Kling 4.0's stated limits become the spec for repeatable video review

The note frames the problem as one that only surfaces once a draft has to survive more than one viewing. A clip passed to a creative lead, bounced back with notes and regenerated invites silent drift: duration, resolution, color handling and the number of keyframes or reference stills can all change between runs without anyone deciding they should. None of it shows up in a single clip — it shows up when two outputs have to be diffed, or when a downstream editor assumes a frame rate and bit depth the draft did not have.

Its proposed remedy is to write the constraints before choosing a tool: a fixed maximum duration per draft, a fixed resolution and color depth target, and known upper bounds on keyframes and on reference items, so a revision can be told apart from a prompt change. Traceability is the fifth constraint — each draft has to carry back which prompt, references and keyframe count produced it. To make that concrete, the note shows a JSON-style intake record with fields such as scene_id, prompt_version, target_duration_sec and model_used, paired with a checklist that tests comparability rather than output quality.

Kling 4.0 enters only after those constraints are written. Its stated ceilings — up to 30 seconds at 4K with 10-bit HDR, up to 10 keyframes and 15 reference items — function as the bound a pipeline can design around, so a scene capped at, say, 6 keyframes has headroom, while a scene needing more reference items than the ceiling becomes a pipeline decision rather than a mid-review surprise. The note is explicit that these are vendor-stated capabilities rather than independently benchmarked figures, and that pricing, credit allotments and batch limits are outside its scope and should be re-checked rather than assumed fixed.