EvoLink Kling O3 Video Edit (Official)

Cloud video generation has a dirty secret: the first take is rarely the final take, and regenerating a whole clip because one detail is wrong is expensive. Kling O3's video-edit node exists to kill that waste. EvoLink_KlingO3Edit takes a video you already have and re-renders it according to a text instruction - the tooltip's example is "把外套换成红色" ("change the jacket to red") - while keeping the output duration and framing locked to the original. Change an outfit, swap a background, add or remove an object. The expensive regeneration happens, but only for the edit.

The output follows the source clip's duration and aspect ratio, so it slots cleanly into an edit-pass workflow.

The inputs that matter

• prompt (required) - the edit instruction. Be specific and target the thing you want changed; Chinese works.
• video_url - the source video URL. This is the key field: wire it to an upstream EvoLink node's result_urls output to build a generate → edit chain in one graph.
• keep_original_sound - defaults to on, preserving the source video's audio.
• quality - 720p (default) or 1080p.
• reference_image1, reference_image2 - optional reference images to guide the edit (e.g. "make it look like this").

Then the pack-standard api_key and timeout_seconds (1200).

How it works

The job posts to POST /v1/videos/generations as kling-o3-video-edit, and the pack polls until the edited clip is done, downloading it as a VIDEO object - connect video to a native Save Video node. Outputs are the pack standard: video, task_id, status_info, result_urls (24h expiry), response_json. Because the source clip is a URL, this is one of the nodes that makes the pack's "result_urls chaining" philosophy concrete.

Installing

No GPU, no model files, only requests:

cd ComfyUI/custom_nodes
git clone https://github.com/deeplearning-goethe/comfyui-evolink.git

then fully restart ComfyUI, or use ComfyUI Manager and search EvoLink.

Troubleshooting

Pack-standard errors (401/402/403/429) and the VPN note - whitelist api.evolink.ai and files-api.evolink.ai. The video-edit-specific gotcha is the source URL expiring: result links live only 24 hours, so an edit job run on a stale result_urls from an old session can fail with a download error - if that happens, regenerate the source and rewire. And if the edit barely changes anything, your instruction is probably too vague; video edit models respond much better to "change the red jacket to blue" than to "make it different."