vided 0.1.2

A stupid-simple local video silence speeder and rectangular blur redactor.

Vided

Vided is a small local tool for editing one screen recording at a time.

It can:

• cut short silent gaps
• speed up longer silent sections
• optionally mute sped-up silent sections
• generate thumbnails
• mark fixed rectangular blur regions in a browser UI
• render a debug preview with visible boxes
• render the final blurred video with ffmpeg

Everything is file-based and local. A project folder contains project.json, redactions.json, generated thumbnails, generated filtergraphs, and rendered outputs.

Best fit

Use Vided for mostly static screen recordings where the sensitive area stays in a predictable part of the screen:

• email addresses
• account IDs
• tokens
• URL bars
• sidebars
• terminal panes

Vided is not a tracker. If the thing you need to hide moves, create multiple redactions across smaller time ranges.

Requirements

For video work:

• ffmpeg
• ffprobe

For local development and source installs:

• Python 3.11 or newer
• uv

Speech-based trimming with VAD is optional. Install the ONNX Runtime extra when you need it:

uv sync --extra vad

Run from this repo

Install dependencies and check the CLI:

uv sync
uv run vided --help
uv run vided doctor

Run the CLI module directly:

uv run python -m vided --help

After the package is published to PyPI, install it as a tool:

uv tool install vided
vided --help

For one-off use after publishing:

uvx vided --help

Quick start

Create a project:

vided init /path/to/original-recording.mp4 --output-dir my-recording-project

If you omit --output-dir, init creates a folder from the input filename, such as original-recording.

Trim silence:

vided trim my-recording-project

Open the annotation UI:

vided ui my-recording-project

If thumbnails are missing, ui generates them from work/trimmed.mp4 before opening the browser.

In the UI:

1. Click a thumbnail in the bottom filmstrip.
2. Click Set start.
3. Drag a rectangle in the large frame view.
4. Click the thumbnail where the blur should end.
5. Click Add redaction.
6. Wait for the save status in the top bar.

Render a debug preview first:

vided render my-recording-project --debug --overwrite

The debug output is:

my-recording-project/output/debug-preview.mp4

Render the final blurred video:

vided render my-recording-project --overwrite

The final output is:

my-recording-project/output/final.mp4

Render a contact sheet from the final video:

vided render my-recording-project --contact-sheet --overwrite

The contact sheet shows sampled frames from output/final.mp4. Frames that overlap redactions get an accent border.

my-recording-project/output/contact-sheet.jpg

Trim behavior

The default trim mode is hybrid:

detector: audio
short silent sections: cut
silent sections 1.5s or longer: speed:8,volume:0
normal sections: keep
margin: 0.2s
smooth: 0.2s,0.1s
audio threshold: 0.04

Short pauses disappear. Long waits are sped up and muted. Speech stays at normal speed. The default detector uses ffmpeg-decoded audio levels.

To use VAD instead:

vided trim my-recording-project --detector vad --overwrite

This creates or refreshes work/vad.wav and work/vad_ranges.json as needed, then runs the same ffmpeg trim renderer.

Project layout

After the full flow, a project looks like this:

my-recording-project/
  project.json
  redactions.json
  input/
    original.mp4
  work/
    trimmed.mp4
    vad.wav
    vad_ranges.json
    filtergraph.txt
    filtergraph.debug.txt
    frames/
      frame_000001.jpg
      frame_000002.jpg
      frames.json
  output/
    debug-preview.mp4
    final.mp4
    contact-sheet.jpg

vad.wav and vad_ranges.json exist only after using VAD.

Commands

The normal workflow uses:

• init
• trim
• ui
• render
• doctor

Check dependencies:

vided doctor

Create a project:

vided init input.mp4

Choose a project folder:

vided init input.mp4 project-dir
vided init input.mp4 --output-dir project-dir

Symlink the original instead of copying it:

vided init input.mp4 project-dir --symlink

Run the default trim pass:

vided trim project-dir --overwrite

Use a different trim mode:

vided trim project-dir --mode cut --overwrite
vided trim project-dir --mode speed --overwrite
vided trim project-dir --mode keep --overwrite

Tune silence handling:

vided trim project-dir --silent-speed 12 --overwrite
vided trim project-dir --no-mute-silent-audio --overwrite
vided trim project-dir --smooth 0.2s,0.1s --overwrite

Show a speed badge on sped-up sections:

vided trim project-dir --speed-indicator --speed-indicator-corner top-right --overwrite

The badge appears only when the sped-up section lasts at least one second in the output. To show it on shorter sped-up sections:

vided trim project-dir --speed-indicator --speed-indicator-min-seconds 0.5 --overwrite

Use VAD:

vided trim project-dir --detector vad --overwrite

Generate denser or larger thumbnails before opening the UI:

vided frames project-dir --interval 0.5 --overwrite
vided frames project-dir --thumbnail-width 960 --overwrite

Render to a custom path:

vided render project-dir --output output/my-final.mp4 --overwrite

Render a contact sheet from the default final video:

vided render project-dir --contact-sheet --overwrite

Render a contact sheet from a custom final video:

vided render project-dir --contact-sheet --final-video output/my-final.mp4 --output output/my-sheet.jpg --overwrite

trim, frames, and render also support --dry-run.

Technical design

The pipeline is:

original video
  -> audio-level or optional VAD activity detection
  -> ffmpeg-backed speed/mute trim pass
  -> work/trimmed.mp4
  -> ffmpeg thumbnail generation
  -> browser rectangle annotations
  -> redactions.json
  -> ffmpeg debug preview
  -> ffmpeg final blurred render

Redactions are created against the trimmed video. That keeps all redaction timestamps in the same timeline as the debug and final renders.

The UI stores rectangles in real video pixels, not thumbnail pixels.

Redaction data

The browser writes redactions.json. Each redaction stores selected times, buffered effective times, a video-pixel rectangle, and a style:

{
  "id": "redaction_001",
  "selected_start_seconds": 10.0,
  "selected_end_seconds": 15.0,
  "buffer_pre_seconds": 0.5,
  "buffer_post_seconds": 0.5,
  "effective_start_seconds": 9.5,
  "effective_end_seconds": 15.5,
  "rect": {
    "x": 1000,
    "y": 42,
    "w": 420,
    "h": 90
  },
  "style": {
    "type": "blur",
    "filter": "boxblur",
    "luma_radius": 18,
    "luma_power": 3
  }
}

Rendering

For each blur redaction, the renderer generates an ffmpeg filtergraph that:

1. splits the video stream
2. crops the target rectangle from a copy
3. applies boxblur to the crop
4. overlays the blurred crop back onto the base video during the redaction time range

Generated filtergraphs are written to:

work/filtergraph.txt
work/filtergraph.debug.txt

The final render must re-encode video because blur is a video filter. Defaults:

{
  "video_codec": "libx264",
  "crf": 16,
  "preset": "medium",
  "pixel_format": "yuv420p",
  "audio_codec": "copy"
}

Lower CRF means higher quality and larger files. Try CRF 12 or 14 for very high quality, or 18 to 23 for smaller files.

Audio is stream-copied during the redaction render. Silence speeding and muting happen during the trim pass.

Limitations

• Fixed rectangles only. There are no keyframed rectangles or object tracking.
• Annotations happen on the trimmed video. Changing trim settings later means regenerating thumbnails and reviewing redactions.
• The UI uses frame thumbnails plus debug render. It does not include a video preview.
• Frame times are based on the thumbnail interval. This is fine for broad redactions with buffers, but not frame-perfect editing.
• Rotation metadata and unusual pixel aspect ratios are not normalized.
• Blur is implemented in the UI. Solid redaction is partially supported by the renderer if you hand-edit style.type to solid.
• The local UI server binds to 127.0.0.1 by default. Do not expose it to the public internet.

Development

Run tests:

uv run pytest

Format, lint, and type-check:

uv run ruff format src tests
uv run ruff check src tests
uv run ty check src tests

Install the pre-commit hook:

uv run pre-commit install

Every commit with package source or release metadata changes must bump the package version with zerover. This includes staged changes under src/ and published package metadata in pyproject.toml, such as runtime dependencies, entry points, Python version support, or build configuration. Tests, workflows, docs, scripts, dev dependencies, and uv.lock-only changes do not require a bump.

uv version --bump patch
# or, for larger changes while still staying on major zero:
uv version --bump minor
git add pyproject.toml uv.lock

After committing, add the matching tag:

version="$(uv version --short)"
git tag -a "v$version" -m "v$version"
git push origin HEAD --tags

Publishing runs from .github/workflows/publish.yml when a v* tag is pushed. In PyPI, configure a trusted publisher for:

• owner: pmbaumgartner
• repository: vided
• workflow: publish.yml
• environment: pypi

Realistic media fixtures

The repo includes two Git LFS-tracked public-domain NASA fixtures:

• tests/fixtures/media/realistic-speech-gaps.mp4
• tests/fixtures/media/realistic-speech-gaps-short.mp4

Use them for realistic trim, VAD, frame generation, render, and UI smoke tests. Source and license notes live beside each fixture as *.LICENSE.md.

Run fixture e2e tests:

uv run --extra vad pytest --run-e2e -m e2e

Install the Playwright browser once before browser e2e tests:

uv run playwright install chromium
uv run --extra vad pytest --run-e2e -m browser --browser chromium