# Super Ninja

- **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:** [Chirantan Patra](https://cerebralvalley.ai/u/cpatra)
- **GitHub:** https://github.com/cpatra11/phantasm
- **Demo video:** https://youtu.be/dC2aPdYvWrY
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4444

What Phantasma is
Phantasma turns a brand URL and one prompt into brand-consistent, editable motion graphics: launch promos and posters, exported as MP4 (9:16 and 16:9) or PNG. Built for small marketers and founders who don't design, don't know After Effects, and don't have weeks to wait.

Link- https://phantasm-lake.vercel.app

The problem
Launching anything online means paying for ads, but pro tools take weeks to learn, template tools ignore your brand, and agencies cost time and money. phantasma closes that gap: real motion graphics, your own brand identity, one sentence.

How it works
Paste "yourbrand.com" and the agent reads the site to build a brand kit: logo, colors, fonts, product photos, and headline voice.

Write one prompt, like "launch promo, 9:16".

Three storyboard cards appear, each a different angle, with live progress you can watch while they generate.

Everything is editable. Every headline, logo, and image is its own layer on a visual board. Select the layers you want changed, say "make this punchier", and only those layers change. Every change is saved as a version you can revert to.

Animate a storyboard into one continuous film, then export as MP4 or PNG.

Under the hood
Firecrawl scrapes the site in one call, and brand assets are mirrored to our own storage so generation never depends on the site's CDN.

Meta models do all the AI work: muse-wonder-1.0 plans each run, muse-spark-1.3 drives a real tool-calling agent loop that authors whole storyboard documents, muse-image-1.0 generates brand-locked visuals, and SAM 3.1 segments product and logo cutouts. The brand's real products stay in every frame.

Design mode reads a flat ad with an image-understanding call and rebuilds it as editable layers using sharp and SAM. Animation follows a motion-preset vocabulary with a deterministic choreographer as fallback.

Every step streams live as typed UI events, so cards land on the board the moment they save. Stack: Next.js on Vercel, Neon Postgres + Drizzle, Vercel Blob, Better Auth, Remotion Lambda.

All model traffic runs through one validated JSON facade with deterministic fallbacks, so the demo never dead-ends.

Demo flow
Sign in, the brand card appears, run a prompt, watch three variants generate, open the editor, select a layer and refine, animate, export your 9:16 MP4.

## More from Meta Global AI Developer Hackathon

- [Jeremy Hobson (3tched)](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4441)
- [DnA-Arkitecs](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4442)
- [NovelSpell Remix](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4443)
- [Deception Check](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4445)
- [Anville](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4446)
- [Doc_Coder](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4447)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4444. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt