MEDIA PREVIEW
Overview & Result
Developer Nailthy Tang created Drape, a real-time voice-driven virtual try-on shopping assistant. Shoppers speak naturally about weather, occasion, and aesthetics. Drape transcribes speech, uses Jev to immediately select the top clothing items from an inventory of 500 options, and dispatches the item to a fast image-generation model (Fal.ai), rendering the user wearing the outfit in 620ms total at $0.0011 per session.
How Jev fits in the loop
- Mobile app streams user voice input through client-side Whisper transcription.
- The transcript and current user profile are evaluated by Jev against wardrobe catalog categories.
- Jev selects candidate jacket, shirt, and trouser item IDs with confidence scores in 115ms.
- Dispatches product assets to virtual try-on diffusion pipeline, rendering the new look on the user's avatar instantaneously.
How to reproduce
- Watch Nailthy Tang's mobile screen recording demonstrating the interactive voice wardrobe.
- Check out sample clothing catalog schema and Jev Choice prompt specification.
- Feed natural language styling requests ('rainy spring weekend brunch in Shibuya') into Jev Choice.
- Observe deterministic catalog SKU output and rapid diffusion rendering.
Why this build matters
Demonstrates how Jev bridges multimodal voice interfaces and e-commerce inventory. It removes the latency penalty from generative shopping experiences.
Reported performance
Reported by authorCost: $0.0011 / session
Latency: 620ms end-to-end (115ms Jev)
115ms Jev inventory lookup latency; 620ms full round-trip from voice query to rendered clothing overlay via Fal.ai.
Limitations
- Image rendering latency depends on Fal.ai / Flux image generation GPU cold starts.
- Clothing catalog attributes must be cleanly structured in the database.