How does the image compressor route different formats — JPG, PNG, WebP, BMP?

The tool uses the widely-used browser-image-compression library (Donaldcwl, MIT licence) which detects MIME type and routes to the appropriate pipeline: JPGs go through Canvas toBlob('image/jpeg', quality) which re-encodes via JPEG's DCT + Huffman quantisation per ITU-T T.81 Annex K (lossy); WebPs go through Canvas toBlob('image/webp', quality) which uses VP8 (RFC 6386 Bankoski et al., November 2011) for lossy mode or LZ77 + Huffman/prefix coding for lossless; PNGs bypass Canvas entirely and use UZIP (a JavaScript DEFLATE implementation per RFC 1951 Deutsch, May 1996) for lossless re-compression with potentially better filter selection.

Compress Image — JPG, PNG, WebP Compressor Online

Q: Does PNG compression work the same way as JPG and WebP compression?

No, fundamentally different. PNG (W3C 2nd Edition, 10 November 2003 / ISO/IEC 15948:2004) is lossless by design — every pixel is preserved exactly through DEFLATE re-encoding. The compression gain comes from optimising filter selection (5 filter types per scanline) and zlib compression level, NOT from discarding visual data. Expected savings: 5–30% for typical PNGs (much less than JPG's 40–70%). For palette-quantised lossy PNG (significant size reduction up to 70%) you'd need a tool using libimagequant/pngquant — this tool does NOT do palette quantisation, only lossless DEFLATE re-encoding.

Q: What does the quality slider actually do for each format?

JPG: quality 1–100 maps to JPEG quantisation tables per ITU-T T.81 Annex K. Higher = smaller quantisation steps = more DCT coefficients preserved = larger file. 85–90 visually indistinguishable from lossless; 75–85 web sweet spot. WebP: quality 1–100 controls VP8 quantiser for lossy mode (Firefox 105+ supports lossless via quality=1.0). PNG: quality slider has NO effect — PNG is lossless regardless. Output size depends only on filter selection and zlib level.

Q: Will my image's EXIF metadata survive compression?

Depends on format and configuration. JPG: by default Canvas toBlob() strips EXIF (JEITA CP-3451X / CIPA DC-X008-2019, Exif 2.32, revised 17 May 2019); browser-image-compression has an opt-in 'preserveExif' flag that retains APP1 markers — privacy benefit is the default behaviour (GPS coordinates, camera serial dropped). PNG: native PNG has no widely-supported EXIF chunk; metadata is typically dropped. WebP: supports EXIF chunk per RFC 9649 §2.7.1.5 but Canvas re-encoding strips it.

Q: Does the compressor run my images in the cloud?

No. The tool processes everything locally via the WHATWG HTML Living Standard Canvas 2D Context (for JPG/WebP) or UZIP (for PNG), running in a Web Worker thread when OffscreenCanvas is available. No file leaves your device — verifiable in browser DevTools Network tab (no upload requests fire). Critical for client work under NDA, personal photos with location data, or any image you'd rather not hand to a remote service.

Image Compressor — JPG via DCT, PNG via DEFLATE, WebP via VP8

Image compression spans three fundamentally different algorithms depending on format. JPG (ITU-T T.81 / ISO/IEC 10918-1, 18 September 1992) uses lossy DCT + Huffman quantisation per Annex K — quality 1–100 maps to quantisation table scaling. PNG (W3C Recommendation 10 November 2003 / ISO/IEC 15948:2004) uses lossless DEFLATE (RFC 1951 Deutsch, May 1996, LZ77 + Huffman with 32 KB sliding window) — compression gain comes from filter selection (5 types per scanline) and zlib level, not pixel discarding. WebP (RFC 9649 Zern, Massimino & Alakuijala, Google LLC, November 2024) supports both lossy (VP8 keyframes per RFC 6386 Bankoski et al., November 2011) and lossless (LZ77 + Huffman/prefix coding + colour cache). This tool uses the widely-used browser-image-compression library (Donaldcwl, MIT licence): Canvas-based for JPG/WebP/BMP, UZIP-based DEFLATE for PNG, with Web Worker offloading via OffscreenCanvas when available.

How to compress an image

Drop a JPG/PNG/WebP/BMP file onto the tool or click to browse — single file or batch.
The tool detects format and routes to the appropriate pipeline (Canvas+JPG/WebP, UZIP+PNG).
Adjust quality (effective for JPG/WebP — PNG is lossless regardless). 80–85 is the standard web sweet spot for photos.
Review before/after sizes (typical: JPG 40–70% reduction, WebP 25–35% smaller than JPG, PNG 5–30% lossless gain), then download.

Common use cases

Optimising mixed-format image library (JPG photos + PNG logos + WebP variants) before uploading to a CMS in batch.
Reducing email attachment size below 10 MB / 25 MB limits without resizing — JPG quality 75 typically does the trick.
Cutting Core Web Vitals (LCP) on image-heavy pages by re-encoding all hero images at quality 80–85.
Trimming photo backups before cloud upload (Google Drive, Dropbox, iCloud) — typical 40–60% storage reduction with no visible quality loss.

Frequently asked questions

How does the compressor route different formats — JPG, PNG, WebP, BMP?

Uses the widely-used browser-image-compression library (Donaldcwl, MIT) which detects MIME type. JPGs/WebPs go through Canvas toBlob() per ITU-T T.81 / RFC 9649. PNGs bypass Canvas and use UZIP (JavaScript DEFLATE per RFC 1951) for lossless re-compression with optimised filter selection.

Does PNG compression work the same way as JPG and WebP compression?

No, fundamentally different. PNG (W3C 2nd Ed) is lossless by design — DEFLATE re-encoding only optimises filter selection + zlib level. Expected savings: 5–30% (vs JPG's 40–70%). For lossy PNG palette quantisation (up to 70% savings) you'd need libimagequant/pngquant.

What does the quality slider actually do for each format?

JPG: quality 1–100 maps to ITU-T T.81 Annex K Q-table scaling. 85–90 visually lossless; 75–85 web sweet spot. WebP: quality controls VP8 quantiser (Firefox 105+ supports quality=1.0 lossless). PNG: quality slider has NO effect — PNG is lossless regardless.

Will my image's EXIF metadata survive compression?

JPG: by default Canvas strips EXIF (JEITA CP-3451X / CIPA DC-X008-2019, Exif 2.32, May 2019); browser-image-compression has opt-in preserveExif flag. PNG: native PNG has no widely-supported EXIF chunk. WebP: supports EXIF chunk per RFC 9649 §2.7.1.5 but Canvas re-encoding strips it.

Does the compressor run my images in the cloud?

No. Local processing via WHATWG Canvas 2D Context (JPG/WebP) or UZIP (PNG), running in Web Worker thread when OffscreenCanvas available. No file leaves device — verifiable in DevTools Network tab (no upload requests fire).

How the compressor routes each format — and what the quality slider actually does

The browser-image-compression library detects MIME type and routes to the appropriate pipeline. For JPG: decode to pixels via Canvas → toBlob('image/jpeg', quality) → re-encode via the browser's built-in JPEG encoder using ITU-T T.81 Annex K quantisation tables. The quality slider (1–100) directly scales the Q-tables — higher quality = smaller quantisation steps = more DCT coefficients preserved = larger file. For WebP: same pipeline but toBlob('image/webp', quality) → VP8 keyframe encoding for lossy mode (Firefox 105+ also supports lossless via quality=1.0). For PNG: bypasses Canvas entirely (since Canvas's PNG output is unoptimised) and uses UZIP — a pure-JavaScript DEFLATE implementation by photopea — to re-encode losslessly with optimised filter selection and higher zlib compression level (typically 5–30% smaller than the source PNG, with zero pixel changes). Critical caveat: PNG-to-PNG compression is strictly lossless; the quality slider has NO effect on PNG output. Browser support asymmetry: JPG output works everywhere; WebP output requires libwebp WebAssembly polyfill in Safari (Safari does not support image/webp via canvas.toBlob as of 2026 — a long-standing WebKit limitation tracked at bugs.webkit.org). EXIF metadata: by default Canvas-based re-encoding strips JPEG APP1 EXIF (JEITA CP-3451X / CIPA DC-X008-2019, Exif 2.32, May 2019) — privacy benefit; browser-image-compression has an opt-in preserveExif flag.

JPG re-encoding via ITU-T T.81 Annex K quantisation (quality 1–100 maps to Q-table scaling)
PNG lossless re-compression via UZIP DEFLATE (RFC 1951) with optimised filter selection
WebP re-encoding via VP8 (RFC 6386) for lossy or LZ77 + Huffman for lossless
Real-time before/after size comparison with percentage savings
Web Worker offloading via OffscreenCanvas when supported (non-blocking batch processing)
EXIF metadata stripped by default for privacy (JEITA CP-3451X Exif 2.32) — opt-in preserveExif available

Free. No signup. No file uploads. Ads via AdSense (consent required).

Sources (5)

Donald (Donaldcwl) (live). browser-image-compression. npm package v2.0.2 (github.com/Donaldcwl/browser-image-compression, MIT license) — widely-used JS image compressor: Canvas-based for JPG/WebP/BMP, UZIP-based DEFLATE for PNG; supports OffscreenCanvas + Web Worker.
WHATWG (live). HTML Living Standard — Canvas 2D Context + HTMLCanvasElement.toBlob(). html.spec.whatwg.org/#2dcontext (toBlob('image/jpeg', quality) for JPG re-encoding, toBlob('image/webp', quality) for WebP — universal Chrome/Edge/Firefox 96+, Safari WebP needs polyfill).
ITU-T (CCITT) Study Group VIII & ISO/IEC JTC 1/SC 29/WG 10 (JPEG) (1992). Information technology — Digital compression and coding of continuous-tone still images. ITU-T Recommendation T.81 (18 September 1992) / ISO/IEC 10918-1:1994 — DCT + Huffman quantisation per Annex K underlies JPG quality slider.
W3C (PNG Working Group) (2003). Portable Network Graphics (PNG) Specification (Second Edition). W3C Recommendation 10 November 2003 / ISO/IEC 15948:2004 — PNG uses RFC 1951 DEFLATE (LZ77 + Huffman); browser-image-compression's UZIP path re-applies DEFLATE losslessly with optimised filter selection.
Zern, J., Massimino, P., & Alakuijala, J. (Google LLC) (2024). WebP Image Format. RFC 9649, IETF (November 2024) — WebP container; lossy mode VP8 (RFC 6386 Bankoski et al., Nov 2011) accepts quality 1–100; lossless mode preserves pixels exactly.

These are the W3C, ISO/IEC, ITU-T, and IETF specifications the tool implements or builds on. Locate them on w3.org, iso.org, itu.int, or datatracker.ietf.org.