zhenzhen-kling-o3-edit-lowprice

Kling's edit models are the "re-shoot this footage" tier of the Kling line - you hand them an existing clip and an instruction, and they re-render it to match. This node wraps the o3 edit variants (kling-o3-std-edit and kling-o3-pro-edit) on Zhenzhen's cheap domestic API, and it's the natural bookend to the big Kling generation node: generate → edit, both in the same graph, both billed per run.

The name is the honest summary: edit. This is not image-to-video from a still and it's not a text-to-video from scratch - the video_url is a required input (you can also wire the input_video VIDEO tensor, but the URL field is the schema's required one). The prompt tells Kling what to change: restyle the lighting, swap the weather, change the motion, erase an object and clean up what's left.

How it works

Standard lowprice mechanics: an api_config input (ZHENZHEN_SEEDANCE2_CONFIG type) from the pack's settings node carries the base URL and key. You give it the source video (as a video_url string, or an input_video tensor if you have one in the graph), the edit instruction in prompt, and seconds (5 or 10) for the output length. It submits, polls, downloads, and returns video (a real VIDEO output you can save), plus video_url, task_id and response.

One required-input trap to flag: video_url is in the required block, so the node will complain if it's empty - even if you've connected input_video. Fill the URL or wire the tensor; whichever you use, the field needs something.

Inputs that matter

• video_url - the source clip (required). For files you already have locally, wire input_video instead and put a placeholder URL in the field.
• prompt - the edit instruction. Specific beats vague: "make it night with neon reflections" beats "make it cooler."
• model - std vs pro. Pro is the better pass; std is the cheap iteration.
• seconds - 5 or 10. Editing doesn't extend infinitely; pick the output length you want.
• api_config - from the settings node; without it nothing connects.

Installing it

Manager → search Comfyui-zhenzhen → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Restart, add the lowprice settings node with your key, wire its api_config in. No model downloads - the render happens on Kling's side.

Common issues

• "Required input is missing" on video_url - even with input_video wired, the URL field is required. Put a value in it (the tensor wins when both are present).
• Edits that don't match the prompt - video-to-video editing is content-aware, and Kling's interpretation leans on the prompt's clarity. Describe the change, not the whole scene.
• Seconds vs source length - the output is the length you ask for; if your source is longer, the edit covers the window it can. Trim upstream if you need precision.
• The pack standard gotchas - empty key in the settings node, domestic-site reachability, vendor 500s needing a re-run, cache-only seed.

The sibling node (Comfly_kling_video_lowprice) is where you go to create Kling video; this is where you go to change it. Two different jobs, two different models, and - worth knowing - two different bills.