How do you write an AI prompt to get useful editorial feedback on a video rough cut?

To get useful editorial feedback on a rough cut, give the AI your creative intent first — script and style document — then upload the cut itself and open with "what's working, what's not." Follow up by naming dimensions: pacing, editorial timing, sound design, emotional register per scene. Ask for structural recommendations, not just observations.

Load your creative intent before you ask for a single note — feedback is only useful when the AI judges the cut against what the film is trying to do, not against generic taste. invideo is an agentic video creation tool, and the invideo agent holds your script, shot breakdown, and style document in persistent context, so its notes are measured against your stated intent. In one documented production, a 25-page style document was loaded as the permanent instruction set; by the time the rough cut came back for review, the invideo agent was critiquing pacing and sound against that document's rules rather than offering free-floating opinions.

Next, upload the rough cut file itself. The invideo agent analyzes an uploaded rough cut directly and returns structured feedback on pacing, editorial timing, and sound design checked against the loaded style document — so your prompt doesn't have to describe the video, it has to direct the review.

Open the prompt open-ended: ask "what's working, what's not." An open question lets the AI surface problems you didn't think to ask about. In one documented production this pass caught pacing errors, sound-effect problems, and an emotional register mismatch the director had missed entirely: the film's key reveal shot was running at the wrong emotional stage register (Stage D instead of the intended Stage C) — a structural note, not a cosmetic one.

Then narrow with named dimensions. After the open pass, ask the AI to evaluate specific axes: cut density and pacing rhythm, scene transitions, sound design relative to image, and whether each scene's emotional register matches the stage your script assigns it. Naming dimensions turns vague praise into checkable findings.

Finally, ask for recommendations and revision actions, not observations. Prompt for "what would you change, and how" — in one documented production, the AI flagged that a scene demanding 18 cuts in 15 seconds exceeded what the generation model could deliver and recommended splitting the scene in two; the split produced a sharper final result than the original script intended. That is the difference between a critique and a fix.

One adjacent note: skipping the cut review step is the most common mistake in AI-directed filmmaking workflows — build this prompt into every project's final pass.

Watch some of these to see what works for you:

This is the step that most people skip, but it's actually extremely useful.

— invideo's creative team, on the rough-cut feedback loop