TrustFrame

⏳ Live app note: Hosted on Render's free tier, which sleeps after inactivity. The first load may take ~30–60 seconds to wake up — please give it a moment, then it runs fast.

Inspiration

Anyone can generate AI media from a prompt today, but two problems go unsolved: the AI never checks whether its output actually matches the request, and once you have a file, there's no way to prove how it was made. TrustFrame solves both.

What it does

You give TrustFrame a simple business brief — brand, product, mood, audience. It assembles an optimized prompt, generates the media (image, video, or audio), and an AI judge scores the result against the brief. If it fails, it automatically retries with the judge's feedback. Every final asset is stored on Backblaze B2 with a cryptographic provenance manifest proving exactly how it was made.

How we built it

A FastAPI backend orchestrates the pipeline through the Genblaze SDK, calling GMI Cloud models for generation and an LLM judge for evaluation. Assets and manifests are persisted to Backblaze B2 and served via presigned URLs. The frontend is a guided brief builder. Deployed on Render.

Challenges

Getting the evaluate-and-retry loop reliable, wiring real provider endpoints, and durable storage with verifiable provenance — all as a solo build.

What we learned

APIs, cloud storage, agentic pipelines, and deployment, end to end.

Built With

• backblaze-b2
• fastapi
• gemini
• genblaze
• gmi-cloud
• html
• inworld
• javascript
• pixverse
• python
• render
• seedream