Image Conversion Mistakes That Ruin Quality

Converting image formats looks dead simple — you are just turning A into B, right? Then you actually do it, and the result comes back soft, or the colors drift, or a perfectly clean photo suddenly grows color blocks. This piece walks through every trap I have fallen into. It is all hard-won practice, so if you work with images too, you can skip the rework.

Why Does Quality Vanish the Moment You Convert?

Start with the core issue: most common formats rely on lossy compression, and every single save throws information away. Save a JPEG as a JPEG again — quality slider cranked to 100 — and you still introduce fresh damage. Miss this mechanism and nothing below will make sense.

Three Classic Recompression Traps

Trap 1: Generational Loss

Every lossy encoding pass sheds a little more detail. JPEG to JPEG piles up the damage even when the settings never change. It is like photocopying a photocopy — each generation is a bit mushier than the last.

Trap 2: Different Formats, Different Math

JPEG uses DCT (discrete cosine transform) while WebP uses predictive coding. The algorithms are not comparable, so conversion cannot map cleanly between them; all you can do is "decode, then re-encode," and that round trip costs you twice.

Trap 3: Quantization Error Accumulates

Quantization always introduces rounding error, and once you convert enough times the accumulated error becomes visible to the naked eye.

So the advice is blunt: start from the original whenever you can (RAW, or lossless PNG/TIFF), and never chain one lossy format into another. If you truly must convert, push the quality slider to maximum. And if the format you are holding is just inconvenient, run it through the Image Format Converter once instead of bouncing it around several tools.

The Zero-Loss Route: A Lossless Conversion Checklist

Is there a path with no loss at all? Yes. Lossless conversion is a zero-loss channel, and every route below is safe:

• PNG to WebP (lossless mode): not a single pixel changes, and the file actually gets smaller.
• PNG to AVIF (lossless): also zero loss, with competitive compression.
• BMP to PNG: lossless to lossless — convert freely.
• Uncompressed TIFF to PNG: same story, no worries.

JPEG's Special Lossless Operations

JPEG is a special case. Its lossless rotation runs through jpegtran, which only reorders MCU blocks instead of re-encoding. Cropping along MCU boundaries (usually 8 or 16 pixels) is lossless too.

Read the caveat carefully though: JPEG to PNG is lossless, yet everything the JPEG already discarded stays discarded and the file balloons. Do not expect this step to "improve" quality.

My Workflow: Keep the Original as the Master

Here is what works for me: move to a lossless format early in the workflow, convert to the lossy target only at final output, and keep the original as your master. No matter how many rounds of edits follow, you always start from the master, so loss happens exactly once.

Lossy Conversion: Setting the Quality Parameter

Back to lossy conversion — pick the right quality number and you get the balance between file size and fidelity:

• JPEG: 75-85 for the web, 90-95 for print. Drop below 70 and blocking artifacts show up immediately.
• WebP: q75 lands roughly at JPEG q85 visually, at a smaller file size.
• AVIF: quality runs 1-63 where lower means higher quality; 30-40 is the web recommendation. CRF 23-32 is common for the web, but encoding is genuinely slow — 10 to 50 times slower than JPEG — so it is not built for real-time conversion.

Stop Guessing: Measure Instead

Do not eyeball quality; put a number on it. SSIM above 0.95 is essentially indistinguishable, DSSIM under 0.015 is acceptable, and Butteraugli under 1.0 goes unnoticed. For an objective verdict, run ffmpeg with -lavfi ssim or Netflix's VMAF — SSIM above 0.95 and VMAF above 90 means almost nobody will spot the degradation.

Adaptive Quality: More Bits Where Detail Lives

Sharp's quality:"auto" and mozjpeg's -quant-table 2 allocate quality adaptively: busy regions get more bits, flat regions get compressed harder. Same file size, noticeably better result.

You do not have to type all of this at a command line. The Image Compressor already ships sensible presets tuned for web use — drag a file in and it is done.

Color Space: Why Colors Drift

Color space is another easy place to crash. Mishandle the ICC profile during conversion and your colors shift.

Three Common Color Spaces

• sRGB: the web standard and what browsers assume by default.
• Display P3: the wide gamut used by Apple devices, about 25% larger than sRGB.
• Adobe RGB: common in print, with a wider green range than sRGB.

Gamut Mapping and the RGB Trap

Converting from a wide gamut down to sRGB requires gamut mapping, and anything outside the target range gets clipped or squashed. The subtler problem: one RGB triplet — say R=200, G=50, B=50 — renders as two different colors in sRGB and Adobe RGB. Never mix those up.

One Consistent Rule for the Web

For web work I convert everything to sRGB and then strip the ICC profile. Browsers already assume sRGB, so removing it trims the file; keep a Display P3 profile only when you actually need wide-gamut display.

Metadata: The Step Everybody Forgets

What Each Metadata Type Stores

• EXIF: shooting parameters and GPS.
• XMP: edit history and copyright.
• IPTC: title and author.

All of it can disappear the instant you convert.

Format Support at a Glance

JPEG carries all three; PNG natively supports XMP only; WebP and AVIF support EXIF and XMP. Decide what you need to keep before you pick a target format.

Keep or Strip? Decide by Use Case

• Web publishing: strip GPS and personal data (privacy), but keep copyright.
• Archiving: keep everything.
• Social media: the platforms strip it on their own, so do not bother.

To remove GPS surgically and leave the rest untouched, run exiftool -gps:all= -overwrite_original image.jpg. In Sharp, .withMetadata() preserves metadata, and ImageMagick keeps everything as long as you do not pass -strip.

Frequently Asked Questions

Does converting JPEG to PNG improve image quality?

No. JPEG to PNG is a lossless conversion, but the information lost during the original JPEG encoding cannot be recovered, and the resulting file is usually much larger. Lossless only guarantees zero loss during the conversion itself; it cannot undo damage that already happened.

What quality setting should I use for web images?

JPEG 75-85 works well for the web and 90-95 for print. WebP q75 looks roughly like JPEG q85 at a smaller file size. AVIF uses a 1-63 scale where lower means higher quality; 30-40 suits most web work. Verify with SSIM or VMAF once you have settled on a value.

Why do colors shift after converting an image format?

Usually the ICC profile was mishandled. The same RGB triplet is two different colors in two different color spaces, and converting a wide gamut space such as Display P3 or Adobe RGB to sRGB requires proper gamut mapping, otherwise out-of-gamut colors get clipped or compressed.

Does converting image formats strip EXIF shooting data?

It depends on the target format and the tool. JPEG carries EXIF, XMP and IPTC; PNG natively supports only XMP; WebP and AVIF support EXIF and XMP. For web publishing, strip GPS and personal fields but keep copyright, and use exiftool to remove GPS precisely without touching anything else.

Wrap Up

If you remember one line, make it this: go lossless when you can, stop fiddling with parameters, and keep an eye on color and metadata — do that and you will rarely get burned. All of the above exists so you do not end up where I did once: hours spent compressing images, only to find them all mushy and start over from scratch.

If you simply need a common format and a smaller file before you send something out, skip installing half a dozen apps — FreeConvertBox's Image Format Converter and Image Compressor cover it. No uploads, no retention. Your privacy is fully protected. Need documents to go with the photos? Follow up with Images to PDF — a dozen more free tools sit right there on the same site.

Note: Some source material was AI-assisted.

---

---