Based in Kyiv

©2026

Four Seconds to 1.9: Rebuilding a Photographer’s Site Without Losing a Single Image

A white feather resting on a dark basalt stone

A Kyiv photographer’s site looked stunning and loaded like a brick. Here is exactly what we measured, cut and rebuilt to get her Largest Contentful Paint under two seconds.

Daryna Lis shoots interiors for architecture studios across Ukraine and Poland, and her old site had one job: show the work at full quality. It did that. It also took 5.8 seconds to reach Largest Contentful Paint on a mid-range phone over a throttled 4G connection, and her bounce rate on mobile was sitting above 60 percent. She came to us not because the site looked bad, but because people were leaving before it finished loading.

The Site That Wouldn’t Load

The old build was a Squarespace template with 340 images spread across twelve galleries, every one of them a full-resolution JPEG exported straight from Lightroom. Some files were 4.2 megabytes. The homepage alone pulled 14.1 megabytes before a visitor saw anything usable. There was no lazy loading, no responsive srcset, and the custom font, a licensed serif called Canela, loaded as four separate static weights totaling 310 kilobytes.

Running the First Audit

We started with Lighthouse and WebPageTest, both set to simulate a Moto G4 on a slow 4G profile, because that’s closer to how half of Daryna’s audience actually browses. The numbers were rough: LCP at 5.8 seconds, Cumulative Layout Shift at 0.34, Total Blocking Time at 890 milliseconds. CLS that high meant images were popping into place after text had already rendered, which is exactly the kind of jank that makes a portfolio site feel cheap no matter how good the photography is.

Setting Real Targets

We didn’t aim for a perfect 100 on Lighthouse, because that number means less than people think. Instead we set targets tied to what mattered for her business: LCP under 2 seconds, CLS under 0.05, and total page weight under 3 megabytes on the heaviest gallery page. Those three numbers became the brief.

Photographic prints fanned out beside a film camera
Every image stayed; only the weight went.

Images: From JPEG to AVIF

Images were 90 percent of the problem, so they got 90 percent of the attention. We re-exported every photo from the original RAW files at three widths, 640px, 1280px and 1920px, and converted each to AVIF with a JPEG fallback for older Safari versions. Average file size per image dropped from 1.8 megabytes to 210 kilobytes without a visible quality loss at normal viewing distance, something we confirmed by running a blind comparison with Daryna herself on her own monitor.

Lazy Loading With a Catch

Native lazy loading via loading="lazy" handled most of the work, but we reserved eager loading for the first two images in any gallery, the ones visible on initial paint. Reserving explicit width and height attributes on every image tag killed almost all the layout shift, since the browser could allocate space before the file arrived.

Building a Real Pipeline

Manually exporting 340 images three times each wasn’t sustainable, so we wrote a small Node script using Sharp that watches a source folder and outputs the full set of sizes and formats on save. Daryna now drops a RAW export into a folder, runs one command, and gets production-ready assets in under four minutes.

Fonts: Cutting Four Weights to One Variable File

Canela came as four static weight files because that’s how the foundry originally shipped it for print use. We switched to the variable version, subset it to only the Latin and Cyrillic glyphs Daryna’s site and captions actually use, and ended up with a single 42 kilobyte file replacing four files that totaled 310 kilobytes.

Font Loading Strategy

We added font-display: swap and preloaded the variable file in the document head, so text renders in a system font immediately and swaps to Canela within roughly 80 milliseconds once it arrives. That swap is fast enough that most visitors never notice it happening.

The biggest lie in web performance work is that you have to choose between looking good and loading fast. Most of the weight on a slow site isn’t the design, it’s unoptimized assets nobody ever audited, says Yuri Palchyk, who led the front-end rebuild.

Motion Without Layout Shift

Daryna’s galleries use a crossfade transition between thumbnail and full image, and in the old build that transition was the main source of visible jank, because it ran on properties that trigger layout recalculation. We rebuilt it using only opacity and transform, both of which the browser can animate on the compositor thread without touching layout at all.

The transition timing also changed. The original used a stock ease-in-out at 400 milliseconds, which felt sluggish once the images themselves were loading instantly. We brought it down to 220 milliseconds with a custom curve that front-loads the movement, so it reads as snappy rather than slow, even though the actual difference is less than two tenths of a second.

What Changed, in Numbers

After six weeks of work the same Lighthouse and WebPageTest runs told a different story. Here is the before and after on the heaviest gallery page, run on the identical throttled connection profile:

  • Largest Contentful Paint: 5.8s to 1.9s
  • Cumulative Layout Shift: 0.34 to 0.02
  • Total page weight: 14.1MB to 2.3MB
  • Total Blocking Time: 890ms to 110ms
  • Time to Interactive: 7.2s to 2.6s

Daryna’s own numbers moved too. Mobile bounce rate dropped from 61 percent to 38 percent over the following two months, and average session duration on the gallery pages nearly doubled. None of that came from new content. It came from the same photographs loading the way they were supposed to all along.

What We’d Do Differently

If we started this project again, we would push for the image pipeline rebuild in week one instead of week three. We spent early time on smaller CSS fixes that felt productive but didn’t move any of the three core metrics, and in hindsight that was a sequencing mistake. Performance work rewards starting with whatever is heaviest, not whatever is easiest to fix first.

The other lesson is softer. Daryna was nervous that compressed images would look like a compromise, and the only way past that was sitting with her at a real monitor, A/B testing files side by side until she trusted the process herself. Numbers convince engineers. Photographers need to see the pixels.

Leave a Reply

Your email address will not be published. Required fields are marked *

Recent posts