GenStudio

AI Content Factory - generate images, video, and audio from a single prompt. Built with Genblaze SDK and Backblaze B2 durable storage.

Comment

Inspiration

Every day, small businesses and content creators spend hours writing marketing copy, creating images, and editing videos for their campaigns. Most of them don't have a team of designers and copywriters. I wanted to build something that lets anyone describe their campaign in one sentence and get a complete content package with images, video, and voiceover automatically.

What it does

GenStudio is an AI Content Factory. Type one prompt like "Create a product launch for my coffee shop Brew Haven" and the system generates:

  • Marketing copy with headline, tagline, description, and social media captions
  • Product images using GMI Cloud (Seedream models)
  • Promo video using GMI Cloud (Seedance models)
  • Voiceover narration using ElevenLabs TTS

Every asset is stored on Backblaze B2 with SHA-256 provenance manifests. All assets get durable, credential-free URLs that never expire.

How I built it

The system uses the Genblaze SDK to chain multiple AI providers through a unified Pipeline API:

  1. Copy Generation - OpenRouter / GMI Cloud chat models write marketing copy
  2. Image Generation - GMI Cloud Seedream models generate product images
  3. Video Generation - GMI Cloud Seedance models create promotional video
  4. Audio Generation - ElevenLabs TTS produces voiceover narration

Every step feeds into a Genblaze Pipeline with an ObjectStorageSink configured for Backblaze B2. Assets are organized with a hierarchical key strategy under runs/{tenant}/{date}/{run_id}/.

The backend is Python FastAPI. The frontend is a Streamlit dashboard for monitoring.

Challenges

The hardest part was understanding the Genblaze SDK provider adapter pattern. Each provider has a different API, different poll lifecycle, and different output format. Genblaze abstracts all of this behind a consistent submit/poll/fetch_output interface.

Another challenge was optimizing for the hackathon budget. Using OpenRouter for chat text was more cost-effective than using GMI Cloud chat, reserving GMI Cloud credits for the media generation where they provide the most value.

Accomplishments

  • Built a complete multi-provider content pipeline in under 48 hours
  • Integrated Backblaze B2 with provenance manifests automatically
  • Deployed the live backend on Railway
  • All code is open source on GitHub with MIT license

What I learned

Genblaze makes multi-provider orchestration incredibly clean. The Pipeline API with chain=True lets me pipe image output directly into video generation. The ObjectStorageSink handles all B2 uploads, URL rewriting, and manifest creation in a single operation.

I also learned that production-ready AI apps need graceful degradation. If one provider fails, the system still delivers results from the others.

What's next

  • Add full image, video, and audio generation with proper API credits
  • Add a scheduling system for recurring campaign generation
  • Build a gallery view for browsing generated campaigns on B2
  • Add team collaboration features

Built With

  • b2
  • backblaze
  • cloud
  • elevenlabs
  • fastapi
  • genblaze
  • gmi
  • openai
  • openrouter
  • python
  • railway
  • sdk
  • streamlit
Share this project:

Updates