JPEG Quality 70 vs 85 vs 95: Which Setting Should You Actually Use?

The JPEG quality slider is one of the most misunderstood controls in image editing. People treat it like a percentage — "95% quality, so only 5% is missing" — but that isn't what the number means. It's a scaling factor applied to the quantization tables defined in the JPEG standard, and the relationship between that number and visible quality is steeply non-linear. This guide walks through what quality 70, 85, and 95 actually do, shows you the file-size curves, and gives concrete recommendations for the content types you're most likely to save.

1. What the JPEG quality slider actually controls

JPEG compression works by splitting the image into 8×8 pixel blocks, running a discrete cosine transform on each block, and then quantizing the frequency coefficients — dividing them by numbers in a quantization table and rounding the results. Low-frequency information (broad tonal shapes) gets smaller divisors so more of it survives. High-frequency information (sharp edges, fine texture) gets larger divisors, so much of it gets rounded to zero and discarded.

The quality slider doesn't represent a percentage of information preserved. It's a single number that scales those quantization tables. At quality 50, the tables from the JPEG standard (Annex K) are used as-is. At quality 100, every entry is scaled down toward 1 (minimal quantization). At quality 10, every entry is scaled up dramatically, throwing away most high-frequency detail. The mapping is defined by libjpeg's formula, which almost every encoder inherits — though with tweaks (see the FAQ about Photoshop vs Squoosh).

This matters because the relationship between quality number and perceived quality isn't proportional. Going from 95 to 85 discards a lot of data that your eyes won't miss. Going from 55 to 45 discards much less data in absolute terms but crosses a perceptual cliff where artifacts become obvious. Treat the slider as a set of labeled presets, not a linear measurement.

2. Quality 95 — the "keep the file for re-editing" setting

At quality 95, the quantization tables are scaled down aggressively, so very little information is discarded. The output is visually near-lossless: on normal displays at normal zoom, you cannot distinguish a quality-95 JPEG from the uncompressed original. Round-tripping through quality 95 a couple of times is still safe enough for most practical purposes, though you'll eventually accumulate generational loss if you keep re-saving.

The trade-off is file size. A quality-95 JPEG is typically 3–5× larger than the same image at quality 80, and you get essentially no additional visible quality for the extra bytes. That makes quality 95 the wrong setting for web delivery — you're paying four times the bandwidth cost for zero perceptible benefit.

Use quality 95 when: you're saving a master copy you expect to re-edit, submitting to a stock photography site with strict quality requirements, or when the JPEG is going to be re-encoded downstream by a service you don't control (and you want to give them the best possible input).

3. Quality 85 — the standard web default

Quality 85 is the quiet consensus default across almost every serious image pipeline. Photoshop's "Save for Web" defaults to 85 on its 0–100 scale (level 8 on the 0–10 scale). WordPress ships with a default JPEG quality of 82. Shopify re-encodes uploaded JPEGs at quality 85. Squoosh defaults to 75 for MozJPEG but 85 is the most common recommendation in Google's own web.dev image-optimization documentation. Adobe's own guidance for web export lands in the 75–90 range depending on content, with 85 as the middle.

At quality 85, the output is typically indistinguishable from quality 95 at 100% zoom on photographic content, while being roughly 30–50% smaller. That's the entire reason 85 is the default: you keep all the visible quality, and you pay half the bytes. The only places the difference shows up are on extreme zoom, on flat skies where 8×8 block boundaries can become faintly visible, and on fine text inside the image.

Use quality 85 when: you're saving JPEGs for the web, for a blog post, for product photography on a storefront, or anywhere the file will be viewed rather than further edited. If you only remember one number from this article, make it 85.

4. Quality 70–75 — the aggressive web budget

Quality 70–75 is where you start to pay a visible cost, but in exchange for major size savings. Relative to quality 85, you'll typically save another 25–40% in file size. The trade-off is that fine detail begins to soften, small coloured text inside the image can bleed, and very gentle gradients (blue skies, skin tones) may start to show faint banding at pixel level.

For most photographic content viewed at normal size, 75 is still perfectly acceptable. MozJPEG — the encoder used by Squoosh and most modern web image pipelines — is tuned to squeeze additional fidelity out of the same quality numbers, and at MozJPEG quality 75 you often get output that rivals libjpeg quality 85 on file size with comparable visual quality.

Use quality 70–75 when: you're saving thumbnails, generating heavy-traffic images where every kilobyte counts, sending chat or messaging attachments, or producing large image grids on a single page. If your site is bandwidth-constrained or you have hundreds of images on a page, this is the tier to consider.

5. Below quality 50 — where artifacts become obvious

Under quality 50, the characteristic JPEG artifacts become visible without needing to zoom. You'll see blockiness — the 8×8 quantization grid becomes perceptible as a faint checkerboard, especially in smooth gradients. You'll see ringing — ghostly halos around sharp edges, caused by truncated high-frequency coefficients. And you'll see colour banding, particularly in dark areas and shadows, where the reduced chroma precision rounds neighbouring pixels to the same value.

Quality 30–40 is sometimes intentionally used for stylistic reasons (the "deep fried" meme aesthetic), but for normal content it's below the threshold of acceptable. If you find yourself pushing quality this low to fit a size budget, the right move is usually to downscale the image resolution first, then re-encode at quality 75–85, rather than keeping the resolution and cranking quality down.

6. File-size vs quality curves are NOT linear

This is the point most tutorials miss. Dropping JPEG quality doesn't save a linear amount of bytes per point — the savings follow a strongly diminishing curve, which is exactly why quality 85 is such a sweet spot. Here's what a typical 1920×1280 photograph looks like across the common settings:

Actual numbers vary ±25% with content, but the shape of the curve is consistent: the savings from 95 → 85 are dramatic (−42%), the next step (85 → 75) is smaller, and each successive step returns less size savings for each point of quality sacrificed. This is why quality 85 is the pragmatic default — you've captured almost all the compression gains that matter, and pushing further costs visible quality for ever-smaller byte savings.

7. Content-dependent recommendations

JPEG quality defaults are always a starting point — the right number depends on what's in the image. Here's a practical guide by content type:

• Photographs (portraits, landscapes, products) → quality 82–88. JPEG was designed for this content and handles it gracefully. Quality 85 is the safe default.
• Screenshots with text or UI elements → don't use JPEG at all. Use PNG or WebP lossless. JPEG's 8×8 block compression and chroma subsampling make text look smeared and halated. If you must use JPEG, push quality to 92+ and turn off chroma subsampling (4:4:4).
• Flat illustrations, vector exports, logos → same as above — JPEG is the wrong format. If the file must be JPEG for compatibility, use quality 90+ with 4:4:4 subsampling.
• Noisy or grainy content (film scans, high-ISO photos, textured surfaces) → quality 78–82. Grain structure actually helps hide JPEG artifacts, so you can push quality a bit lower than you'd normally use on a clean image.
• Thumbnails (under 400px) → quality 70–75. The reduced resolution already hides most of the detail JPEG discards.

Frequently asked questions

References

• ISO/IEC 10918-1 — JPEG (the original 1992 standard, including Annex K quantization tables)
• MozJPEG project — github.com/mozilla/mozjpeg (perceptually-tuned libjpeg fork used by Squoosh)
• libjpeg-turbo documentation — libjpeg-turbo.org (reference encoder; quality slider formula originates here)
• Squoosh.app research — Google Chrome Labs, comparative image-encoder benchmarks
• Adobe — Save for Web defaults and JPEG export settings (Photoshop documentation)
• web.dev — Use imagemin to compress images (Google's web performance guide)