How optimization works

Content-aware optimization, explained.

Most optimizers apply one setting to everything: a fixed quality number, a fixed format, applied identically to a product photo, a logo, and a screenshot. That's why generic optimization either leaves savings on the table or visibly damages some of your assets.

Inverity looks at each asset first, then picks the method that fits that specific content.

The pipeline

1. Analysis. Inverity inspects the asset and classifies its content type. For images that's photo, screenshot, icon, or transparent graphic; for video it's talking head, screen recording, motion graphic, or archive master. See Content detection.

2. Routing. Based on that classification, your quality profile, and the plan you're on, Inverity selects an engine and settings. You don't choose the engine — Inverity chooses per asset.

3. Encoding. The asset is processed. Nothing has touched your library yet; this all happens on a copy.

4. Quality check. The result is compared against your original. Inverity is looking for a file that is both smaller and visually equivalent. Both conditions have to hold.

5. Write-back, or not. If the result passes, it's written back to your platform. If it doesn't, your original stays exactly as it was and the asset is marked skipped — not smaller. See How write-back works.

Quality is protected by default

The quality check is not optional and can't be turned off. You will not get a worse-looking file to save a few kilobytes.

This is also why savings vary so much between libraries. A library of unoptimized camera exports might average 60–70% savings. A library that's already been through a build-time optimizer might average 10%, with a high skip rate. Both are correct results — the second library simply had less room in it.

The engines

Inverity routes each asset to the most effective engine available on your plan, from fast modern-format encoding up to advanced neural compression on higher tiers.

Higher plans don't optimize better in the sense of producing worse quality more cheaply — they give the router more options for content that the simpler engines can't improve. In practice that shows up as a lower skip rate and better savings on difficult assets, not as a visible difference in the assets that were already being handled well.

See Compare plans for which engines each tier unlocks.

What Inverity does not do

• It doesn't resize your images. Optimization changes the encoding, not the dimensions. A 4000px image stays 4000px.
• It doesn't strip your assets from your platform. Everything stays where it lives; Inverity writes back in place.
• It doesn't serve your media. Inverity is not a CDN or an image proxy. Your existing delivery path is unchanged — which is also why you may need to clear a cache to see the new files.