# monofilm

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Morteza nokhodian esfahani](https://cerebralvalley.ai/u/monoes)
- **GitHub:** https://github.com/nokhodian/monofilm
- **Demo video:** https://youtu.be/2KdxXoeaYDE
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3502

Monofilm is an AI film studio where the AI proposes and a human directs.

https://monofilm.nokhodian.workers.dev/
Username: judges@monofilm.dev
Password: EJcUAqXbJg7tIpE8

AI video tools are fast and unreliable at the same time. A model puts a character on the wrong side of the frame, changes a costume between shots, or ignores the camera you described. You find out after you've paid for it. There is no crew, no continuity and nobody in charge.

Monofilm gives an AI film the structure of a real production. You bring a story, which can be a line, an excerpt or notes. A guided wizard turns it into everything a video generator needs: Start, Understand, Look, People, Places, Scenes and Ship. The AI makes each piece for you to review, and you approve it or click it and say what's off. The server enforces that only a person can approve, so the AI can propose and comment but cannot approve. If you change something, everything built on it is marked out of date, and the price of redoing it is shown first. Each production has a credit budget.

Key features:
- Margin card: click any sentence, face, place, scene or picture. Tap what's wrong, drop a numbered pin on the image, tick what must not change, or speak your feedback. The AI redoes only that one item.
- Stage: drag people, objects and the camera on a floor plan, pick a lens and a move. That layout is what the start and end pictures are generated from, so the model is told where everyone stands
- Layout check: after pictures are generated, SAM 3.1 finds each person and object and compares them with the planned layout.
- Pack: Meta has no video model, so each scene leaves as ar. It holds the prompt and negative prompt, start and endpictures, stage controls, transcript and subtitles, and a manifest. You bring the generated clip back, and the studio checks its first and last
frame against your pictures.
- Roll camera: one button lets the AI run every step end to end, recorded as "Approved by Roll camera". Everything stays editable afterwards.
- MCP server: any outside AI agent can read the human's feorking on the same versioned artifacts as the wizard.

The demo story, "Ask First", is about an AI that must ask h is the product itself.

Built on Cloudflare Workers, D1 and R2 with vanilla JavaScttps://monofilm.nokhodian.workers.dev

What is not built: video generation, and voice or sound since the meta model wouldn't support, its possible o quickly connect it to higgsfiled as or any other voice or video local models as I did to make a sample productions.

## More from Meta Global AI Developer Hackathon

- [StepSignal](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3499)
- [oguirado](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3500)
- [Agentic Appetite Adventurers](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3501)
- [Strativu](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3503)
- [TraceLoop](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3504)
- [Snap Part](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3505)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3502. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt