147483647 | Seed controls whether the node should re-run; results are non-deterministic regardless of seed. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| VIDEO | VIDEO | — |