Use Kling 4.0 video editing when you want to modify existing footage while asking the model to retain its action, timing, framing, or subject. The request should identify what changes and what stays. A reference video used to inspire new generation is a different workflow from a source video being edited.
This guide uses PoYo's Kling 4.0 API and a saved, completed PoYo example: a red fox walking through snow, edited into an autumn woodland. This is a scene-change example, not evidence that every product swap or identity-preserving edit will succeed.
Watch the source and edited result
Source: winter wildlife shot. Recorded task 9DRXDJ5XGW41B9IW, requested five seconds at 720p, generated from a first frame. Recorded charge: 142.5 credits.
Edited result: autumn woodland. Recorded task 84PEDOWZ40C5D7VM, requested five seconds at 720p, using the source video and an autumn reference image. Recorded charge: 187 credits.
Download the exact editing request and sample metadata. Compare the original and edited clips directly: the fox's markings, gait, head movement, ground contact, camera track, and foliage boundaries all matter. A changed background alone is not enough if the action you needed has drifted.
Choose editing explicitly with video_url
The following shortened template demonstrates the fields; the downloadable file contains the recorded task's full prompt.
{
"model": "kling-4.0",
"input": {
"video_url": "https://storage.poyo.ai/blog/kling-4-0-series-20261024/fox-video-d0100387a0e0.mp4",
"reference_image_urls": ["https://storage.poyo.ai/blog/kling-4-0-series-20261024/autumn-reference-0773c88cbeb3.png"],
"prompt": "Edit @Video1 into the autumn woodland shown in @Image1. Replace snow with dry fallen leaves and distant conifers with silver birches. Preserve the same fox, its markings, proportions, walking and head movements, timing, framing and camera tracking. Match shadows and light to the new environment. Return one coherent live-action scene, not a montage or transition.",
"duration": 5,
"resolution": "720p"
}
}
The source is addressed as @Video1; the image reference is @Image1. Images specify the new appearance while the source supplies the motion to preserve. video_url selects editing ahead of generation modes, so do not add it to a request that is supposed to generate a new shot from a motion reference.
What the PoYo edit schema accepts
The source must be 3–30 real seconds. Editing requires prompt and supports 720p or 1080p. It can use reference images and image-defined subjects, but rejects image_urls, reference_video_urls, video-defined subjects, multi_prompt, and aspect_ratio.
Do not send unsupported fields with null or empty values in an attempt to disable them. Omit them. PoYo's current editing interface accepts one source-video field; the broader creator product's multi-video workflow is not a reason to invent a video_urls array here. Verify the current API documentation for updates.
How to write a constrained edit prompt
Separate the instruction into two clear statements. First, name the desired change: replace the season, change a jacket, or replace a product. Second, name the retained structure: subject identity, action timing, camera movement, framing, and untouched objects.
For product replacement, provide a clean reference with a clear silhouette and write which object should be replaced. Check contact with hands and surfaces, moving occlusions, shadows, and reflections. The model may change more than you requested, so compare the entire shot and verify brand details before approval. This article's fox example does not demonstrate a measured product-replacement success rate.
Avoid changing the character, background, outfit, camera and dialogue in the same first test. If an edit fails, a smaller change makes the next diagnosis more useful.
Estimate the cost from source and output duration
Editing uses video-input pricing. At the current PoYo list rate, 720p is 17 credits per billed second, and 1080p is 25.5. Billable seconds are output + max(ceil(source duration), output).
For a source of exactly five seconds and a five-second requested output, the estimate is 10 billed seconds: 170 credits / $0.85 at 720p. Real files can run slightly longer than their requested generation duration. The saved autumn task's recorded 187-credit charge corresponds to 11 billed seconds; use server-probed source duration and task records rather than assuming every “five-second” file is exactly five seconds.
When duration is omitted, PoYo uses the ceiling of source duration for output billing while leaving duration omitted upstream. See the pricing guide for fractional-duration examples.
Review before publishing
Watch the two clips at matching points, then at normal speed. Check object identity, motion continuity, edge flicker, texture stability, and any unintended content. Listen to the complete soundtrack as a separate check; preserved visuals do not establish preserved audio.
For fresh generation with references, use the Omni Reference guide. For submission and polling, use the API tutorial. The official creator guide offers additional creative workflows, subject to platform availability.