AetherFlow Creative Engine

Agentic multimodal media studio and provenance aware pipeline powered by Backblaze B2 cloud storage, Genblaze workflow orchestration, gmi cloud model router and Google Gemini engines.

Comment

Inspiration

Modern digital content creation requires juggling fragmented tools for image generation, video synthesis, voiceover audio, background score composition and asset provenance tracking. Creators often struggle with siloed generative models, unpredictable latency, lack of immutable audit trails and clunky manual workflows. AetherFlow Creative Engine was inspired by the vision of a unified, agentic multimodal media studio a single orchestrator that seamlessly connects Google Gemini, Veo 3.1 video generation, Lyria music streams and GMI Cloud routing into an immutable, provenance-aware pipeline powered by S3 compatible cloud object storage.

What it does

AetherFlow Creative Engine is a full stack multimodal media production and orchestration studio. Key features include; Text to Image and Image to Image: High-fidelity visual synthesis powered by Google Gemini and custom model routers. Veo 3.1 Video Generation: Cinematic video rendering with real time polling and progress tracking. Gemini TTS Voiceover: Multi-speaker script synthesis with dynamic voice timbres. Lyria 3 Music Composition: High-quality backing track generation tailored to prompt moods. Agentic Campaign Execution: One-click execution of multi-step DAG workflows (visuals + copy + script + audio + music). Real-Time Job Tracker Tand toast System: Interactive stepper progress tracking real-time status, model routing, quality verification and cloud commit phases. S3 and Cloud Storage Vault: Persistent storage interface with live bucket stats, SHA-256 verification, and file download management. C2PA Provenance Audit Trail: Cryptographic lineage records tracking model versions, prompt seeds, execution latency, and quality verification scores for every asset. Local File Queue and Batch Ingest: Drag-and-drop raw media ingest queue for processing local assets into pipeline operations. AetherFlow AI Copilot: Interactive multi-turn assistant for designing pipeline scripts, optimizing prompts and managing storage settings.

How we built it

Frontend Architecture: Built with React 18, TypeScript, Vite and Tailwind CSS. We integrated lucide-react for iconography and custom animated components for smooth stepper progress and modal dialogs. Backend Service Layer: Node.js/Express server handling server-side API proxy routes to protect credentials, process generative AI payloads, manage video generation polling operations and compute SHA-256 asset hashes. Model Integrations such as @google/genai TypeScript SDK for Google Gemini multimodal generation and TTS audio synthesis. Veo 3.1 long-running operation polling handlers. Custom Lyria generative audio composition handlers. Storage and Provenance: Persistent object vault indexing system storing asset metadata, execution timings, and C2PA cryptographic signatures.

Challenges we ran into

Async Long-Running Operations: Video generation models like Veo 3.1 operate asynchronously. Designing a non blocking background polling mechanism that reports intermediate progress back to the user without UI freezes required careful state orchestration. Multimodal State Synchronization: Combining images, TTS voice tracks and Lyria music into a cohesive campaign pipeline required building robust DAG orchestrator logic and state handlers. Immutability and Provenance Verification: Ensuring every generated media asset carries a valid cryptographic SHA-256 hash, execution timestamp and provider lineage while preserving low-latency UI updates.

Accomplishments that we're proud of

Seamless Multimodal Pipeline: Successfully unified 4 distinct generative AI domains (Image, Video, Voice, Music) into a single agentic workflow. Real-Time Pipeline Stepper: Built a responsive, live job tracking system that gives creators complete visibility into pipeline execution (Router → Model → Quality Guard → Cloud Commit). Production-Ready Provenance: Created an accessible, visual C2PA audit tool that makes AI asset verification transparent and trustworthy.

What we learned

Agentic Pipeline Design: Breaking complex media tasks into modular, decoupled DAG steps simplifies error handling and recovery during model execution. User Experience for Async AI: Real-time feedback (percentage bars, step badges and live execution timers) drastically improves creator trust when waiting for heavy model renders. Server-Side API Security: Enforcing server-proxied API routes is essential for shielding sensitive model credentials while maintaining high-performance streams.

What's next for AetherFlow Creative Engine

Real-Time Canvas Editor: Visual drag-and-drop node graph for visually chaining AI inputs and outputs. Enhanced Multi-Speaker Audio Editing: Timeline-based audio mixing for overlapping voice tracks and background Lyria scores. Multi-Cloud Vault Replication: Automatic cross-region object replication and automated C2PA blockchain notary signing.

Built With

Share this project:

Updates