Optimization
Image optimization
How Inverity handles photos, screenshots, icons, and graphics.
Images are the bulk of most libraries and the easiest place to see results. Inverity classifies each one — see Content detection — and encodes it accordingly.
By content type
Photos
Camera imagery and product shots have continuous tone and fine detail, which compresses extremely well. This is where the largest savings usually come from, and where aggressive profiles are safest — the eye is poor at spotting small errors inside busy detail.
Watch for smooth gradients: skies, studio backdrops, and soft shadows are the first place banding appears if a profile is pushed too far. If you review one photo before widening a profile, review one with a gradient in it.
Screenshots
Screen captures are mostly flat colour with hard edges and text. They need edge-preserving treatment, because the standard lossy approach that flatters a photograph puts visible ringing around every letter.
Savings on screenshot-heavy libraries are typically lower than on photo libraries. That's the content, not a failure of the optimizer.
Icons and logos
Small dimensions, few colours, hard edges. There is often very little room here — a well-made 4 KB logo may already be at its floor, and will legitimately be skipped.
Transparent graphics
Anything with an alpha channel: overlays, badges, cut-out products. Inverity preserves transparency through the round trip. A transparent asset will never be written back with its alpha channel flattened onto a background.
What Inverity doesn't change
- Dimensions. A 4000×3000 image comes back 4000×3000. Optimization changes the encoding, not the size on screen. If you need responsive variants, that's a job for your CMS or CDN.
- Colour appearance. Colour profiles are preserved so your assets don't shift in appearance.
- Filenames and references. The optimized file takes the place of the original in your platform, so existing links and embeds keep working. See How write-back works.
Metadata
Non-essential metadata may be stripped during encoding, which is part of where the saving comes from on assets that carry large embedded thumbnails or editing history.
If your workflow depends on embedded metadata
Some teams rely on embedded IPTC, XMP, or copyright fields for rights management or downstream automation. If that's you, verify a sample asset round-trips with the fields you need before you run a large backfill, and keep master libraries on a conservative profile or outside the watched scope.
Realistic expectations
Savings depend almost entirely on how optimized your library already was:
| Starting point | Typical outcome |
|---|---|
| Raw camera or design-tool exports | Large savings, very few skips |
| Uploaded straight from a phone or stock library | Substantial savings |
| Already run through a build-time optimizer | Modest savings, higher skip rate |
| Hand-tuned by a designer | Mostly skips — nothing left to take |
A high skip rate on an already-optimized library is Inverity working correctly. See Results and savings.