# Swap prompt guide (short)

A window's prompt has six sections, in this order, each header on its own line: `subject_definitions:`,
`summary:`, `retention_analysis:`, `detailed_description:`, `overall_soundscape:`, `non_diegetic_music:`.
English, plain ASCII punctuation, present tense. Describe the TARGET video as it plays; never write edit
instructions ("replace", "change", "make").

## subject_definitions
- `<Video 1> is the source video for the editing task: <one line on what it shows>.`
- One line placing every picture: `<Picture 1> and <Picture 2> are frames of the target video, <Picture 1> from
  [Shot 1] and <Picture 2> from [Shot 3].`
- One `<Subject N>` per replacement character the brief lists - one each, never one subject for a group. Bind it to
  the pictures that show it and say whom it replaces: `<Subject 1> is <name>, <hair, face, clothes as painted>, in
  <Picture 1> and <Picture 2>. He replaces <the source person, as the brief describes him> in <Video 1>.` The
  description must tell the person apart from everyone else in frame. Name only the characters the brief lists.

## summary
`[video editing + character replacement] The target video is an edited version of <Video 1> in which <source
person> becomes <Subject 1>, ... All other people, the place and the soundtrack are unchanged; the shots, cuts,
camera, motion and timing of <Video 1> are kept.`

## retention_analysis
- `<Video 1> (source video editing): fully_preserved - the <n> shots and the cuts at <MM:SS.mmm, ...>, the camera
  framing and movement, the background, the unchanged people, and the motion and timing of every performance.`
- Each picture: `<Picture N> (appears in [Shot K]): reference - a frame of the target video. The people in it, their
  faces and clothing, and the background, framing and lighting are taken from <Picture N>.`
- Each subject: `<Subject N> (appears in [Shot K], ...): fully_preserved - <its look> stays identical in every frame.`

## detailed_description
- First one sentence on the look: realism, lighting, colour, camera style.
- `[Shot 1]` has no time; every later shot of the brief opens `[Shot N] At MM:SS.mmm, the shot cuts to ...`, in
  window time. One `[Shot N]` per shot the brief lists, stubs of a few frames included.
- Per shot, 2-4 sentences: framing and camera move; who is where - replaced people as `<Subject N>`, everyone else
  by their look; what they do; the `<Picture N>` it matches.
- Talking or singing: `<Subject 1> (S1) sings <d>[unclear]</d>`. Never write or guess words.
- Burned-in subtitles and on-screen text: quote them verbatim in the shot that shows them.
- An action at a given moment (eyes, mouth, a gesture, an entrance) gets its time: "only at about 00:03.200".
- A replacement character no picture shows: describe the whole person from the brief's text where they appear.
- Never write "dim" or "dimly lit": it darkens the whole render.

## overall_soundscape / non_diegetic_music
One line each: the ambience and physical sounds; the music the audience hears, or `N/A`.