ambhora-brandkit

Generates every distribution variant of the Ambhora brand assets from the pristine sources, which are never modified.

pip install .[render]
brandkit build -o dist

No Inkscape required

The original plan was to shell out to Inkscape for everything. That turned out to be unnecessary. The work splits into four parts, and only one of them actually needs a renderer:

So SVG-only builds run with nothing but lxml. PNG and PDF need exactly one of:

• cairosvg — a pip-installable library, so pip install .[render] gives a complete install with no external binaries. Writes true vector PDFs. Default.
• rsvg-convert — from librsvg. Excellent SVG conformance, very fast, and already present on most Linux systems and via Homebrew.
• inkscape — still supported. It is the slowest of the three, but it is the only backend that can measure and render live text, which makes it the right choice for the source set (see below).

brandkit info reports which backends are usable on the current machine, and --engine picks one explicitly.

Why the bounding box did not need Inkscape either

Padding has to be measured from the ink, not from the page, so the one thing genuinely worth calling inkscape --query-all for was the bounding box. svgbbox.py computes it directly instead: Bézier segments are bounded by their real stationary points, elliptical arcs are converted to cubics first, and affine transforms are composed down the tree and applied to control points before extrema are solved — which keeps the result exact under rotation and skew. Against a 4000-point resampling of random cubics it agrees to 1.5e-5.

On the real assets it reproduces the page geometry exactly: the icon measures 1600×1600 to the last decimal.

The one thing it cannot do is measure live <text>, because that needs font metrics and shaping. This only affects the source set. When it happens the build says so and falls back to the document page, or you can pass --engine inkscape on a machine with the brand font installed.

The variant matrix

Three marks × three inks × three backgrounds × three formats, minus the two illegible pairings (black on black, white on white):

Seven legible combinations per mark, per padding rule, in svg, pdf and png (at each requested raster width). With the default two padding rules and three PNG widths that is 210 files.

Output is laid out as dist/<mark>/<format>/<name>, with self-describing names:

dist/icon/svg/ambhora-icon-white-on-black-clearspace.svg
dist/icon/png/ambhora-icon-color-on-transparent-none-512px.png
dist/wordmark/pdf/ambhora-wordmark-black-on-white-clearspace.pdf

A dist/manifest.json records every file with its size, SHA-256, canvas geometry and the digest of the pristine source it came from, so a build is auditable and reproducible.

Padding

Yes, padded variants work — and they are exact, because padding is applied as a change of window, not a transform of the artwork. The viewBox is widened around the ink box; not a single coordinate in the drawing moves. A test asserts this by re-measuring the ink after padding and comparing it to the original.

Clear space is expressed as a fraction of the mark's ink height, applied equally on all four sides. Height is the right reference here because the wordmark is roughly 7:1 — a width-derived margin would swallow it, whereas a height-derived margin is how exclusion zones are specified in practice, as a multiple of the cap height.

brandkit paddings prints this at the terminal. Select rules with repeated --padding flags.

The share directory

The pristine assets live in share/ambhora-brandkit/{primary,source} and are byte-for-byte identical to the originals. They install as shared data, to <prefix>/share/ambhora-brandkit/, rather than being buried inside the Python package — so a Makefile, a LaTeX build or a shell script can reach them just as easily as Python can.

At runtime they are looked up in this order:

1. $AMBHORA_BRANDKIT_SHARE
2. <sys.prefix>/share/ambhora-brandkit — a normal or virtualenv install
3. <site.USER_BASE>/share/... — a pip install --user
4. /usr/local/share/..., /usr/share/... — system packages
5. share/ambhora-brandkit above the source tree — editable installs and plain checkouts, so the tool works straight from a clone

primary has all type converted to outlines and is what you should build from. source keeps editable text and needs the ADLaM Display brand font installed to render correctly.

Usage

brandkit info                          # assets, palette, backend status
brandkit list                          # print the variant matrix
brandkit paddings                      # explain the clear-space rules
brandkit build -o dist                 # everything
brandkit build -n                      # dry run, writes nothing

brandkit build --asset icon --format png --padding avatar --png-width 512
brandkit build --format svg            # needs no rendering backend at all
brandkit build --engine rsvg -j 8
brandkit build --asset-set source --engine inkscape

Also available as python -m ambhora_brandkit.

Layout

pyproject.toml
share/ambhora-brandkit/     pristine assets, installed to <prefix>/share/
  primary/                  type outlined — build from this
  source/                   live text — needs the brand font
src/ambhora_brandkit/
  geometry.py               affine matrices, rectangles, transform parsing
  svgbbox.py                analytic ink bounding boxes
  svgdoc.py                 recolour, recanvas, background plate
  padding.py                clear-space rules
  spec.py                   the variant matrix and naming scheme
  engines.py                cairosvg / rsvg / inkscape backends
  build.py                  planning, execution, manifest
  cli.py                    command line interface
tests/

Licensing

The brand assets under share/ carry their original CC-BY-4.0 headers and are unchanged. The Python code is marked MIT, since CC licences are not intended for software — change it if you would rather it matched the assets.