quietshrink

Compress screen recordings without making your fans spin.
_{70-90% smaller files. Visually lossless. Computer stays silent.}

_{▶ Watch the 30s animated promo}

---

The problem

Mac screen recordings are huge. A 3-minute screencast easily hits 100+ MB at retina resolution. Existing compressors (HandBrake, ShrinkVideo, Compressor) use software encoders that pin your CPU at 100% for minutes — fans roar, battery drains, your laptop becomes a heater.

The fix

quietshrink runs entirely on Apple Silicon's Media Engine — the dedicated chip that handles HEVC encoding without touching the CPU. Combined with smart frame deduplication and tuned GOP settings, it produces files typically 80-90% smaller than the original while staying visually indistinguishable from the source.

The computer stays silent. Even on a fanless MacBook Air, you won't notice it's running.

Real numbers

A typical macOS screen recording (3024×1964 @ 120fps, 3 minutes):

Tool	Output Size	Encoding Time	Computer Load
Original	105 MB	—	—
HandBrake (default)	42 MB	4 min	🔴 Fans on
ShrinkVideo	35 MB	5 min	🔴 Fans on
libx265 CRF 18 (manual ffmpeg)	25 MB	10 min	🔴 Fans on
quietshrink	12 MB	90 sec	🟢 Silent

Quality: SSIM 0.99+ (visually identical to source).

Install

curl -fsSL https://raw.githubusercontent.com/achiya-automation/quietshrink/main/install.sh | bash

Or manually:

# Requires ffmpeg
brew install ffmpeg

# Clone and link
git clone https://github.com/achiya-automation/quietshrink.git
sudo ln -s "$(pwd)/quietshrink/bin/quietshrink" /usr/local/bin/quietshrink

Usage

# Default: transparent quality (visually lossless)
quietshrink recording.mov

# Tiny preset — for chat / email sharing
quietshrink -q tiny recording.mov

# Replace original
quietshrink --replace recording.mov

# Batch a whole folder
for f in ~/Desktop/*.mov; do quietshrink "$f"; done

Quality presets

Preset	Use case	Typical reduction	SSIM
tiny	Chat / email	~90%	0.95
balanced	Documentation	~88%	0.99
transparent (default)	Anything important	~87%	0.99+
pristine	Archival / editing	~84%	0.997

How it works

Three optimizations layered together:

ffmpeg -i input.mov \
  -vf "mpdecimate,setpts=N/FRAME_RATE/TB" \
  -c:v hevc_videotoolbox -q:v 60 \
  -prio_speed 0 -spatial_aq 1 \
  -g 600 \
  -tag:v hvc1 \
  -c:a aac -b:a 96k \
  -movflags +faststart \
  output.mov

1. hevc_videotoolbox — Apple's hardware HEVC encoder. Runs on the Media Engine chip, not the CPU. This is why your fans don't spin.

2. mpdecimate — Detects and drops duplicate frames. Screen recordings at 60-120fps have many identical consecutive frames (no on-screen change between captures). Removing them is lossless and shrinks the file dramatically.

3. spatial_aq + long GOP — Apple's adaptive quantization sends more bits to text and edges, fewer to flat areas. Long GOP (600) reduces redundant keyframes for static content. Together, these two tweaks cut another 50-70% off the size.

For the deep-dive on why each parameter matters, see WHY.md.

Requirements

• macOS with Apple Silicon (M1, M2, M3, M4)
• ffmpeg 6+ (brew install ffmpeg)

Intel Macs work but without hardware acceleration — the computer will get loud, defeating the purpose.

Comparison vs alternatives

Tool	Encoder	Hardware accelerated	Smart dedup	Polished CLI
HandBrake	x264/x265 (sw)	optional, low quality	❌	GUI only
ShrinkVideo	libx265 (sw)	❌	❌	⚠️
Compresto	libx265 (sw)	❌	❌	GUI only, paid
Apple Compressor	videotoolbox	✅	❌	paid, complex
quietshrink	hevc_videotoolbox	✅	✅	✅

Agent integration (CLI-Anything)

quietshrink ships with an agent-native CLI compatible with CLI-Anything — AI agents (Claude Code, Cursor, OpenClaw) can install and use it via a single pip command:

pip install cli-anything-quietshrink

The agent harness wraps the bash CLI with structured JSON output and skill files agents can read.

FAQ

Why is the file smaller than HandBrake's output?

HandBrake uses general-purpose presets. quietshrink is specifically tuned for screen recordings: long GOP (most screen content is static), aggressive frame deduplication (consecutive frames are often identical), and Apple's adaptive quantization (more bits to text). Together these layer multiplicatively.

Will it work on my old Intel Mac?

It will compress fine, but hevc_videotoolbox falls back to software on Intel — your CPU will be busy and fans will spin. If you have an Intel Mac, you might be just as well off with HandBrake.

Can I use this for camera footage / vlogs / screen + camera?

It works but the savings will be smaller. Camera footage doesn't have duplicate frames, so mpdecimate doesn't help. You'll still get ~50% reduction from videotoolbox + long GOP, just not the 80-90% screen recordings see.

Why not AV1?

AV1 compresses ~25% better than HEVC at the same quality, but Apple Silicon (through M4) has no hardware AV1 encoder. Software AV1 encoding (SVT-AV1) is loud and slow. When Apple adds hardware AV1 encoding (likely M5), quietshrink will switch to it.

How is quality verified?

We use SSIM (Structural Similarity Index) — a perceptual quality metric. The default preset achieves SSIM > 0.99, which is below the threshold of human visual detection.

Contributing

PRs welcome! See CONTRIBUTING.md.

Bug? Open an issue.

License

MIT. See LICENSE.

Credits

Built on top of:

• FFmpeg — the universal video toolkit
• Apple VideoToolbox — hardware encoding
• mpdecimate filter — frame deduplication

Distributed via CLI-Anything for AI agent ecosystems.

---

_{If quietshrink saved you a fan spin-up, consider giving it a ⭐}

---

_{Built by Achiya Automation — automation & AI tooling.}