JPEG remains the default format for camera output and everyday photo sharing because its lossy compression can shrink a typical photo to roughly one-tenth of its uncompressed size while looking nearly identical to the eye at normal viewing distance. What happens to the file size is a negotiation between the JPEG quality setting, the image's actual detail, and chroma subsampling, and small changes in quality can swing the saved file dramatically.

Think of a 24-megapixel photo straight off a mirrorless camera: sensor data at 14 bits per pixel runs to tens of megabytes before any processing. JPEG is the mechanism that turns that into a 4–10 megabyte file you can email, upload, or fit a thousand of on a memory card. Its popularity isn't an accident of history alone; it's a set of engineering tradeoffs that happened to align with how human vision actually works.

How JPEG Compression Actually Discards Data

JPEG doesn't throw away pixels randomly. It converts each 8Γ—8 pixel block from RGB into a luma/chroma color model, then runs a discrete cosine transform that expresses the block as a sum of spatial frequencies. High-frequency terms β€” the fine detail β€” get divided by larger quantization values, and many of those coefficients round to zero, which is where the data disappears.

Human vision is far more sensitive to brightness detail than to color detail, so JPEG stores luma at full resolution and chroma at reduced resolution through a process called chroma subsampling. Common camera settings like 4:2:0 keep one color sample for every four luma samples, cutting chroma data to a quarter. That single decision accounts for a large share of the size reduction before any quality slider moves.

The compression is lossy and irreversible. Once those high-frequency coefficients quantize to zero, no decoder can recover them, which is why re-saving a JPEG at high quality cannot restore detail. Each subsequent save re-quantizes an already-degraded image, and generation loss accumulates.

A useful mental model: quality 100 doesn't mean lossless. It means the quantization tables are scaled down toward 1, minimizing but never eliminating loss. Adobe's Photoshop documentation has long described its quality scale as 0–12, with 12 the maximum, and even at 12 the file is still lossy.

Why Small Quality Changes Swing File Size So Hard

The quantization step is the single biggest lever on output size, and its effect is nonlinear. Doubling the quantization values doesn't double compression β€” it collapses whole bands of coefficients to zero, so file size can drop by half or more for a modest visual change. This is why a quality 95 export and a quality 80 export of the same photo can differ by threefold in bytes.

Image content matters as much as the slider. A photo of a clear blue sky, a smooth wall, or a blurred background compresses to a fraction of the size of a detailed forest canopy or a crowd scene at identical settings, because those smooth regions have few high-frequency coefficients left to store. Camera noise is the hidden cost: every speck of sensor noise is real high-frequency detail JPEG must preserve, so a noisy ISO 6400 shot can produce a file twice the size of a clean ISO 100 shot of the same scene.

Consider a worked example. A 24-megapixel image with roughly 24 million pixels at 3 bytes per pixel in uncompressed RGB is about 72 megabytes raw. At a quality setting that yields a 10:1 compression ratio, the JPEG lands near 7 megabytes. Push to a 20:1 effective ratio through a lower quality setting or smoother content, and you're near 3.6 megabytes. That range β€” 3.6 to 7 MB from the same sensor and lens β€” is the practical everyday spread.

Quality setting (typical) Approx. compression vs. uncompressed 24 MP file size (derived) Visible artifact at 100% zoom
100 (Photoshop 12) ~5:1 ~14 MB Minimal
90 ~10:1 ~7 MB Faint blocking in high-detail areas
80 ~15:1 ~4.8 MB Softening of fine texture
60 ~25:1 ~2.9 MB Obvious 8Γ—8 block edges
40 ~40:1 ~1.8 MB Visible ringing and color blotches

The compression ratios above are derived from the 72 MB uncompressed baseline, not measured on a specific camera; treat them as order-of-magnitude guidance and confirm on your own files. The key point stands: the file size relationship to quality is steep and content-dependent.

The Chroma Subsampling Setting Most Photographers Ignore

Chroma subsampling is a separate dial from the quality slider, and many cameras and editors expose it without fanfare. The common modes are 4:4:4 (no subsampling, full color resolution), 4:2:2 (half horizontal chroma), and 4:2:0 (quarter chroma). The notation describes how many chroma samples exist per luma sample block, and moving from 4:4:4 to 4:2:0 typically cuts file size by 20–30 percent on its own.

The visible cost of 4:2:0 shows up at sharp red or magenta edges β€” text, thin colored lines, saturated signs. It's usually invisible in natural scenes at normal viewing size, which is exactly why it became the default. Cameras from Canon, Nikon, Sony, and Fujifilm generally write 4:2:0 for standard JPEGs and reserve 4:4:4 or 4:2:2 for higher-end or RAW-adjacent modes.

For a straightforward workflow: keep 4:2:0 when the destination is a screen, a photo book at normal size, or social media; switch to 4:4:4 or 4:2:2 when the image contains fine colored type, product shots with saturated packaging, or line art that will be enlarged. Editing a JPEG after capture is the other case β€” subsampled chroma is harder to push in saturation and color grading without artifacts.

A subtle edge failure worth knowing: repeatedly opening a 4:2:0 JPEG, editing, and re-saving at 4:2:0 can stack chroma subsampling on top of itself, producing color fringing around high-contrast edges that didn't exist in the original. Save the master at 4:4:4 if you plan multiple rounds.

What the 8Γ—8 Block Structure Does to Your Image at High Compression

JPEG divides the image into 8Γ—8 pixel tiles, and the discrete cosine transform operates on each tile independently. At low quality, adjacent tiles can reconstruct with slightly different average brightness, producing the grid of visible squares known as blocking artifacts. That's not a bug; it's the direct consequence of independent block coding.

There's a second flaw called ringing or mosquito noise, where sharp edges in a block create oscillating halos because the limited set of cosine terms can't represent a true step edge. These halos cluster around high-contrast boundaries β€” tree branches against sky, text on a sign β€” and get worse as quality drops.

The block grid is also why JPEG resists certain edits. Cropping and then compressing again can misalign a new 8Γ—8 grid with the original, and heavy sharpening amplifies both blocking and ringing because sharpening boosts exactly the high frequencies quantization destroyed. Rotating a JPEG by non-multiples of 90 degrees forces a resample, which softens the image.

For archival workflows that survive repeated editing, this block structure is the argument for shooting RAW and exporting a JPEG only at the end. For everything else β€” sharing, printing at normal sizes, uploading β€” the block structure rarely shows, which is the tradeoff JPEG was designed around.

Why Cameras and Phones Still Write JPEG First

JPEG's dominance is a compatibility story. Nearly every device, operating system, browser, printer, photo kiosk, and image service on earth decodes it without additional software, a level of ubiquity no newer format has matched. The format dates to the Joint Photographic Experts Group's work standardized in 1992 by ISO/IEC 10918, and that institutional backing plus three decades of embedded support make it the safe universal default.

Cameras output JPEG because it's the fastest path from sensor to shareable file, and because JPEG files open instantly on a phone or tablet with no conversion step. Professional bodies like the Associated Press and most news wires specify JPEG delivery for exactly this reason: the receiving desk can use the file immediately.

Modern phones add a twist. Many Android and iOS devices save JPEG but apply computational processing β€” multi-frame stacking, HDR tone mapping, sharpening β€” before encoding, so the "JPEG" is a processed output rather than a raw sensor read. Apple and Google also offer HEIC/HEIF as a higher-efficiency alternative, but the shareable export frequently reverts to JPEG for compatibility with older systems and email.

The version distinction matters: there are no meaningful "JPEG versions" in the way there are numbered standards like H.264 and H.265. What shifts is the encoder and the settings β€” the JPEG standard itself has been stable since 1992, with only incremental extensions like arithmetic coding and JPEG 2000 as separate formats.

Where JPEG Falls Short: Cases That Break the Format

JPEG handles photographs superbly and handles certain other content badly. Images with flat color β€” logos, screenshots, line art, charts β€” grow visible ringing around text and sharp boundaries even at high quality, and the format can't represent a sharp two-color edge efficiently. PNG or WebP are the right calls for that content.

Images needing transparency can't use JPEG at all; the format has no alpha channel. Images needing extreme enlargement β€” a small JPEG blown up to billboard size, or a crop pushed far past 100 percent β€” reveal block structure and soften quickly, because the high-frequency data is gone permanently.

Repeated editing is the quiet killer. Every save re-quantizes, and ten generations of editing a JPEG at quality 90 can leave visible mush compared to the original, while the same edits on a TIFF or RAW intermediate stay clean. Photo restoration professionals treat JPEG as a delivery format, not an editing format, for this reason.

The other failure case is overconfidence in "high quality." A smartphone exported at quality 100 can still be softer than a properly processed JPEG at quality 90 from a larger sensor, because file size and quality aren't the same thing β€” a point that trips up people who assume the biggest file wins.

When a Smaller File Size Costs You Later

Compression is a bet that the discarded detail won't matter. It's worth naming when that bet loses. If a photo will be printed larger than roughly A4 or 8Γ—10 inches, the high-frequency loss from aggressive compression starts to show as softness in fine texture, especially in foliage, fabric, and hair. Print workflows typically want JPEG quality 90–95 or a lossless format.

If a photo is evidence β€” insurance documentation of damage, a real-estate record, a scientific capture β€” the compression artifacts and the inability to recover original detail become a liability. Capture at the highest quality the camera offers, or use RAW.

If the image is the start of a composite, a crop-heavy edit, or a color grade, JPEG's block grid and subsampled chroma put a ceiling on how far the edit can go. Working from RAW and exporting once at the end avoids compounding the loss.

Storage economics often push the other way. A memory card the size of a postage stamp holds thousands of JPEGs but far fewer RAWs, and cloud backup costs scale with total bytes, so a 10:1 compression ratio frees roughly nine-tenths of the storage a RAW workflow consumes. That single ratio is the core argument for JPEG in volume capture.

Re-saving, Re-exporting, and the Generation Loss Trap

Every save of a JPEG re-encodes the already-quantized data, which means the second save quantizes coefficients that were already rounded. The damage compounds fastest at low quality settings; saving at quality 90 twice is nearly invisible, but saving at quality 60 ten times can visibly destroy detail.

A practical rule: edit from a lossless or RAW intermediate, and export a JPEG exactly once for delivery. If a JPEG is all you have, minimize saves by doing all edits in a single session and exporting once, and choose a high quality setting β€” quality 90 or above β€” for that final export. Avoid quality 100 as a default; the file grows large with little visible gain over 90–95 in typical photographs, because it mostly preserves noise.

Re-exporting from one platform to another also matters. Many social platforms re-compress uploads, so a file already saved at quality 70 that gets platform-compressed again lands closer to 40 in effective quality. Uploading at quality 85–90 gives the platform room to compress without visible damage.

A quick test you can run: export the same photo at quality 100, 90, and 80, then compare at 100 percent zoom on a decent monitor. You'll likely see the 100 and 90 versions side by side with no clear winner, while the 80 starts to soften texture. That confirms 90 as a sensible delivery default without needing a lab.

Matching the Quality Setting to the Destination

The right compression level depends entirely on where the image is going. For email attachments, quality 80–85 keeps a 24-megapixel image near 4–5 MB, well within most mail servers' limits, and looks fine on a laptop screen. For social media, quality 85–90 accounts for the platform's own re-compression and preserves more headroom.

For print, quality 90–95 or a lossless format is the safe zone; the increments are cheap because the print pipeline and the eye both benefit from retained fine detail. For archival delivery alongside RAW, quality 95 or higher, ideally with minimal subsampling, gives a faithful reference file without the RAW's storage cost.

There's a cost computation worth doing once. If you store 10,000 photos and each JPEG averages 6 MB, that's 60 GB. The same set as RAW at roughly 30 MB each would be 300 GB β€” five times the storage and a much slower cloud sync. Saving to quality 80 instead of 90 could cut the 60 GB to roughly 40 GB, a 20 GB reduction, at the cost of texture in large prints. The exact savings depend on content, since smooth scenes compress further than detailed ones.

For a photographer shooting events and delivering web galleries, the workflow that holds up is RAW capture, single export at quality 90 with 4:2:0 for web galleries below 2000 pixels on the long edge, a separate quality 95 or 4:4:4 export for print, and no re-saving of the web files. A gallery of 500 images at 1.5 MB each lands near 750 MB β€” easily hosted and fast to load.

FAQ

Is JPEG lossless at quality 100?

No. Quality 100 scales the quantization tables to their smallest values but still quantizes, so a tiny amount of detail is permanently discarded. Adobe's Photoshop quality scale runs 0–12, and even 12 is lossy. If you need truly lossless, use PNG, TIFF, or a lossless JPEG 2000 mode, and confirm the format's spec on the maker's or standards body's site.

Does a bigger JPEG file always mean better quality?

Not necessarily. File size grows with image noise, texture, and fine detail as much as with quality setting, so a noisy phone photo at quality 100 can be larger than a clean camera photo at quality 90 and still look softer. Judge quality by comparing at 100 percent zoom, not by file size alone.

Why did my JPEG get bigger after I cropped it?

Cropping can misalign the 8Γ—8 block grid and force the encoder to re-quantize the full cropped area at your chosen setting, and heavy sharpening amplifies high-frequency coefficients that cost bytes. The result can exceed the original file size despite fewer pixels. Export at a slightly lower quality or unsharpened for delivery.

Can I recover original detail by re-saving at quality 100?

No. Quantized coefficients are gone; re-encoding at 100 cannot recreate them and only adds bytes. The only path to original detail is working from the RAW or a lossless intermediate saved before the first JPEG export.

What's the difference between JPEG quality and chroma subsampling?

Quality controls how aggressively the luma and chroma coefficients are quantized; chroma subsampling separately reduces the color resolution by storing fewer chroma samples per luma sample. 4:2:0 cuts file size roughly 20–30 percent versus 4:4:4 on its own, and the two settings interact, so check both in your export dialog.

Is HEIC replacing JPEG for phone photos?

HEIC/HEIF offers better compression per unit of visible quality and is the default save format on many recent iPhones and some Android devices, but compatibility is the reason JPEG persists: many email clients, older systems, and uploading services still accept JPEG more reliably. Many phones therefore export JPEG when sharing.

How many times can I re-save a JPEG before it degrades visibly?

It depends on the quality setting. At quality 90, several generations are near-invisible; at quality 60, three or four saves can produce obvious blocking and ringing. The safe rule is one export from a lossless intermediate; if you must re-save, keep the setting high and compare at 100 percent zoom.

What's the largest JPEG file size I can safely email?

Most email providers cap attachments around 20–25 MB, and base64 encoding inflates the effective size by roughly a third, so keep individual JPEGs under about 10–15 MB and use a gallery link for larger batches. A 24-megapixel photo at quality 80–85 typically lands near 4–5 MB and clears every common cap.