The problem
Most WordPress sites have an image problem in both directions: someone uploads a 12MB photo straight off a phone and the page crawls; someone else stretches a tiny logo and it comes out blurry. And most "optimize everything" plugins make it worse in a quieter way — they claim they've optimized everything when they haven't, and they make destructive changes you can't undo.
What we built
A two-part product built around honest, reversible optimization.
- Optimizer service. A stateless Node +
sharpservice recompresses each image and emits modern WebP/AVIF derivatives, guaranteeing no output is ever larger than the original — "already optimal" is reported honestly rather than faked as a saving. - ML upscaling, charge-once. Undersized images run through a Topaz upscaler before the same pipeline. The credit-charging call is guarded server-side so it can never bill twice, with a "needs review" queue for any run interrupted mid-flight.
- The plugin. Bulk optimize, an embedded-image scanner that repoints oversized
<img>tags to right-sized variants with conflict-safe undo, a per-image report with CSV export / restore / re-optimize, and a<picture>delivery layer serving WebP/AVIF with graceful fallback. - Reliability as a feature. A self-healing job queue re-enqueues stuck work, reclaims stale locks, and re-syncs the ledger — and a versioned API contract fails CI if the service and plugin ever drift apart.
Status
In final testing — the plugin is packaged and the service runs on Fly.io. Same standards as our client work: a small, well-instrumented backend that tells the truth about what it did.
