Skip to content

Commands

m4bwav edited this page Nov 6, 2026 · 2 revisions

cine.py has seven command groups. The outputs below come from a clone of the v0.2.0 tag (the commands did not change in 0.3.0), run from its root. Every skill ships the same runtime as scripts/cine.py; the repository's scripts/cine.py adds maintainer commands (Development).

usage: cine.py [-h] {kb,cards,compile,continuity,qc,takes,voice} ...
Group Subcommands
kb index, search, show: the knowledge base
cards new, validate, list, graph, export: shot cards
compile card plus bibles to one model's prompt
continuity diff: script supervisor checks
qc sheet, spec, loud, measure, rubric, judge, agree: rendered takes (needs ffmpeg)
takes log, lastframe: take records
voice measure, ref, check: character voices (needs ffmpeg)

cards

python scripts/cine.py cards validate examples/three-shot
cards validate: 3 cards, 0 errors

cards list prints the shot list in cut order (Getting started shows it). cards export --film-json prints the shot list as a film.json.

continuity

python scripts/cine.py continuity diff examples/three-shot
continuity diff: 3 cards, 0 errors, 0 warnings

With the planted card swapped in, it exits 1:

python scripts/cine.py continuity diff examples/three-shot --with examples/three-shot/planted/1C.json
ERROR 1C AXIS: camera is on side B of the 1 axis (between Maren and Tomas, across the lens): it crosses the line, so every left and right flips. Move it to side A, or set crosses_axis with a cross_reason ...
ERROR 1C EYELINE: Tomas looks frame-left at Maren; from side B the eyeline must be frame-right
continuity diff: 3 cards, 2 errors, 0 warnings

compile

python scripts/cine.py compile examples/three-shot --model veo --out compiled/veo
== 1A  veo-3.1-generate-001  6s 16:9 720p  214 words  est. $2.40
== 1B  veo-3.1-generate-001  4s 16:9 720p  163 words  est. $1.60
== 1C  veo-3.1-generate-001  4s 16:9 720p  146 words  est. $1.60
wrote 3 prompts to compiled/veo (model card veo-3-1.md, last_checked 2026-10-03)
est. $5.60 for one take of each (list price on 2026-10-03; a render needs the user's go)

Each prompt starts from the card and pastes the bibles verbatim (1A, first words): "Wide shot, eye-level angle, 24mm lens, static camera. Maren, a lean woman in her sixties with deep-set grey eyes, silver hair in a single braid and a thin white scar through her left eyebrow...". --sequence compiles several cards as one multi-shot prompt for models that take one; --style FILE applies a style; --resolution picks the size.

An unknown model exits 1:

no model card for 'nosuch'. Install cinewright-genvideo, or check `kb search nosuch`.

kb

python scripts/cine.py kb search eyeline
eyelines (cinewright-continuity): Which way each character must look in singles and over-the-shoulder shots so two separately generated shots read as people looking at each other.
screen-direction (cinewright-continuity): The fixed phrases for travel, eyeline and frame position that every card and compiled prompt uses, so direction never drifts between shots.

kb show NAME prints one entry or one section of it.

qc, takes and voice

These read clips you already have through ffmpeg and ffprobe. qc loud final.mp4 --preset web meters integrated loudness and true peak against a preset; qc spec checks fps, size, aspect, length and audio against the card; qc judge asks a local Ollama vision model for a second opinion and qc agree reports how often it agrees with you. Not run for this page: they need rendered media, which the repository does not keep.

Clone this wiki locally