A large gemstone inventory built over years — but imaged at resolutions and framing standards that no longer meet current e-commerce expectations. We built an automated batch tool that upgrades every image: sharper, higher resolution, consistently composed. No re-shooting required.
The retailer's legacy gemstone images were captured at 750–1080p with inconsistent framing — stones appearing at varying sizes within the frame, large empty backgrounds, and resolution that fell short of current display standards. Re-shooting hundreds of stones was not practical. The images needed upgrading without returning a single stone to the photography studio.
A batch processing tool that automates the upgrade end-to-end. It runs AI super-resolution to upscale each image 4x, applies AI gemstone segmentation to detect the stone boundary precisely, then reframes the image so the gemstone occupies a defined, consistent proportion of the output frame. The operator configures the settings once — every image in the batch is processed from there without manual intervention.
- AI 4x super-resolution — conservative upscaling that preserves edge detail without artifacts
- AI gemstone detection and boundary extraction for precise, consistent reframing
- Automatic reframing to a configurable stone-to-frame ratio, applied uniformly across the catalog
- Batch CLI with INI configuration file — no per-image labor
- Optional sharpening pass
- CUDA GPU acceleration for fast batch throughput
Conservative upscaling preserves what is already there. The goal is not to invent detail — it is to restore what compression and low resolution obscured.
Each pair shows the same gemstone image before and after processing — higher resolution, consistently framed, stone occupying a defined portion of the output.
Before
After
Before
After
Before
After
Hundreds of legacy inventory images upgraded to current e-commerce presentation standards — without re-shooting a single stone. Every image now meets the same resolution and framing specification, restoring visual consistency across the full catalog.