Kling 4.0 video edit: an "Edit @Video1" prompt with a keep-list in the official ComfyUI template

What this is: a reading of one workflow file from the official Comfy-Org workflow_templates repository, pinned at commit 0cc0fba5 (Kling 4.0: Video Edit). Only the JSON was inspected. No workflow was run, no Kling API call was made and none of the example media was opened, so nothing here describes an actual output. Declared graph (4 nodes, 2 links): - LoadVideo (node 1) feeds model.video_1 of Kling4EditVideoNode (node 2, link 1). - model.reference_videos.video_2 and model.reference_images.image_1 are present on node 2 but unconnected. - Node 2 sends its VIDEO output to SaveVideo (node 3, link 2), format mp4, codecs auto. Stored settings on node 2 (positional list; the named copy agrees for this node): model kling-4.0, resolution 720p, aspect ratio adaptive, seed 42, and both control_after_generate entries set to fixed. Unlike the Kling 4.0 reference-to-video and Genjutsu templates, the seed is not set to randomize. There is no duration setting; the note says the output follows the input. The prompt has the pattern "Edit @Video1:" plus one change plus a keep-list: - The change: replace the frozen lake and overcast sky with a rainy neon city street at night (wet asphalt reflecting pink and cyan signage, rain streaks, low mist). - The keep-list: the woman, her black tiered gown, wide black hat, gloves and jewellery unchanged; her movements, the three shots and the camera work exactly as they are; the same rim-light direction so she still reads as a silhouette. - Sound: rain on asphalt and distant traffic replace the wind. A mismatch inside the file: the positional values load kling4_gown_input.mp4 and save with prefix video/kling4_video_edit, while the widgets_values_named copy says kling4_gown_edit_input.mp4 and video/kling4_edit/kling4_gown_rain. The note's input asset entry names the positional file. The file does not say which copy a loader uses. What the template's note says (author statements, not verified here): the prompt must open with "Edit @Video1:" and everything after the change is the keep-list. Listing what must survive (subject, costume, movement, shot count, camera work, light direction) stops the model from redesigning the person or re-staging the camera. The output inherits the input's length, aspect ratio (adaptive) and three shots. Input clips must be 3-30 s, and the example input is 5 s. Extra reference images or videos can be connected when the new element must match something specific. Not established: what the edited output looks like; whether the keep-list is honoured; whether audio is generated or replaced; whether a fixed seed reproduces a result with this partner model; or the cost of a run. Example media terms and Kling service terms are not covered. Related topics: prompt-driven video editing with an explicit keep-list, reference inputs for edits, and reading ComfyUI workflow JSON when positional and named widget values differ.

{
  "citations": [
    {
      "url": "https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/templates/api_kling4_video_edit.json",
      "note": "Pinned original; exact hashes and byte ranges retained in source pins."
    },
    {
      "url": "https://raw.githubusercontent.com/Comfy-Org/workflow_templates/0cc0fba562efe9f68e48050c2b82ed6495050bc2/LICENSE",
      "note": "Pinned original; exact hashes and byte ranges retained in source pins."
    }
  ]
}