proofcut

# proofcut

**An AI video editor that proves its cuts.** Your recordings in, a finished,
mastered film out: cut by transcript, with b-roll, cards, music and captions,
and every step an agent can call. proofcut renders on your own machine, then
transcribes the render and checks that it says what the edit says.

https://github.com/user-attachments/assets/4153d180-3d7c-4c70-af5f-54d63d0a8bd5

Above: an agent cutting a demo video, unattended. The two runs it was cut
from, uncut (silent: the recorder took frames only, and the voice is in the
film the agent cut): [the workspace](https://github.com/tydude001/proofcut/releases/download/v0.23.0/proofcut-v0.23.0-uncut-workspace-run.mp4)
(2:26) and [Claude Code with the proofcut plugin](https://github.com/tydude001/proofcut/releases/download/v0.23.0/proofcut-v0.23.0-uncut-claude-code-run.mp4)
(3:09).

That is not a staged run. [TRIAL.md](docs/TRIAL.md) scores three unattended
ones, each handed a goal and no steps, and each passed every one of its checks.
- **The demo cut**, the one above.
- **Real footage:** 96 seconds of narration with its fluffed takes left in,
  plus four clips of film footage. The agent cut it to 45 seconds, chose
  footage by what each line was about, burned captions, and checked its own
  render: all 123 expected words heard back.
- **A whole film:** the demo material plus a score, briefed as a finished film
  ready to upload. In 192 seconds and $2.27 the agent cut it, laid the music
  under the voice, ended on a card, mastered it to −16 LUFS, and checked it.
  The score, the level and the end card were each measured in the delivered
  file, not taken from the project.

> **Have a Mac or a Windows PC and half an hour?** Nobody has run proofcut on
> a Mac yet, and on Windows only its author has. One script runs the whole
> test and removes what it installed:
> [Mac](#help-wanted-the-first-run-on-a-mac) or [Windows](#help-wanted-a-windows-run-by-someone-else).

## From recordings to a finished film

One project, and proofcut's own commands from the first import to the
delivered file. No NLE finishes the film, and nothing else touches the render.
Each stage is one command. The [manual](docs/MANUAL.md) walks every one, and
its [§ Music, holds, a cold open and the master](docs/MANUAL.md#music-holds-a-cold-open-and-the-master)
covers the sound:

| Stage | Command |
|---|---|
| Bring in the voiceover and footage, and transcribe | `import`, `transcribe` |
| Cut retakes and asides by naming their words | `cut vo 111:114` |
| Hang b-roll and cards off the lines they belong to | `cue add`, `card new` |
| Open cold on a scene, end on a card | `head`, `tail` |
| Play the footage's own lines in a gap, or under the narration | `hold add`, `hold under` |
| Score it: placed passages, crossfaded, levelled under the voice | `music` |
| Pull breaths down without cutting them | `attenuate` |
| Render and master to a loudness target | `export --render --loudness -16` |
| Burn in captions | `captions --burn` |
| Check the render says what the edit says | `verify`, `frames`, `hold check` |

Two video essays of five to six minutes have been rebuilt this way and
measured against their originals, which had been finished outside proofcut.
One came out the same length to the frame. The other matched its original's
63 voiceover ranges to the millisecond, with the voice aligned to the sample,
starting from a retake pass made in Kdenlive. Both master at the original's
−16 LUFS (HISTORY.md § The Lambs/Longlegs native rebuild, § The Scream native
rebuild).

proofcut makes no footage and writes no script. It takes what you recorded to
a film, and proves the film matches the edit.

## Why proofcut

- **It checks its own work.** `verify` transcribes the finished file and
  diffs it word by word against the timeline, so a retake left in the film is
  caught before anyone watches it. Frame counts and the picture are measured
  too, because ffmpeg, melt and auto-editor all exit 0 on some failures.
- **Cuts stay addressable.** Every word in a recording keeps a fixed index
  that never renumbers, so `cut vo 111:114` names the same words however many
  cuts came before it. `--plan` prints what a range says before anything
  changes, and `restore` and `undo` walk it back.
- **Built for agents.** 90 MCP tools with typed inputs and structured
  returns, so Claude Code, Codex or your own agent can drive it. Tools like
  `shot-sheet` return an image of the edit, not a file path the agent can't
  open.
- **One engine, three ways in.** The MCP server, the `proofcut` command line
  and a browser workspace all call the same operations. Every tool has a
  matching command (the test suite enforces it), so anything an agent does,
  you can script or re-run by hand.
- **Local-first.** Commercial AI editors are apps around a metered cloud
  service. proofcut transcribes, edits and renders on your machine, and calls
  no cloud service of its own: no account, no per-minute billing. The only
  thing that talks to a model provider is the agent you choose to run (the
  optional footage-description and voice models download once, on first use).
- **No lock-in.** The timeline is OpenTimelineIO. Export a `.kdenlive` or
  OTIO file, finish in Resolve, Premiere or Kdenlive, and bring your trim back
  with `import-edit`.

![The proofcut workspace on the demo project: the transcript with a retake struck
through, the preview drawing the shot under the playhead with its captions, the
side rail on its agent tab reporting a finished render against the timeline,
and the layered timeline below — picture, waveform and captions as three
projections of one edit](docs/img/edit-mode.png)

## How it works

Every hard part of an editor already exists as mature open source. proofcut
is the layer that lets an agent drive them, and check what they produced:

| Job | Done by |
|---|---|
| Cutting, concatenating, captions, rendering | ffmpeg |
| Word-timed transcription (30+ languages) | openai-whisper |
| Silence and bad-take removal | auto-editor |
| The timeline, and export to other editors | OpenTimelineIO |
| Layered rendering (b-roll, cards, music) | MLT |
| Title and end cards | SVG templates, rendered by ImageMagick |

## Try it

Check your machine first, before cloning anything. `proofcut doctor` probes
every tool proofcut uses and prints the fix for anything missing
([§ Requirements](#requirements) has the list). With
[uv](https://docs.astral.sh/uv/) and git installed:

```sh
uvx --from git+https://github.com/tydude001/proofcut proofcut doctor
```

The first run downloads Python 3.13 if uv has none, plus proofcut's
dependencies, about 230 MB together. The demo and your own recordings run
from a checkout:

```sh
git clone https://github.com/tydude001/proofcut && cd proofcut
uv sync
```

### The two-minute demo

No footage needed. [docs/DEMO.md](docs/DEMO.md) generates a voiceover with a
real retake, b-roll and a score, then walks a whole small film: cut the retake
by naming its words, hang b-roll off a phrase, lay the score under the voice,
render and master it, end on a card, and check the render against the
timeline.

```sh
uv run python scripts/make_demo.py ~/proofcut-demo
```

### In Claude Code

The plugin registers proofcut's MCP server, so all 90 tools are available
with no setup of your own:

```
/plugin marketplace add tydude001/proofcut
/plugin install proofcut@proofcut
```

The first start downloads about 175 MB of Python dependencies, and Claude
Code gives a server 30 seconds to connect. On a slow connection, start that
first session as `MCP_TIMEOUT=300000 claude`. If `/mcp` already shows proofcut
as failed, reconnect it there; the download keeps what it fetched.

Any other MCP client runs the same server, from a checkout or with no
checkout at all:

```sh
uv run --project /path/to/proofcut proofcut mcp
uvx --from git+https://github.com/tydude001/proofcut proofcut mcp
```

### On your own recording

```sh
uv run proofcut init myproject
uv run proofcut -C myproject import VO.wav --clip-id vo
uv run proofcut -C myproject transcribe vo                  # whisper, word-timed
uv run proofcut -C myproject seed vo                        # auto-editor strips silences
uv run proofcut -C myproject transcript vo --search "here's the thing"
uv run proofcut -C myproject cut vo 111:114 --plan          # what do those indices say?
uv run proofcut -C myproject cut vo 111:114 --pad 0.1       # inclusive word range
uv run proofcut -C myproject export final.mp4 --render      # or a .kdenlive to finish in an NLE
uv run proofcut -C myproject verify final.mp4               # did the render say what you edited?
```

To watch the edit instead, open the workspace. It plays the source through
the edit, so seeing a cut costs no render:

```sh
uv run proofcut -C myproject open        # server plus an app window
uv run proofcut -C myproject web --open  # the same page in a browser tab
```

## What it does

One line each. The [manual](docs/MANUAL.md) covers every command and the
reasoning behind it.

**Editing**

- **Cut by transcript.** Word ranges, phrases, or spans of timeline time, all
  undoable.
- **The workspace.** Edit (transcript, preview, drag-trim and razor), Frame
  (review every crop in place) and Finish (presets, verify, and the finished
  file). The truth strip warns while you edit if the film would ship wrong.
- **Captions from the timeline.** They stay right after cuts, and the style
  is saved with the project. Karaoke highlight, sidecar ASS, or burned in.
- **Multi-mic recordings.** A file with two mics is refused until you say how
  to use it (`--mix` or `--audio-stream k`), and `attribute-speakers` labels
  each word with who said it.

**Sound**

- **A score, placed.** `music` lays a bed under the voiceover from a word,
  as passages that crossfade into each other or assets that rotate. Its level
  is measured against the voice, and it moves with every cut because it stores
  words, not seconds.
- **The footage's own lin