Jewelry photos on a model without a photoshoot

A jewelry studio or store adds new designs every week. A white-background product shot does not show scale or how a pendant sits on the neck. A model photoshoot requires a studio, photographer, model and retouching, so it usually covers only part of the collection.

The app reads a product page from any online store or uses photos from your phone. It extracts the product dimensions from the description, lays out a shot plan: scene, half-length shot, close-up, detail, portrait, and generates the frames one by one with a progress preview. The finished set is delivered as a ZIP archive.

The application uses Tauri 2 (Rust) with a Node.js engine. The HEXART server issues licenses, by default for one workstation, stores the key for the model provider, and bills calls in PLN. Add funds by card through Stripe or by bank transfer against a pro forma invoice.

Images are generated by GPT Image 2.5 (OpenAI) via OpenRouter. A separate, smaller language model reads stores and product descriptions.

Sales are also handled by the phone line 58 500 40 64. An AI assistant with a synthetic voice answers calls through the phone system we built on Asterisk.

The studio’s first commit was made on September 11, 2026. On September 15, the terms and conditions reached version 1.0, and by September 22, the desktop app was on build 201. The karatvue.com website is now on build 299.

Tasks KV-01 to KV-21 were handled by AI agent sessions, each on its own branch. The main branch has 59,143 lines of TypeScript across 239 files, 16,392 lines of CSS, and 53 test files.

Code signing (Apple notarization and Windows certificate) is pending developer accounts. Planned features include processing the entire collection in one run, an MCP server and REST API, and video and Reels mode.