Edit video with an AI prompt — type the change, keep the shot

The phrase has two meanings. In transcript editors, it means cutting a video by deleting words from its transcript. In generative tools like Genjutsu, it means changing what's in the frame by describing the edit in text. Genjutsu does the second: it re-renders the clip with your change and keeps the motion, camera and timing — it doesn't cut or rearrange footage.

Almost anything you can see: the character, an object, an outfit, the location, the weather, the season or time of day, or several of these at once. Free edit is the open mode for this; the other four modes — Replace character, Swap object, Change outfit and Change background — are presets for the common cases, with a built-in instruction behind each.

Each reference image you add is numbered in order, so the first is @image1, the second @image2. Writing "the character from @image1" tells the engine which picture to copy. Mentions are converted into "reference image 1" before the engine reads them, and a mention past the end of your list is dropped instead of guessed.

Up to 1,500 characters. Short, precise prompts usually work best: a verb, the target and the new look, plus what to keep if it's at risk. On Kenerate Genjutsu and Genjutsu Pro, the Enhance prompt toggle can expand a short instruction into a fuller one for you; Seedance 2.5 reads your words as written.

No. Genjutsu changes what appears in the frame and returns the clip at its own length; it doesn't trim, cut, reorder or add captions. For captions, use the Add subtitles hand-off after the result. For assembling several clips, use a timeline editor and bring each finished clip into it.

Usually because the target was ambiguous or the change implied more — "make it night" affects the light on everyone. Name the target precisely, add "keep the faces and the outfits the same" when they matter, and split big changes into two runs. Genjutsu Pro and Seedance 2.5 are the tiers built for cleaner, more faithful results.

Only footage you own or are licensed to use, and only with the consent of the people in it. Don't prompt real people into situations that never happened, don't impersonate anyone and never use celebrities. Label AI-edited clips when you share them, especially when they could be mistaken for real events.