Snap-to-Sell

One phone photo → marketplace-ready listing in 30 seconds.

Comment

Inspiration

I run an e-commerce development gig, and the same problem comes up with every client. They have inventory, but they don't have listings — because writing a good marketplace listing takes about 30 minutes per product. Multiply by 200 SKUs and you've got a hundred hours of copywriting, photography, and video work standing between a small store and revenue. So a lot of products just never get listed.

What it does

Snap-to-Sell takes a single phone photo of a product and generates a complete marketplace-ready listing in about 30 seconds.

How we built it

Backend: Node.js 20 + Express on a DigitalOcean droplet, fronted by nginx and exposed via a Cloudflare tunnel for HTTPS. pm2 manages both the Node process and the tunnel. The pipeline runs four AI services, with three of them in parallel:

Gemini 2.5 Flash analyzes the photo and returns structured listing JSON Nano Banana (Gemini 2.5 Flash Image) generates a polished studio hero shot from the original photo ElevenLabs (eleven_multilingual_v2, Bella voice) generates the voiceover from the ad script FFmpeg assembles the final 15-second MP4 with crossfade, drawtext overlays, and the voiceover audio track

Challenges we ran into

The hardest single problem was a filesystem corruption bug in Lovable's build environment. For about two hours, files I could see in Lovable's code editor weren't actually on disk — Vite's build kept failing with ENOENT on components that visibly existed. The chat agent kept "fixing" them by writing files that didn't persist. Eventually I cloned Lovable's GitHub-synced repo locally, manually recreated the four missing components in Claude Code, and finished the build on my own machine.

Accomplishments that we're proud of

Shipped a real end-to-end product solo in less than 36 hours. Not a prototype — a working app at a public URL that anyone can use right now.

What we learned

AI orchestration is the real product. Any one of these services (Gemini, ElevenLabs, FFmpeg) is impressive on its own. The interesting work was making them act like one system. Also, CLAUDE.md is worth its weight in gold.

What's next for Snap-to-Sell

Cinematic ad videos. Multi-angle photo intelligence Platform-tailored copy Direct API publishing Batch upload A/B copy variants Agency mode

Built With

Share this project:

Updates