WayCast

Turn any company's URL into a polished vertical brand‑film — script, voiceover, city backgrounds, final MP4 — without opening an editor.

Open‑source · self‑hosted · local‑first. No avatars, no stock‑footage clichés — clean institutional motion‑graphics. The AI agent driving it is the writer, so it runs with zero cloud keys if you want.

_{中文说明 → README.zh.md · MIT OR Apache‑2.0 · Node ≥ 22 · pre‑alpha}

---

60‑second try

npx waycastai doctor                                            # what your machine can do
npx waycastai make https://stripe.com --reuse-bg wjdigital     # URL → brand package in ./brands/
npx waycastai render <slug> --quality standard                 # → ./output/*.mp4

Everything lands in your folder. Prefer a UI? npx waycastai console → http://127.0.0.1:4178.

What it makes

A 1080×1920, 8‑shot institutional brand film — cover → who‑we‑are → what‑we‑do → why → how → who‑it‑powers → beliefs → CTA — data‑driven from a reusable block library, EN + ZH, real‑voice narration, per‑shot city backgrounds.

Why WayCast is different

Most "URL → video" tools are paid SaaS, or wrap a cloud API behind talking‑head avatars. WayCast is the opposite end:

	WayCast
Agent‑native	Driven by Claude Code / any agent via MCP — the agent writes the storyboard itself, so good copy needs no LLM key
Local‑first	Renders locally; bundled local TTS (Kokoro EN / CosyVoice ZH same‑voice clone, Apache‑2.0) — or BYO cloud key
Zero‑dependency	Pure Node + ffmpeg. No framework lock‑in
Style	Clean text‑motion brand films — no avatars, no random stock B‑roll
Bilingual	English + 简体中文 first‑class (most OSS video tools are EN‑only)
Yours	Self‑hosted, dual‑licensed, keys stay in your .env

Three ways to drive it

1. CLI / npx — npx waycastai make <url> → render.
2. Any coding agent — clone, point Claude Code at it; it reads CLAUDE.md and authors + renders the film itself. See docs/agent-usage.md.
3. MCP — claude mcp add waycast -- npx waycastai mcp → 7 tools (waycast_scrape / write_brand / render / …). See docs/mcp.md.

How it works

URL ─▶ scrape ─▶ brief (+brand color) ─▶ script (EN/ZH copy) ─▶ pick city images
                                                                      ▼
        brand package (brand.json + storyboard.json + vo.json + bg)
                                                                      ▼
        local TTS  ─▶  retime to narration  ─▶  render (hyperframes)  ─▶  MP4

Every stage has a keyless local/deterministic fallback and an optional BYO‑key upgrade. Anti‑fabrication is enforced — the copy only uses facts from the site.

Backends (all optional)

Capability	Local (offline)	Cloud (BYO‑key)
Voiceover (required, pick one)	Kokoro (EN) / CosyVoice (ZH)	OpenAI / ElevenLabs / Azure
Copywriting	deterministic draft or the driving agent	Anthropic / OpenAI
City images	--reuse-bg / drop your own	Pexels / Unsplash

Keys go in .env (.env.example). npx waycastai doctor shows what's ready. Full setup: docs/install.md.

Status — pre‑alpha (honest)

Core pipeline (URL → standard MP4) is verified end‑to‑end via the local path. Cloud LLM/TTS/image adapters are implemented but not yet live‑tested; a one‑click Docker image for the local models is WIP; Windows is untested (use WSL). It's a self‑hosted dev tool, not a turnkey consumer app. Changelog: CHANGELOG.md.

License

Dual‑licensed MIT OR Apache‑2.0 — your choice. Third‑party components (local TTS models, fonts, CC‑BY sample images, GSAP via CDN) in NOTICE.

Contributing

Early days — issues and ideas welcome. See CONTRIBUTING.md. If it's useful, a ⭐ helps others find it.