video-compress-to-size git:20260812.225d65dA

Immutable. This exact content is served forever at /api/v1/blob/c58faff40c8a4b0c.

---
name: video-compress-to-size
description: Compresses video files to stay under a user-specified max file size using FFmpeg. Prefers GPU encoders (NVENC / AMF / QSV) with VBR bitrate targeting; falls back to CPU two-pass H.264/HEVC. Use when the user wants to compress video, reduce video file size, shrink MP4/MKV/MOV under N MB, GPU encode, NVENC compress, 压缩视频, 缩小视频体积, or batch-compress clips to a size limit.
---

# Video Compress To Size

Re-encode supported videos so each output is **at or under** a given max file size.

**Default:** probe and use a GPU encoder when available (NVIDIA NVENC → AMD AMF → Intel QSV), single-pass VBR. If no GPU works, fall back to **CPU two-pass** (`libx264` / `libx265`).

## Rules

When this skill applies, read and follow [skill-dependency-manager](../skill-dependency-manager.md) — run scripts as documented, install missing tools into `.dependency/`.

## Quick Start

```bash
.dependency/python/python .cursor/skills/video-compress-to-size/scripts/compress.py path/to/video_or_folder --max-size 50MB
```

Bare numbers mean **MB** (`--max-size 50` ≡ `50MB`):

```bash
.dependency/python/python .cursor/skills/video-compress-to-size/scripts/compress.py assets/intro.mp4 --max-size 50
```

Force CPU (slow, more precise two-pass):

```bash
.dependency/python/python .cursor/skills/video-compress-to-size/scripts/compress.py clip.mp4 --max-size 50MB --cpu
```

## Size Syntax

| Input | Meaning |
|-------|---------|
| `50` / `50MB` / `50M` | 50 mebibytes (1024² bytes) |
| `500KB` / `500K` | 500 kibibytes |
| `1GB` / `1G` | 1 gibibyte |
| `52428800B` | exact bytes |

`--max-size` is **required**.

## Format Defaults

| Setting | Default | Notes |
|---------|---------|-------|
| Encoder | GPU first | `h264_nvenc` → `h264_amf` → `h264_qsv` → `libx264` |
| HEVC | `--hevc` | Same order with `hevc_*` / `libx265` |
| Container | `.mp4` | Always MP4 |
| Audio | AAC 128k | Lowered automatically on tiny budgets |
| Preset | `medium` | Mapped (e.g. NVENC `p4`); override with `--preset` |
| Already under limit | Skipped | `[skip] … (already under limit)` |
| Safety margin | GPU ~90% / CPU ~92% | Headroom for mux / VBR overshoot |

## Common Flags

`-r` · `-o` / `--output-dir` · `--max-size` · `--audio-bitrate` · `--preset` · `--hevc` · `--cpu` · `--overwrite` · `--dry-run`

**Never overwrite source files.** Writes only to `compressed/` (or `-o`). Supported inputs: `.mp4`, `.mkv`, `.mov`, `.avi`, `.webm`, `.wmv`, `.flv`, `.m4v`, `.mpeg`, `.mpg`, `.ts`, `.mts`, `.m2ts`, `.3gp`, `.ogv`, `.ogg`.

## Agent Notes

1. Use the bundled script, not hand-written `ffmpeg` commands.
2. Always pass `--max-size` from the user (ask if missing). Prefer their unit wording; bare numbers are MB.
3. **Prefer GPU** — do not pass `--cpu` unless the user asks or GPU encode fails.
4. Do **not** downscale or change fps unless the user asks.
5. Sources already ≤ max size are skipped.
6. Tell the user where `compressed/` files are; they swap assets manually when ready.
7. Missing Python/FFmpeg → populate `.dependency/` per skill-dependency-manager, retry same command.
8. Pipeline details: [reference.md](reference.md)