💡 Inspiration

Live commerce has exploded into a multi-billion dollar market, yet human streamers face burnout, high operational costs, and physical limits to scaling 24/7. Traditional video generators produce static, non-interactive video files that cannot respond to customer questions in real time.

We created AntStudio to bridge this gap: the world's first fully autonomous AI Streamer designed specifically for e-commerce. AntStudio dynamically generates scripts, synthesizes neural audio in multiple languages, renders realistic AI avatar clips, and orchestrates live interactive selling — all backed by high-throughput media storage on Backblaze B2 and workflow management with Genblaze.


🎬 What It Does

AntStudio transforms raw product links or catalogs into interactive, high-converting live streams and short-form marketing videos:

  • 🤖 24/7 Autonomous AI Streamer: AI sales avatars (Aidol) pitch products, respond to live viewer comments, dynamically handle objections, and trigger flash-sale countdowns.
  • 💰 Commerce Intelligence Engine: Real-time dynamic QR code generation for instant checkout, scarcity overlays, and conversion tracking.
  • 🎤 Neural Voice & Instant Dubbing: Multi-language text-to-speech and instant dubbing to reach global buyers in their native language.
  • 📸 Live Highlight Clipping: Autonomous capture of top sales moments, automatically edited into TikTok/Reels/Shorts-ready clips.
  • 🎨 Client-Side Decentralized Video Rendering: High-performance browser-based timeline editor and video assembly engine utilizing @webav/av-cliper and WebWorkers.
  • 📦 Durable Storage & Data Orchestration via Backblaze B2: All generated avatar assets, script clips, provenance metadata, thumbnail variations, and stream archives are stored in Backblaze B2 Cloud Storage.

📸 Platform Showcase

Dashboard AntStudio Central Workspace Dashboard

AI Sale Streaming AntStudio Sale Streaming

Timeline-based Video Editor AntStudio Timeline-based Video Editor

All in One Video Tools AntStudio All in One Video Tools

Multi-AI Providers AntStudio Central AI Providers Configuration

Blaze B2 Storage AntStudio Blaze B2 Storage Configuration

GMICloud and Gemini models AntStudio AI Model Configuration


🛠️ How We Built It

AntStudio is built as an enterprise-grade monorepo combining a modern web interface, robust backend services, and a dedicated AI media orchestration layer:

Architecture Stack:

  • Frontend: Vue 3 + TypeScript + Vite + TailwindCSS + Element Plus. Client-side canvas rendering powered by Fabric.js, WebGL/Three.js for virtual stages, and MediaPipe for AI face tracking.
  • Backend & Storage: Node.js, Express, MongoDB. Native storage abstraction layer supporting Backblaze B2 Cloud Storage via @backblaze-labs/b2-sdk (B2StorageAdapter).
  • AI Media Pipeline & Genblaze: Uses Genblaze SDK (genblaze package) with GMICloudProvider as our primary unified multi-resource AI engine in GMICloudClient to generate all media assets — Text (LLM Reasoning), Image (Gemini 3.1 Flash-Lite Image), Video (Gemini Omni Flash / Kling V2.1), Audio (Inworld TTS 1.5 Mini), and Music (Minimax Music 2.5) — alongside ElevenLabs and Google Gemini.

🛠 Core Capabilities

  • AI Streaming for Sale: 24/7 automated sessions with real-time inventory sync.
  • AI Streamer Builder: Custom face, voice, and brand personality.
  • Product Ads Gen: High-converting video ads generated from product links.
  • AI Video from Storyboard: Transform sketches into full marketing videos.
  • Omni-channel Integration: Native support for YouTube, Facebook, TikTok, and AMS.

🚧 Challenges We Ran Into

  1. Python-Only SDK Ecosystem (Building the Full Genblaze JS/TS SDK): The official Genblaze SDK was strictly available for Python. Since AntStudio is built on a high-performance Node.js & Vue 3 TypeScript stack, we undertook the massive challenge of porting the entire Genblaze SDK 1:1 from Python to TypeScript/JavaScript (genblaze package), including libs/core, 15 AI model connectors, spec, meta, tools, cli, and 30 runnable examples.
  2. Low-Latency Live Response: Generating video/audio fast enough to reply to live viewer comments required pre-generating modular avatar clip libraries, storing them in Backblaze B2, and stitching them on-the-fly.
  3. Serverless Heavy Rendering: Traditional server-side video rendering requires expensive GPU clusters. We solved this by creating a Decentralized Client-Side Rendering Architecture using WebWorkers and @webav/av-cliper, relying on Backblaze B2 to serve raw assets directly to the browser.

🏆 Accomplishments That We're Proud Of

  • Complete 1:1 Line-by-Line Port of Genblaze SDK to TypeScript: Expanded the Genblaze ecosystem into Node.js/TypeScript, allowing JavaScript developers to orchestrate multi-provider GenAI media workflows natively.
  • Built a Working 24/7 Autonomous Live Streamer: An AI sales avatar that pitches products and answers questions in real-time without human intervention.
  • Seamless Backblaze B2 Integration: Embedded B2StorageAdapter into the storage pipeline with automatic fallback and authorization token signing.
  • Standardized Multi-AI Orchestration: Built full pipeline adapters across Google Gemini, GMICloud, Imagen, Veo, TTS, and Google Agent via Genblaze.

📚 What We Learned

  • Genblaze significantly simplifies building multi-provider AI workflows by providing a clean, unified abstraction for generative media models.
  • Backblaze B2's S3 compatibility and transparent pricing make it the ideal storage foundation for bandwidth-heavy media AI applications.

🔮 What's Next for AntStudio

  • Interactive 3D AI Avatars: Integrating WebGL/Unreal Engine avatars rendered directly in browser.
  • Multi-Tenant B2 Analytics: Adding detailed storage analytics and CDN distribution tracking inside the AntStudio enterprise admin dashboard.
  • Expanded Genblaze Integrations: Adding support for newly released video generation models (Sora, Runway Gen-3, Luma Dream Machine) via Genblaze provider adapters.

📬 Get in Touch


Built With

Share this project:

Updates

posted an update —

Integrate Blaze B2 storage + Port GenBlaze JS SDK from Python + GMICloud Platform + Google Agent ADK Build an Electron app for Windows, Linux, and macOS Upgrade Sale Live Stream Engine Fix bugs Join Genblaze Hackathon!

Submission history