Batch-convert images (includes vector art), audio, and video — tracing, SVG splitting, and compression with no visible or audible loss.
ConvertAll is a desktop utility that does four media chores which normally need four different tools, in one window: convert a file to a specific format, squeeze a file down without visible or audible loss, trace artwork into vectors, and split a layered SVG into its parts.
It runs entirely on your machine. Nothing is uploaded, nothing phones home, and no original file is ever overwritten.
Two tools do most of the work — Conversion when you need a particular file type, Compression when you just want the file smaller. Vector work lives in its own group.
| Tool | What it does |
|---|---|
| Conversion | For when you need a specific file type. Add images, audio or video and pick the target — the dropdown only offers formats every selected file can actually become. Covers .png → .webp, .heic → .png/.jpg, .png/.jpg → .ico, and .wav → .mp4 or .flac. |
| Compression | For when a smaller file is the goal and the format is not. Each type gets the encoder settings that suit it, tuned to stay visually and audibly lossless. |
| Tracing (Raster to Vector) | Auto-traces artwork into scalable .svg. Two Potrace-based engines: posterised colour layers for artwork, pure black & white for line art. |
| Split SVG Layers | Splits any layered .svg into one standalone file per layer, group or shape. Universal — no assumptions about which editor made the file. |
Tracing, with the engine and detail controls:
- Python 3.10 or newer
- FFmpeg — optional. If it is not on your
PATH, ConvertAll uses the binary bundled in theimageio-ffmpegwheel, so there is nothing to install by hand.
Everything else installs from PyPI - eight packages, all pure wheels. There is no Potrace binary to track down, no ImageMagick and no Inkscape.
git clone https://github.com/alirisner14/ConvertAll.git
cd ConvertAll
python -m venv .venvActivate the environment:
.\.venv\Scripts\Activate.ps1source .venv/bin/activateThen install and run:
pip install -r requirements.txt
python -m convertallpython -m convertall --contrast maximum --text-scale 1.3| Flag | Values | Meaning |
|---|---|---|
--contrast |
dark, maximum |
Palette to start in |
--text-scale |
0.85–1.6 |
Starting text size |
--version |
Print the version and exit |
- Pick a tool in the left sidebar (or
Ctrl + 1…Ctrl + 4). The first two cover most work; the Vector art tools group holds tracing and SVG splitting. - Add files — drag them onto the list, or use Browse / Add folder. Folders are searched recursively and filtered to the types that tool accepts.
- Adjust the options. The defaults are the safe choice for every tool.
- Choose where output goes, or accept the default
ConvertAll Outputfolder next to your first input file. - Press Convert (
Ctrl + Enter). Progress, per-file results and the size saved show up in the activity log.Escstops after the current file.
ConvertAll never overwrites anything. If a name is taken it writes
filename (2).ext instead.
"Smaller" is easy; "smaller and indistinguishable" is the whole point. The rules the pipeline follows:
| Input | Treatment | Why |
|---|---|---|
| Flat artwork / logos (PNG) | True lossless WebP | Under ~256 colours, lossless WebP beats lossy WebP on size and is bit-exact. |
| Photographs | WebP q95 (method=6) |
Above the visually-lossless threshold; typically 40–70% smaller than PNG. |
| JPEG | q95, progressive, no chroma subsampling | 4:4:4 keeps coloured text and fine edges clean. |
| PNG | optimize, compress_level=9 |
Lossless by definition. |
| WAV / AIFF | FLAC | Mathematically lossless, usually ~50% smaller. |
| MP3 / AAC / OGG | Left untouched | Re-encoding lossy audio only destroys it. |
| Video | H.264, H.265 or AV1 at CRF, audio stream copied | Your choice of codec — see below. Audio is copied rather than re-encoded, so it never loses a generation. |
| SVG | Editor metadata stripped, coordinates rounded to 2dp | Geometry is never altered. |
And a safety net: if an "optimised" file comes out larger than the original, ConvertAll keeps the original and says so in the log. This matters most for video — anything already efficiently encoded, or in a newer codec than H.264, can grow when re-encoded, so it is left alone.
The Compression tool lets you pick the codec. Measured on screen-recording content (a lecture capture, slides with slow movement) against an H.264 source:
| Codec | Size saved | Encode speed | Use it when |
|---|---|---|---|
| H.264 | — | ~3× realtime | The file has to play on anything, including old hardware |
| H.265 / HEVC | ~17% | ~2× realtime | You want a decent saving without a long wait |
| AV1 | ~34% | ~0.5× realtime | You are archiving and size matters more than time |
AV1 saves roughly twice what H.265 does and costs about four times the encode time. For a one-hour recording that is roughly two hours of encoding — fine overnight, painful if you are waiting.
Two things worth knowing before you compress a large library:
- Nothing is ever overwritten. Output goes to a separate folder, so you need room for both copies while the job runs. Check the results before you delete any originals.
- Playback support. H.264 plays everywhere. AV1 needs a recent player — VLC, any current browser, or Windows with the AV1 Video Extension. H.265 on Windows may prompt you to buy Microsoft's HEVC extension; VLC plays it free.
ConvertAll/
├── convertall/
│ ├── __main__.py # entry point and dependency check
│ ├── app.py # the window, tool panels, event loop
│ ├── widgets.py # accessible buttons, lists, log view
│ ├── theme.py # contrast-verified palettes and type scale
│ ├── jobs.py # background worker thread + event queue
│ └── core/ # all processing — GUI-free and unit-tested
│ ├── convert.py # what a file can become, and how to get it there
│ ├── images.py # PNG/HEIC/JPG/WebP/ICO
│ ├── media.py # audio → MP4, A/V re-encoding (FFmpeg)
│ ├── vectorize.py # Potrace auto-tracing (colour + mono)
│ ├── svgsplit.py # layer/group/shape splitting
│ └── compress.py # type-aware smart compression
├── tools/ # icon generator, GUI smoke test / screenshots
├── tests/ # pytest suite, no display required
└── .github/workflows/ # CI and release automation
Everything in convertall/core/ is a plain function that takes paths and
returns a TaskResult. You can import and script any of it without touching the
GUI:
from pathlib import Path
from convertall.core import convert_image, split_svg
convert_image(Path("logo.png"), Path("out"), target="ico")
split_svg(Path("artwork.svg"), Path("out"), mode="auto")Ordinary desktop software, built carefully. Nothing here is a mode you have to turn on.
-
Two palettes. The default is a charcoal ground with off-white text and a mint accent. Maximum contrast is white on black, where the selected item inverts rather than relying on a hue at all.
Ctrl + Dswitches.
-
The colours are measured, not guessed. Every pairing clears WCAG 2.1 AAA (7:1), and
tests/test_theme.pyrecomputes them on every build, so a colour tweak that hurts legibility fails CI instead of shipping. -
No saturated yellow. It scores well on a contrast chart and is tiring to look at for an hour. Both palettes hit their numbers without it.
-
Text scales 85–160% with
Ctrl + +/Ctrl + -. The layout reflows rather than clipping. It starts at normal desktop size. -
Hover and focus never look alike. The accent is mint, the focus ring orange — about 125° apart in hue, which a contrast ratio cannot express and a test checks directly. Focus outranks hover, so the mouse never hides where the keyboard is.
-
Everything works from the keyboard. Tab moves, Enter or Space activates, every action has a shortcut (
F1lists them). -
Help text is visible, not hovered. Options carry printed descriptions; tooltips are unreachable by keyboard and screen readers.
-
Native list and log widgets, so assistive technology sees real text.
pip install -r requirements.txt -r requirements-dev.txt
pytest # run the test suite
ruff check . # lint
ruff format . # formatVS Code users get interpreter, test, lint and debug configuration out of the box
from .vscode/. Press F5 to launch the app under the debugger.
See CONTRIBUTING.md before opening a pull request.
"HEIC support missing" — pip install pillow-heif.
"FFmpeg was not found" — pip install imageio-ffmpeg, or install FFmpeg and
put it on your PATH.
Drag and drop does nothing — pip install tkinterdnd2. The app works without
it; the buttons do the same job.
"Only one layer found" when splitting — the SVG has no separate groups to split. Try Every individual shape in the Split by menu.
Tracing a photo produces a huge SVG — that is expected. Auto-tracing suits flat artwork, logos and line art, not photographs. Use Simplified detail, which posterises to four colours before tracing.
Tracing produced the negative of my artwork — the black & white engine fills everything darker than the threshold. Raise or lower the threshold slider, or tick Invert for light artwork on a dark background.
ConvertAll follows Semantic Versioning. Releases are recorded in CHANGELOG.md.
MIT.
ConvertAll builds on Pillow, FFmpeg, Potrace (via the pure-Python potracer port), lxml and CustomTkinter. FFmpeg is distributed under the LGPL/GPL; see ffmpeg.org/legal.