The JPEG quality setting, decoded

The quality slider is a UI on top of a table

When you drag a JPEG quality slider, you are scaling the entries in an 8×8 quantization table. Higher quality means smaller table values, which means less aggressive rounding in the lossy step of the encoder. Lower quality means larger table values, which means more zeros after quantization and a smaller file.

The deep walkthrough lives in the JPEG compression explainer. What you actually need for day-to-day work is a clear mapping from slider numbers to expected file size and quality, plus a feel for where the steep parts of the curve are.

Why there is no universal scale

Every image editor maps its slider to the underlying quantization table in its own way. Photoshop uses 0–12. libjpeg, the reference open-source encoder, uses 0–100. Adobe Camera Raw uses 1–12. WebP, when used as a JPEG replacement, uses 0–100 but with a completely different curve.

The same numeric setting in two different tools produces different files. Photoshop quality 12 is not the same as libjpeg quality 100, and "quality 80" in WebP is not equivalent to "quality 80" in JPEG.

For the next sections, "quality 80" means libjpeg quality 80 unless noted, since libjpeg is the closest thing to a baseline.

The actual quality-vs-size curve

For a typical 3000×2000 photograph, encoded with libjpeg using default 4:2:0 chroma subsampling, file sizes look roughly like this:

The curve is steep on both ends. Going from quality 100 to 90 cuts the file size by more than 4× with no visible cost. Going from 75 to 60 cuts maybe a quarter off the size for a noticeable hit to quality. Below about 50, the file barely shrinks any further but the artifacts get rapidly worse.

The practical implication: for almost any use case, the right setting is somewhere between 75 and 90. Below 70 is a deliberate quality compromise. Above 90 is an archival or print decision, not a web decision.

Photoshop's scale and how to translate

Photoshop maps its 0–12 slider to underlying quality factors roughly like this:

Save for Web uses a different mapping, slightly more aggressive at the high end and gentler at the low end, because it is tuned for web delivery rather than archival.

Adobe Camera Raw's 1–12 scale is closer to Photoshop's, with the same caveats.

Per-use-case recommendations

Photos for the web, hero images, blog illustrations. libjpeg 80–85, or Photoshop quality 8. WebP at quality 80 is a strictly better option if your audience uses modern browsers; the file is typically 25%–35% smaller at the same visual quality.

Photos for print. libjpeg 95 minimum, or save as PNG or TIFF if you are doing further editing. JPEG quality below 90 starts showing up as banding in midtones once your image is rasterized by a high-DPI printer.

Photos for archival. Do not use JPEG. Use the camera RAW or save as a 16-bit TIFF or lossless WebP. JPEG is a delivery format, not a storage format. If you must use JPEG, quality 95 with 4:4:4 chroma is the closest you can get to lossless.

Email attachments. libjpeg 75 is the standard "looks fine, fits in an email" sweet spot. Most mail servers will silently re-encode anything bigger anyway.

Thumbnails. libjpeg 70 with 4:2:0 chroma. Thumbnails are small enough that artifacts are mostly invisible. Lower quality saves bandwidth and load time.

Screenshots, line art, logos. Do not use JPEG at all. Use PNG or lossless WebP. JPEG was designed for photographic gradients and produces terrible artifacts on hard-edged content at any quality below 95. The format comparison post covers what to use instead.

Meme aesthetic, deep-fried look, vaporwave. Cranking the quality slider down is the entire point. The JPEG Artifact Generator gives you fine-grained control over quality plus chromatic aberration and noise; the deep-fried meme guide covers the cultural side.

The role of chroma subsampling

The quality slider is one knob. Chroma subsampling is another, and most encoders tie them together unhelpfully.

By default, most encoders use 4:2:0 chroma (one chroma sample per four luma samples). At quality 90 and above, many encoders switch to 4:4:4 (full-resolution chroma) automatically. This is why a Photoshop quality 12 file can be substantially larger than a quality 11 file: not because the quantization changed much, but because the encoder also disabled chroma subsampling.

If you are saving graphics with saturated colors, text rendering, or fine color detail, 4:4:4 chroma matters more than the exact quality number. Most professional tools let you control chroma independently. mozJPEG and squoosh.app expose this directly.

Tools that help

For batch processing, jpegoptim or mozJPEG's cjpeg command is the standard. For one-off comparisons, squoosh.app lets you see the trade-off directly with a draggable comparison.

What about "JPEG progressive"?

A progressive JPEG and a baseline JPEG encode the same coefficients with the same quantization. The only difference is the order they are written. Progressive saves are a few percent smaller on average and load with a perceived speed advantage on slow connections (you see a blurry preview while the file is still arriving).

There is no quality difference. If your tool offers progressive, turn it on. mozJPEG defaults to progressive.

FAQ

What is the lowest "safe" JPEG quality?

For photos, libjpeg 75 (Photoshop 7) is the floor where most viewers do not notice damage. Below that, artifacts start being obvious on detailed content. For screenshots and graphics, JPEG is the wrong format at any quality.

Does saving at 100 give me a lossless file?

No. JPEG is always lossy because the DCT-and-quantize pipeline introduces small rounding errors even at the highest quality setting. Quality 100 makes those errors tiny enough to be visually invisible, but the file is not byte-identical to the original. For true lossless, use PNG, lossless WebP, or JPEG XL.

Why does saving the same image at the same quality produce different file sizes?

Encoders make small choices that differ across versions and platforms. The Huffman tables might be optimized for the specific image, the chroma subsampling might switch behavior at certain quality thresholds, and metadata (Exif, ICC profile, thumbnail) varies. Variations of 5–15% at the same nominal quality are normal.

Should I use WebP instead?

For new content delivered over the web in 2026, yes. Browser support for WebP is universal. At the same perceived quality, WebP files are about 25%–35% smaller than JPEG. The compression artifacts are also slightly less visually objectionable when they appear. AVIF is even smaller but the encoder is slower and tooling is less mature.

Can I lose quality just by opening and resaving a file?

Only if you re-encode it. Most image viewers (Preview, the Photos app, browsers) decode the JPEG for display but do not re-save it. Editors that re-save on every operation accumulate damage. This is generational loss, covered in the generational loss post.

How does the quality slider in cjpeg relate to Photoshop's?

cjpeg -quality 80 is roughly equivalent to Photoshop's quality 8 (High). Each tool maps its slider differently, but libjpeg's 0–100 is the closest thing to a baseline scale, and most quality comparisons online are reported in libjpeg numbers.

Sources

• ISO/IEC 10918-1:1994 / ITU-T Recommendation T.81. The quality-factor curve is defined informally in Annex K.
• libjpeg-turbo, the standard reference encoder used in most software.
• mozJPEG, Mozilla's optimized fork.
• squoosh.app, Google Chrome team's open-source compression comparison tool.
• Alakuijala, J. et al. (2017). Guetzli: Perceptually Guided JPEG Encoder, arXiv:1703.04421. Background on perceptual quality metrics for JPEG.