Most image format comparisons online quote the same theoretical figures: "WebP is 30% smaller than JPEG," "AVIF saves 50%." These come from codec specifications and lab studies, not from converting real files through a real pipeline.

This benchmark is different. Every figure below comes from actual conversions run on Convertify's production server (Rust + libvips) on June 18, 2026. The test set is 26 PNG images (a mix of screenshots, illustrations, and photographs), 24 iPhone HEIC photos straight from the camera, plus JPG, AVIF, and WebP versions derived from those sources. Each number is an average across the full set.

The results confirm some common wisdom and overturn other parts of it. Converting HEIC to JPG, for example, makes files larger, not smaller. Here is what the measurements actually show.

WebP is 64% smaller than JPG at Q80, with near-identical visual quality.

AVIF is 5 to 15% smaller than WebP, but encodes about 7 times slower.

PNG to WebP is the biggest single win at 92% smaller.

HEIC to JPG increases file size by 14%, because HEIC's codec is more efficient than JPEG's.

JPG encodes fastest (0.07s), WebP is the best all-rounder (0.19s), AVIF is slowest but smallest (1.30s).

For most websites, convert to WebP at Q80. For maximum compression where encode time is not a constraint, use AVIF. The full measurements are below.

Starting from 26 JPG images averaging 171 KB (created from PNG sources at Q92 to simulate typical camera output), here is what each modern format produces:

WebP Q80 and AVIF Q63 land in the same neighborhood for file size (61 KB vs 55 KB, a 10% gap), but WebP encodes nearly 7 times faster: 0.19 seconds versus 1.30 seconds per image. For a single conversion this is irrelevant. For a pipeline processing thousands of images, it is the difference between minutes and hours.

AVIF pulls ahead at aggressive compression. AVIF Q50 at 33 KB beats WebP Q60 at 41 KB by 20%. If your priority is the smallest possible file and you can absorb the encode cost, AVIF wins on size. If you want the best balance, WebP at Q80 is the safer default.

One result worth flagging: converting JPG to lossless WebP inflates the file by 177%, from 171 KB to 474 KB. Lossless compression has to encode a full pixel grid, and reconstructing that from already-lossy JPEG data produces a larger file than the lossy original. Lossless WebP only makes sense when the source is itself lossless, such as PNG or TIFF.

PNG is lossless and often large, which makes it the best candidate for aggressive conversion. Starting from 26 PNG images averaging 742 KB:

PNG to WebP and PNG to AVIF both hit 92% file size reduction at comparable visual quality. The separator is speed: WebP finishes in 0.21 seconds, AVIF takes 1.49 seconds, roughly 7 times longer.

PNG to JPG at Q85 delivers an 86% reduction in just 0.07 seconds, the fastest conversion in the entire benchmark. The cost is losing transparency and accepting lossy compression. For photographic PNGs with no alpha channel, JPG is the quickest route to small files.

Lossless WebP shrinks PNG by 32% while preserving every pixel exactly. This is the option when you need web-compatible images with no quality loss: WebP lossless works in all modern browsers and is consistently smaller than the equivalent PNG.

iPhone photos arrive as HEIC, and the common advice is to "convert them to JPG for compatibility." The measurements show why that advice costs you storage. Starting from 24 real iPhone HEIC photos averaging 2,557 KB:

Converting HEIC to JPG increases file size by 14% on average in this test. The reason is codec efficiency. HEIC wraps the HEVC codec, which compresses roughly twice as efficiently as JPEG's older DCT-based method. Decoding HEIC and re-encoding as JPG moves the image from a more efficient codec to a less efficient one, so the file grows.

If the goal is smaller files, JPG is the wrong target. HEIC to WebP cuts size by 43%, and HEIC to AVIF cuts it by 57%. JPG only makes sense when you specifically need compatibility with software that cannot read HEIC, and the 14% size increase is the price of that universal support.

HEIC to PNG produces a startling 12-fold size increase, from 2.5 MB to 31 MB. PNG stores every pixel losslessly, and a 12-megapixel photo holds roughly 36 MB of raw pixel data. PNG compresses that to about 31 MB, but it is still enormous compared to the HEVC-compressed original. Convert HEIC to PNG only when you genuinely need lossless quality for editing or print.

One caveat on timing: HEIC to AVIF averaged 14.52 seconds per photo, by far the slowest conversion measured. This combines slow HEIC decoding with slow AV1 encoding on full-resolution camera images. For batch HEIC conversion, WebP is the practical choice.

Sometimes you receive AVIF or WebP files and need JPG for an older system. These conversions always increase file size, because you are moving back to a less efficient codec:

AVIF to JPG grows the file by 80%, WebP to JPG by 60%. The outputs land close together (95 to 96 KB) because both sources are similar in size and both target the same JPG encoder. The conversions are fast, both finishing in under 0.2 seconds, and they do not lose information beyond what JPG encoding normally discards.

AVIF to PNG inflates by more than 12 times, mirroring the HEIC to PNG result for the same reason: PNG reconstructs a full lossless pixel grid from a heavily compressed source. Avoid this conversion unless you specifically need lossless PNG output.

Rasterizing a PDF page to an image format is not compression, it is rendering the page at a chosen resolution. Higher DPI means more pixels and larger files. Here is a single-page PDF rendered to three formats at four DPI settings:

PDF to WebP is 58 to 76% smaller than PDF to JPG, and the gap widens as DPI rises. At 600 DPI, print quality, a single page is 4.2 MB as JPG but only 1 MB as WebP.

PDF to PNG is consistently the largest output because it is lossless. At 600 DPI a single page reaches 17 MB as PNG. Reserve PNG for cases that demand pixel-perfect rasterization, such as OCR preprocessing or archival.

For everyday use, 300 DPI with WebP output is the sweet spot: sharp on any screen, small enough to email or upload. 150 DPI is fine for screen-only viewing and produces files about 70% smaller than 300 DPI.

File size is only half the picture. Encoding time determines whether a format is practical for batch work. Average encode time across the test images, measured as wall-clock round-trip time:

AVIF compresses 5 to 15% better than WebP at similar visual quality, but encodes about 7 times slower. For one image, the difference between 0.2 and 1.3 seconds is unnoticeable. For a batch of thousands, it is decisive.

JPG encoding is fastest by a wide margin, which makes it the right pick when speed beats size: real-time processing, live thumbnail generation, or legacy compatibility.

For most workflows, WebP at Q80 is the best general choice: 64% smaller than JPG, 7 times faster than AVIF, and supported in every modern browser since 2020. Reach for AVIF when you control the delivery pipeline and want the smallest files, and you can encode ahead of time rather than on demand.

All conversions ran on June 18, 2026, on Convertify's production server, the same Rust + libvips pipeline that handles user uploads. The test set was 26 PNG images (screenshots, illustrations, and photographs), 24 iPhone HEIC photos, and derived JPG (Q92), AVIF (Q63), and WebP (Q80) inputs created with the libvips command-line tools.

Timings are wall-clock round-trip on localhost, which includes HTTP parsing but no network latency, so real user times will be slightly higher because of transfer. File sizes are reported by the backend after conversion and verified against disk.

The pipeline runs libvips 8.15 with libjpeg-turbo for JPEG, libwebp for WebP, libaom for AVIF, libheif with libde265 for HEIC, and poppler for PDF rasterization. Metadata is stripped from every output. You can reproduce any figure here by converting the same kinds of files through the corresponding Convertify tool: JPG to WebP, PNG to WebP, HEIC to WebP, or PDF to WebP.