Free to self-host. Metered on what actually costs.
The platform is Apache-2.0 and yours to run. The meter only counts render minutes, storage and model-gateway spend — published machine-readably at pricing.md so your agent can budget a season before it spends a cent.
Self-host
The whole platform, Apache-2.0. Bring your own model keys.
- All seven primitives
- MCP + CLI + SDK
- Your own render/model spend
- Community support
Cloud Pro
Hosted MCP, managed model gateway, metered on what actually costs.
- Remote MCP over OAuth
- Managed model gateway (one key)
- Metered render + storage
- Agent-readable pricing.md
- Continuity Advisor
Studio
For teams running serials at volume, with rights & provenance.
- Everything in Pro
- SSO + content provenance
- Priority render capacity
- Batch season pipelines
- Dedicated support
What the meter counts
Indicative gateway rates — the numbers your agent reads before it renders.
| Video render | ≈ $0.84 / shot | veo-3 class, 9:16, bound to canon refs |
| Image render | ≈ $0.04 / image | fable-5-image class stills & boards |
| Audio render | ≈ $0.02 / clip | elevenlabs-v3 class dialogue & score |
| Storage | metered | masters, takes and canon reference sets |
Rates are indicative and model-dependent; the live, exact numbers are what your agent reads from the gateway before ordering.
Questions, answered
What's the difference between self-hosting and Cloud Pro?
Self-hosting gives you the entire platform under Apache-2.0 — one docker compose up, your own model keys, your own infrastructure, free forever. Cloud Pro adds the parts that are annoying to run yourself: a hosted remote MCP endpoint with OAuth, a managed model gateway behind one key, and metered render/storage so you never manage model accounts.
What exactly is metered?
Only what actually costs money: render minutes (video, image, audio through the model gateway) and storage. The platform itself — primitives, MCP server, CLI, continuity Advisor — is not metered. Indicative gateway rates: video ≈ $0.84/shot, images ≈ $0.04, audio clips ≈ $0.02.
Can my agent read the pricing?
Yes — that's the point. Rates are published machine-readably (pricing.md and /llms.txt), so a coding agent can budget a whole season before ordering a single render, and stop when it hits the budget you set.
Do I need my own model API keys?
Self-hosting: yes, you bring keys for the render models you want (veo, fable, elevenlabs, …). Cloud Pro: no — the managed gateway fronts every model behind one lumilib key, and spend lands on one metered bill.
What license is the platform under?
Apache-2.0, including the seven primitives, the MCP server and the CLI. No open-core gotchas on the pipeline itself; cloud is a hosting convenience, not a license unlock.
When can I sign up?
lumilib is launching shortly. Every button on this site will light up the moment it does — until then they say 'coming soon'.