With Google just announcing it would be releasing full support for JPEGXL in Chrome version 155, I thought I’d give JPEGXL a try. Here are the results compressing a 16-bit 24 megapixel JPEG. The original is an 8-bit JPEG from a Nikon D3200. Original image dimensions is 6016 x 4000 pixels. AI enhanced to 9024 x 6000 pixels. To judge the image quality fairly, I tried to keep the target sizes roughly the same.
WARNING: These are HUGE FILES. So may take a while to download or choke your Internet browsing experience. Hold tight.

The test image is from an 8-bit JPEG from a Nikon D3200 that was converted to 16-bit, then AI enhanced (sharpened and upscaled). Filesize: 52.2MB
Converted to JPEGXL format via ImageMagick. Note JPEGXL retains the 16-bit color depth. Settings used was effort = 9 and distance = 3. Filesize: 2.62MB

Converted to WEBP format via FFmpeg. Note WEBP reverted back to 8-bit color. Settings used was quality = 50. Filesize: 2.29MB

Converted to AVIF format via FFmpeg. Note AVIF supports max of 12-bit color. Default settings. Filesize: 2.79MB
I dunno bout you, but the results are AMAZING. JPEGXL looks vastly superior over WEBP at around the same filesize. The only problem is it isn’t supported universally and not on older Android systems. You can install the JPEGXL codec for free through Microsoft Store for Windows 10 and 11, but viewing it in Windows Photos app is still clunky. Even on a relatively modern powerful machine, it takes a few seconds to decompress/decode before you see the first pixels. Womp womp. It does load relatively fast in Google Chrome though on a live page.
But if you’re a web designer worth your salt and have done the benchmarks, you’ll know the best format for general images meant for web viewing is still JPEG. But if you must post a higher resolution image, WEBP is the next best because it decodes faster than AVIF and has the most support of the next-gen web image formats. AVIF quality might be better than WEBP, but very slow to decode and power hog. JPEGXL most superior quality, fast to decode, but no universal support. Thems the beans, folks.


















