CoverForge

AI makes the key art. CoverForge makes it publishable.

Deterministic Rust compositor + editable show styles for podcast thumbnails and social covers.

Français · 简体中文

Why

Image models are good at scenes and bad at repeated typography, logos and exact branding. CoverForge keeps those jobs separate:

1. generate native key art for the target ratio;
2. apply exact typography / logo / palette with a deterministic renderer;
3. edit the entire visual identity of a show from one JSON file.

No fake 16:9 → blurred background → inset image conversion.

What ships in v0.3

• ⚡ Rust 2024 + Axum + resvg/tiny-skia
• 🧩 one JSON show-style containing every layout
• 🖥️ Svelte 5 editor with live previews
• 🔤 large redistributable font pack
• ⬆️ custom font upload from the UI
• 📁 custom font folder-drop support
• 🐳 Docker Compose deployment
• 🔌 API-first, with a compatibility endpoint for existing automation
• 🎯 normalized 0–1 coordinates
• 🧱 image / rectangle / text layers
• 💾 atomic template saves

Native social formats

A show style can hold all production outputs in the same file:

Key	Output
youtube	3840×2160
square	1200×1200
feed	1080×1350
vertical	2160×3840
acast	3000×3000

A recommended automation pipeline is:

transcript/editorial direction
        ↓
16:9 key art ──→ YouTube layout
1:1 key art  ──→ Square + Acast layouts
4:5 key art  ──→ Feed layout
9:16 key art ──→ Vertical layout

CoverForge only crops when you explicitly ask a layer to use fit: cover; it does not invent blurred letterboxing.

One file per show

Example production tree:

Templates/
├── cf.json
├── dpafm.json
├── lcfp.json
└── my-new-show.json

Each show JSON contains:

• brand.display_name
• brand.visual_summary
• brand.palette
• brand.fonts
• brand.notes
• formats.youtube
• formats.square
• formats.feed
• formats.vertical
• formats.acast

See docs/SHOW_STYLES.md.

Fonts

CoverForge ships with a broad libre font set and exposes a searchable font browser.

Custom fonts can be added either:

• from the UI;
• by dropping .ttf, .otf, .ttc or .otc files into the persistent Custom folder.

The public repo does not contain private/commercial font binaries.

See docs/FONTS.md.

API

GET    /health
GET    /v1/templates
GET    /v1/templates/{show}
PUT    /v1/templates/{show}
GET    /v1/fonts
POST   /v1/fonts
DELETE /v1/fonts/{filename}
POST   /v1/render
POST   /v1/render/batch
POST   /api/generate

Render one format

{
  "template": "dpafm",
  "variables": {
    "title": "LE STAND-UP EST MORT",
    "eyebrow": "DPAFM #48",
    "background": "/srv/storage/path/native-16x9.png",
    "logo": "/srv/storage/path/DPAFM.png"
  },
  "variants": ["youtube"]
}

Legacy automation adapter

Existing calls such as DPAFM - YT still work. They are resolved to show dpafm, format youtube.

Docker

docker compose up -d --build
curl -fsS http://127.0.0.1:3099/health

The provided Compose file is intentionally Cloud9-flavoured; change the bind mounts for another machine.

Development

cargo test --locked

cd web
npm ci
npm run check
npm run build

No-BS rules

• key art should be generated for the actual target ratio;
• typography belongs to the compositor, not the image model;
• one show = one style file;
• private fonts and production assets stay outside Git;
• a template edit must not require an app rebuild.

License

MIT