Most "image compressor" benchmarks ask the wrong question: which tool makes the smallest file?

But a lot of real jobs have a hard cap, not a "smaller is better" goal. A CMS field that rejects anything over 100KB. An email banner. A form attachment limit. For those, the question is: which tool can actually land under my limit, and what does it cost in quality?

So I ran that exact task on three tools — same inputs, same 100KB target — and measured exact output bytes plus SSIM.

The setup

Four synthetic source images (a noisy photo, a gradient photo, a UI screenshot, a noisy graphic), all over 100KB. Target: ≤102,400 bytes (100 KiB), judged on exact downloaded bytes, not the rounded number each tool shows.

  • DigImg — has a target-size feature: type the cap, it binary-searches encoder quality to land under it in one click. Browser-local.
  • TinyPNG (free web) — just compresses. No target-size control, no resize.
  • Squoosh — browser-local, but you hit a target by manually dragging the quality slider.

The results

Tool Input Output Hit ≤100KB? SSIM
DigImg noisy photo 93.7 KB ✅ 0.672
DigImg gradient photo 93.6 KB ✅ 0.631
DigImg UI screenshot 82.7 KB ✅ 0.978
DigImg noisy graphic 97.6 KB ✅ 0.674
TinyPNG noisy photo 484.7 KB ❌ 0.960
TinyPNG gradient photo 424.4 KB ❌ 0.953
TinyPNG UI screenshot 67.0 KB ✅ 0.996
TinyPNG noisy graphic 336.5 KB ❌ 0.925
Squoosh noisy photo 97.8 KB ✅ 0.676

Measured results

Three things this shows

1. "Just compress" doesn't hit a hard cap. TinyPNG kept quality high (SSIM 0.92–0.996) but landed at 336–485KB on three of four inputs. The free web tool has no way to force a target — it only hit 100KB on the screenshot, which was already the smallest input. If your job is "must be under 100KB," a tool without target control can't guarantee it.

2. Hitting the cap costs quality — and that's honest. To force a noisy 652KB photo under 100KB, DigImg dropped to quality 60 and SSIM fell to 0.67. Squoosh, manually tuned to q61, landed at the same place. There's no free lunch on hard, detailed content. Any tool claiming "smallest file AND best quality" is selling something.

SSIM trade-off

3. The real difference is workflow. DigImg and Squoosh both reached the target with MozJPEG at similar quality. DigImg did it in one click (its target-search binary-searches quality internally); Squoosh took ~7 manual slider adjustments. DigImg automates the iteration you'd otherwise do by hand.

DigImg target-size workflow

The privacy angle (scoped, verified)

I instrumented fetch/XHR/sendBeacon in the page across four DigImg runs: zero image-payload egress. The only non-analytics traffic was same-origin WASM encoder fetches and local blob: fetches. That's a scoped observation of these test workflows — not a packet-level audit, and not "no network requests at all" (analytics still runs).

TinyPNG, by contrast, uploaded the image: a POST to /backend/opt/store carrying the full image blob, then /backend/opt/process with job metadata. To be fair, TinyPNG's docs say uploads are encrypted, links expire in 8h, and files are deleted within 48h — server-side, but not proven insecure. Squoosh is also local-first per its docs; I didn't run an equivalent capture on it here.

Privacy data path

Caveats

Four synthetic, noise-heavy inputs — not a universal benchmark. Real photos compress more gracefully. SSIM is one metric, not human judgment. Single runs. The "~1–1.5s" search time and "~7 steps" are my operator observations, not logged JSON fields.

Try it

DigImg runs in your browser, no account, no image upload: img.digdevbox.com

Full method, all six evidence figures, and the raw data: the canonical write-up on Forge Notes.

Disclosure: DigImg is my product. TinyPNG and Squoosh are independent. The comparison page has a ShortPixel affiliate link (commission possible). Inputs were synthetic; no personal images.