Evals / video / Speed and stabilise

Steady a shaky clip

shaky.mp4 was filmed by hand. Stabilize it and save it as steady.mp4.

normalsingledeft video stabilize

The task

The reference answer shows the job can be done with deft. It is not an agent run.

deft video stabilize shaky.mp4 -o steady.mp4

Input

shaky.mp4 3.5 MB · h264 1280x720, 10.0s

Reference output

steady.mp4 1.6 MB · h264 1280x720, 10.0s
How a result is judged (2 checks)
  • media: file = "steady.mp4", has_video = true
  • python: code = "import re, subprocess\nout = subprocess.run(['ffmpeg', '-i', 'steady.mp4', '-vf', 'tblend=all_mode=difference,signalstats,metadata=print:key=lavfi.signalstats.YAVG:file=-', '-f', 'null', '-'], capture_output=True, text=True).stdout\nchange = [float(v) for v in re.findall(r'YAVG=([0-9.]+)', out)]\nassert sum(change) / len(change) < 6, 'the picture still shakes'\n"

Runs

CodexWithout deftPassed2026-11-07

Time
39s
New tokens
13.2k
Total input
126.1k
Output tokens
0.7k
Commands
3
Failed
0

Trace

  1. I’ll inspect the video and available local tools, then stabilize it and write `steady.mp4` in the current directory.

  2. pwd && rg --files -g 'shaky.mp4' -g 'AGENTS.md' -g '*.mp4' | head -80 && command -v ffmpeg && ffmpeg -version | head -2
    output
    shaky.mp4
/opt/homebrew/bin/ffmpeg
ffmpeg version 7.1.1 Copyright (c) 2000-2025 the FFmpeg developers
built with Apple clang version 15.0.0 (clang-1500.1.0.2.5)
  3. ffprobe -v error -show_entries format=duration:stream=codec_name,width,height,r_frame_rate -of default=noprint_wrappers=1 shaky.mp4 && ffmpeg -hide_banner -filters | rg 'vidstab|deshake'
    output
    codec_name=h264
width=1280
height=720
r_frame_rate=30/1
duration=10.000000
 ... deshake           V->V       Stabilize shaky video.
 ... vidstabdetect     V->V       Extract relative transformations, pass 1 of 2 for stabilization (see vidstabtransform for pass 2).
 ... vidstabtransform  V->V       Transform the frames, pass 2 of 2 for stabilization (see vidstabdetect for pass 1).

  4. The clip is 10 seconds at 1280×720, and FFmpeg includes the two-pass `vidstab` filters. I’m running motion analysis and applying the transforms with a small crop/zoom to hide edge gaps.

  5. ffmpeg -y -i shaky.mp4 -vf " vidstabdetect=shakiness=5:accuracy=15:result=<tmp> " -f null - && ffmpeg -y -i shaky.mp4 -vf "vidstabtransform=input=<tmp> unsharp=5:5:0.5:3:3:0.0" -c:v libx264 -preset medium -crf 18 -c:a copy steady.mp4
    output
    ffmpeg version 7.1.1 Copyright (c) 2000-2025 the FFmpeg developers
  built with Apple clang version 15.0.0 (clang-1500.1.0.2.5)
  configuration: --prefix=/opt/homebrew/Cellar/ffmpeg/7.1.1_1 --enable-shared --enable-pthreads --enable-version3 --cc=clang --host-cflags= --host-ldflags='-Wl,-ld_classic' --enable-ffplay --enable-gnutls --enable-gpl --enable-libaom --enable-libaribb24 --enable-libbluray --enable-libdav1d --enable-libharfbuzz --enable-libjxl --enable-libmp3lame --enable-libopus --enable-librav1e --enable-librist --enable-librubberband --enable-libsnappy --enable-libsrt --enable-libssh --enable-libsvtav1 --enable-libtesseract --enable-libtheora --enable-libvidstab --enable-libvmaf --enable-libvorbis --enable-libvpx --enable-libwebp --enable-libx264 --enable-libx265 --enable-libxml2 --enable-libxvid --enable-lzma --enable-libfontconfig --enable-libfreetype --enable-frei0r --enable-libass --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-libopenjpeg --enable-libspeex --enable-libsoxr --enable-libzmq --enable-libzimg --disable-libjack --disable-indev=jack --enable-videotoolbox --enable-audiotoolbox --enable-neon
  libavutil      59. 39.100 / 59. 39.100
  libavcodec     61. 19.101 / 61. 19.101
  libavformat    61.  7.100 / 61.  7.100
  libavdevice    61.  3.100 / 61.  3.100
  libavfilter    10.  4.100 / 10.  4.100
  libswscale      8.  3.100 /  8.  3.100
  libswresample   5.  3.100 /  5.  3.100
  libpostproc    58.  3.100 / 58.  3.100
Input #0, mov,mp4,m4a,3
… 12973 more characters

  6. Stabilized the video and saved it as [steady.mp4](steady.mp4). It’s 10 seconds at 1280×720.

Output

steady.mp4 7.1 MB · h264 1280x720, 10.0s

CodexWith deft, skill installedPassedRead the skillUsed deft2026-11-07

Time
44s
New tokens
9.6k
Total input
166.1k
Output tokens
1.0k
Commands
8
Failed
0

Trace

  1. I’ll inspect the video and available media tools, then stabilize it and write `steady.mp4` in the current directory.

  2. cat .agents/skills/deft/SKILL.md && rg --files -g 'shaky.mp4' -g 'steady.mp4' -g 'AGENTS.md'
    output
    ---
name: deft
description: "Use deft instead of writing a script for file jobs: PDF merge/split/compress/watermark, image resize/convert/compress, video and audio trim/convert/compress, spreadsheets, Markdown to PDF/HTML, and YouTube downloads and subtitles."
---

# deft

One local command for file jobs. Prefer it over writing Python or shell scripts.

Syntax: `deft <group> <command> [inputs] [-o output] [--options]`

Groups:
- `pdf`: Merge, split, compress, encrypt, forms, OCR, compare, convert
- `image`: Resize, crop, convert, compress, filters, montage, OCR
- `video`: Trim, crop, speed, convert, compress, GIF, frames (ffmpeg)
- `audio`: Convert, trim, merge, fade, speed, tag, normalize (ffmpeg)
- `doc`: Word/ODT/RTF: to text, Markdown, HTML, PDF; find and replace
- `slides`: PowerPoint: extract text, Markdown outline, PDF handout
- `data`: JSON/YAML/TOML/XML: convert, query with jq syntax, merge, diff
- `sheet`: Excel/CSV/ODS tables: filter, sort, join, pivot, convert
- `md`: Markdown to PDF, HTML, Word, EPUB; table of contents
- `html`: HTML to Markdown, text, Word, PDF; screenshots
- `epub`: E-books: info, to text, Markdown and PDF
- `email`: EML files: headers, text, attachments
- `contact`: VCF address books: to and from CSV
- `calendar`: ICS calendars: to and from CSV
- `gps`: GPX and KML tracks: convert, distance and time
- `youtube`: Info, download, audio, subtitles, thumbnail (yt-dlp)
- `file`: Any file: hash, duplicates, batch rename, base64, encoding
- `qr`: Cre
… 738 more characters

  3. The `deft` skill covers video conversion but doesn’t list stabilization, so I’m checking its command set and the local FFmpeg availability to choose the direct stabilization path.

  4. ls -lh shaky.mp4
    output
    -rw-r--r--@ 1 user  staff   3.5M Nov  5 21:54 shaky.mp4
  5. deft video
    output
    deft video - commands (input -> output)
  info               video -> report      Duration, size, frame rate, codecs
  add-audio          video -> video       Replace or mix the soundtrack: --audio song.mp3
  add-subtitles      video -> video       Add subtitles: burned into the picture, or --soft
  chunk              video -> video...    Cut into equal parts: --length 10:00
  compress           video -> mp4         Smaller file: --level, --target-size, --max-height
  convert            video|gif -> video   Change container and codec: --to mp4|webm|mov|mkv
  crop               video -> video       Cut a region: --box or --aspect 1:1
  download           url -> video         Save a video: --resolution 720p, --to mp4|webm|mkv [net]
  extract-audio      video -> audio       Save the soundtrack: --to mp3|m4a|wav|flac|ogg
  extract-subtitles  video -> srt|vtt     Save an embedded subtitle track: --to srt|vtt
  fade               video -> video       Fade: --fade-in 1 --fade-out 2
  frames             video -> png|jpg...  Save frames as images: --fps 1 or --at 0:05,0:10
  from-images        image... -> video    Slideshow: --duration, --fps, --audio
  merge              video... -> mp4      Join videos; --transition fades
  mute               video -> video       Remove the audio track
  pad                video -> video       Fit into a frame with bars: --size 1080x1920
  repeat             video -> video       Repeat a video: --count 3 plays it three times
  resize             vid
… 891 more characters
  6. ffmpeg -version
    output
    ffmpeg version 7.1.1 Copyright (c) 2000-2025 the FFmpeg developers
built with Apple clang version 15.0.0 (clang-1500.1.0.2.5)
configuration: --prefix=/opt/homebrew/Cellar/ffmpeg/7.1.1_1 --enable-shared --enable-pthreads --enable-version3 --cc=clang --host-cflags= --host-ldflags='-Wl,-ld_classic' --enable-ffplay --enable-gnutls --enable-gpl --enable-libaom --enable-libaribb24 --enable-libbluray --enable-libdav1d --enable-libharfbuzz --enable-libjxl --enable-libmp3lame --enable-libopus --enable-librav1e --enable-librist --enable-librubberband --enable-libsnappy --enable-libsrt --enable-libssh --enable-libsvtav1 --enable-libtesseract --enable-libtheora --enable-libvidstab --enable-libvmaf --enable-libvorbis --enable-libvpx --enable-libwebp --enable-libx264 --enable-libx265 --enable-libxml2 --enable-libxvid --enable-lzma --enable-libfontconfig --enable-libfreetype --enable-frei0r --enable-libass --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-libopenjpeg --enable-libspeex --enable-libsoxr --enable-libzmq --enable-libzimg --disable-libjack --disable-indev=jack --enable-videotoolbox --enable-audiotoolbox --enable-neon
libavutil      59. 39.100 / 59. 39.100
libavcodec     61. 19.101 / 61. 19.101
libavformat    61.  7.100 / 61.  7.100
libavdevice    61.  3.100 / 61.  3.100
libavfilter    10.  4.100 / 10.  4.100
libswscale      8.  3.100 /  8.  3.100
libswresample   5.  3.100 /  5.  3.100
libpostproc    58.  3.100 / 58.  3.100
  7. ffmpeg -h filter=vidstabtransform
    output
    ffmpeg version 7.1.1 Copyright (c) 2000-2025 the FFmpeg developers
  built with Apple clang version 15.0.0 (clang-1500.1.0.2.5)
  configuration: --prefix=/opt/homebrew/Cellar/ffmpeg/7.1.1_1 --enable-shared --enable-pthreads --enable-version3 --cc=clang --host-cflags= --host-ldflags='-Wl,-ld_classic' --enable-ffplay --enable-gnutls --enable-gpl --enable-libaom --enable-libaribb24 --enable-libbluray --enable-libdav1d --enable-libharfbuzz --enable-libjxl --enable-libmp3lame --enable-libopus --enable-librav1e --enable-librist --enable-librubberband --enable-libsnappy --enable-libsrt --enable-libssh --enable-libsvtav1 --enable-libtesseract --enable-libtheora --enable-libvidstab --enable-libvmaf --enable-libvorbis --enable-libvpx --enable-libwebp --enable-libx264 --enable-libx265 --enable-libxml2 --enable-libxvid --enable-lzma --enable-libfontconfig --enable-libfreetype --enable-frei0r --enable-libass --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-libopenjpeg --enable-libspeex --enable-libsoxr --enable-libzmq --enable-libzimg --disable-libjack --disable-indev=jack --enable-videotoolbox --enable-audiotoolbox --enable-neon
  libavutil      59. 39.100 / 59. 39.100
  libavcodec     61. 19.101 / 61. 19.101
  libavformat    61.  7.100 / 61.  7.100
  libavdevice    61.  3.100 / 61.  3.100
  libavfilter    10.  4.100 / 10.  4.100
  libswscale      8.  3.100 /  8.  3.100
  libswresample   5.  3.100 /  5.  3.100
  libpostproc    58.  3.100 / 58.  3.100
Filter vidstabtransform
… 2396 more characters

  8. `deft` has a dedicated stabilization command. I’m checking its options, then I’ll run it with the default steadying settings and confirm the output file exists.

  9. deft video stabilize --help
    output
    deft video stabilize - Steady shaky footage, or --denoise grain

Usage: deft video stabilize <video>... [-o <video|dir>] [options]
In: video (each handled separately)   Out: video (default <name>.steady.<ext>)

Options:
  --denoise                    Also reduce grain
Common: -o --force --dry-run --json -j   (deft help common)

Examples:
  deft video stabilize clip.mp4 --denoise -o steady.mp4
Needs: ffmpeg, ffprobe

  10. The stabilization pass is still running; it processes the footage in multiple passes. I’m waiting for the encoder to finish before checking `steady.mp4`.

  11. deft video stabilize shaky.mp4 -o steady.mp4
    output
    ok video.stabilize shaky.mp4 -> steady.mp4  1.6MB  21.82s
  12. ls -lh steady.mp4 && deft video info steady.mp4
    output
    -rw-r--r--@ 1 user  staff   1.6M Nov  6 17:31 steady.mp4
ok video.info steady.mp4  0:10, 1280x720
duration=10.0 width=1280 height=720 fps=30.0 video_codec=h264 bitrate=1359649

  13. Stabilized `shaky.mp4` and saved the 10-second, 1280×720 result as [steady.mp4](steady.mp4).

Output

steady.mp4 1.6 MB · h264 1280x720, 10.0s

Claude Opus 5.5Without deftPassed2026-11-07

Time
119s
New tokens
14.6k
Total input
110.3k
Output tokens
3.0k
Commands
0
Failed
0

Trace

    Output

    steady.mp4 7.9 MB · h264 1280x720, 10.0s

    Claude Opus 5.5With deft, skill installedPassedRead the skillUsed deft2026-11-07

    Time
    38s
    New tokens
    10.7k
    Total input
    91.0k
    Output tokens
    0.8k
    Commands
    0
    Failed
    0

    Trace

      Output

      .claude/skills/deft/SKILL.md 2.1 KB
      ---
name: deft
description: "Use deft first for any simple, direct operation on local files, whatever the type: convert, read, extract, edit, compress, merge, split, compare or inspect. Try it before writing a script or one-liner. If unsure, run deft find; if it says no command does this, use another tool. Not for programming work: to read, write or change source code, skip it."
---

# deft

One local command for file jobs. Prefer it over writing Python or shell scripts.

Syntax: `deft <group> <command> [inputs] [-o output] [--options]`

Groups:
- `pdf`: Merge, split, compress, encrypt, forms, OCR, compare, convert
- `image`: Resize, crop, convert, compress, adjust, montage, OCR
- `video`: Trim, merge, convert, compress, subtitles, GIF, frames (ffmpeg)
- `audio`: Convert, trim, normalize, denoise, cut silence (ffmpeg)
- `doc`: Word/ODT/RTF: to text, Markdown, HTML, PDF; find and replace
- `slides`: PowerPoint: extract text, Markdown outline
- `sheet`: Excel/CSV/ODS tables: SQL query, pivot, merge, split, convert
- `json`: JSON: convert, format, merge, diff, validate
- `toml`: TOML: convert, format, merge, diff, validate
- `xml`: XML: convert, format, merge, diff, validate
- `yaml`: YAML: convert, format, merge, diff, validate
- `md`: Markdown to PDF, HTML, Word, EPUB; table of contents
- `html`: HTML to Markdown, text, Word and PDF
- `epub`: E-books: info, to text, Markdown and PDF
- `gps`: GPX and KML tracks: convert, distance and time
- `qr`: Create QR codes and barcodes; read QR codes

How to use it:
1. Run `deft <group>` to list commands, or `deft find "<what you want>"` to search. If it says `no command does this`, use another tool.
2. Run `deft <group> <command> --help` for options and examples.
3. `deft convert <in> -o <out.ext>` converts between formats by extension.
4. Results are one line starting with `ok`, `partial` or `error`; add `--json` to parse them. Errors include a `hint:` with the fix.

Inputs are never overwritten; pass `--force` or `--in-place` to allow it. Globs and directories work as inputs; several inputs run in parallel.

If `deft` is not installed, ask the user, then follow https://github.com/firstprayer/deft/blob/main/docs/INSTALL.md
      steady.mp4 1.6 MB · h264 1280x720, 10.0s

      Claude CodeWithout deftNot run yet

      Claude CodeWith deftNot run yet