figforge

Turn Figma SVG exports into email-safe, pixel-faithful HTML assets.

figforge is the engine distilled from a real Figma → HTML-email rebuild. The one-off fixes that job needed — a baked hero image, transparent white icons, a silhouette rim-glow, a cut-out aligned flush to a band — turned out to be instances of a handful of general operations. figforge is those operations, so the next email is a short recipe, not a forensic expedition.

Why it exists

Browsers render Figma's CSS faithfully; email clients do not. Outlook's Word engine ignores background-size, CSS gradients and position; Gmail strips base64 images and clips large HTML. The reliable move is to bake design into flat images and keep the HTML to tables — but doing that 1:1 with the design, by hand, is fiddly. figforge automates the fiddly parts and verifies the result.

The five parts

Install

pip install figforge                 # core: forensics, image ops, effects, lint
pip install "figforge[render]"       # + SVG/HTML rendering

The [render] extra (cairosvg, weasyprint, pymupdf) needs system libraries that pip can't install:

# Debian/Ubuntu
sudo apt-get install libcairo2 libpango-1.0-0 libpangocairo-1.0-0
# macOS
brew install cairo pango

lint and forensics (and the pure-Pillow helpers) work without the extra.

Quick start

figforge lint email.html                                   # catch client gotchas
figforge rasterize icon.svg --width 48 --recolor FFFFFF -o icon.png
figforge extract master.svg --image-id image1 -o photo.png # decode an embedded raster
figforge render-svg master.svg --width 680 -o target.png   # 1:1 reference
figforge render-html email.html --width 680 --height 1862 -o actual.png
figforge diff target.png actual.png -o heat.png

import figforge as ff

doc = ff.parse_svg("master.svg")
band = next(r for r in doc.rects if doc.gradient_for(r))    # a gradient stripe
grad = doc.gradient_for(band)
shadow = next(s for s in doc.shadows.values() if s.color == (255, 255, 255))

man = ff.dematte(ff.clean_alpha(ff.extract_embedded_image("master.svg", "image1")))
glow = ff.rim_glow(man, shadow, source_card_px=(300, man.width))
hero = ff.bake([(glow, (0, 0)), (man, (0, 0))], size=man.size, background="#2D2C2C")
hero.save("hero.png")

A complete, dependency-light example is in examples/quickstart.py.

Principles it encodes

1. Bake, don't layer. One flat <img> renders identically everywhere; CSS backgrounds + cover + overlap do not.
2. Pin widths. Give images and banner tables explicit pixel widths; never rely on inline-block to fill width.
3. No base64 in email. Host assets; reference them by https URL.
4. Verify against the source. Render the design and diff — don't trust an in-app preview, which lies about Outlook/Gmail.

Use it as an agent skill

figforge ships an Agent Skill that teaches AI coding agents the full Figma → email-safe HTML method (the figforge engine plus the client-robustness principles). Install it into your agent:

npx skills add AzalDaniel/FigForge --skill figma-to-email-html

The skill (skills/figma-to-email-html/) is the method; this package is the engine it drives.

Contributing

Issues and PRs welcome — especially new lint rules for client gotchas you've hit. See CONTRIBUTING.md and the Code of Conduct. Where this is headed — a Figma REST ingester, an intermediate representation, DTCG token extraction and an MJML emitter — is in ROADMAP.md.

License

Apache License 2.0. The Apache-2.0 grant includes an explicit patent license, which is the safer default for a rendering/diffing tool.