Quality 90 can map to different quantization tables and encoder settings in different applications. Export once from the same lossless source and compare visible artifacts, file size, color handling, and service recompression.
PNG → JPG
Treat the quality value as an application control
JPEG reduces data partly by quantizing transformed image components, but the JPEG standard does not assign one universal result to a user-interface value such as 80 or 90. Each program maps its control to quantization tables and other encoder decisions. One application may offer 0 to 100, another 0 to 1, and another only named presets. Even controls that display the same range can use unrelated internal mappings.
A note saying that an existing image was saved at quality 90 is therefore not enough to reproduce it. Record the application and version, output dimensions, color mode, chroma-subsampling option, optimization setting, and any sharpening or resizing. Define the comparison goal before testing: the best appearance under a file limit, readable small text, natural skin, smooth gradients, or the smallest acceptable download may lead to different choices. Use the number only to generate candidates.
- Do not read the number as a standard grade
- Record encoder and related settings
- Define the practical quality target
The same number does not guarantee the same quantization tables, file size, or visible quality.
Inspect quantization and chroma subsampling
JPEG encoders can use different quantization tables for luminance and chroma components. Those choices affect fine texture, gradients, and ringing around sharp boundaries. A high displayed quality can still treat certain frequencies differently from another encoder. Examine candidates at 100 percent for block boundaries, mosquito noise, ringing, banding, and lost texture. When suitable tooling is available, inspect the quantization tables as supporting evidence rather than inferring them from the slider.
Chroma subsampling reduces color resolution to save space and is especially visible around small red or blue text, interface captures, and saturated edges. One tool may switch to 4:4:4 at a high setting while another remains at 4:2:0. A test set containing only photographs can hide this difference. Include portraits, foliage, gradients, colored text, and flat graphics. If the lossless source has transparency, standardize the background used before JPEG encoding as well.
- Examine artifacts and quantization data
- Identify chroma-subsampling choices
- Use photographic and graphic samples
Encode once from the same lossless source
For a fair tool comparison, create every candidate directly from the same PNG or lossless editing master. Re-saving an existing JPEG mixes the first generation's losses with the new encoder's decisions, making attribution unreliable. Keep pixel dimensions, crop, color profile, background, and metadata policy identical. Disable automatic resizing, enhancement, denoising, or sharpening unless that behavior is intentionally part of the workflow being compared.
Name outputs by tool and settings and keep their checksums. Compare both at 100 percent and at the real presentation size. Enlargement helps reveal blocks and ringing, while actual size tests readability, skin texture, and overall perception. Objective difference maps or metrics can supplement the review, but they cannot decide which artifacts are acceptable for a particular product page, document, or social post. Use file size as one outcome, not the sole ranking.
- Use one lossless master
- Avoid generational JPEG loss
- Review enlarged and at actual size
A recompressed JPEG is not a neutral source for comparing encoders.
Control color, metadata, orientation, and API behavior
Differences outside compression can make one JPEG appear worse. If an ICC profile is removed or the color space changes, a color shift may be blamed on quality. Exif Orientation handling can alter displayed dimensions or direction. Progressive output, metadata retention, and optimized entropy coding can also change file size without a proportional visual change. Inspect profiles, orientation, dimensions, metadata policy, and progressive status alongside visible artifacts.
Some APIs, including browser canvas export interfaces, pass a quality hint to an implementation that chooses the actual encoding behavior. The same numeric argument can produce different details across browser engines or versions, and unsupported values may fall back to defaults. Pin the runtime and preserve representative outputs for automated work. If the requirement is a target byte size, find a suitable setting separately for each encoder rather than assuming equal numbers create equal compression.
- Match ICC and orientation conditions
- Record progressive and metadata choices
- Pin browser and encoder versions
Approve the final file after service processing
Upload the candidate JPEGs to the actual CMS, marketplace, messenger, or social platform. The service may resize and recompress them, so a locally superior setting can become indistinguishable or even worse after processing. Download the preview and full rendition that users receive, then compare dimensions, file size, color, text edges, gradients, and photographic detail. Check both mobile and desktop presentation when their delivered sizes differ.
Record the lossless source, tool and version, displayed quality value, quantization or subsampling information, profile, dimensions, and service result. If a candidate fails, generate a new file from the lossless source rather than re-saving the derivative. The final practical check is to approve portrait, landscape, colored-text, and gradient samples through the complete delivery path and set a minimum acceptable result for each content class. Only then standardize the export setting.
- Inspect service recompression
- Compare mobile and desktop renditions
- Approve several content classes
Choose the setting from the file users actually receive, not from the number shown in the export dialog.
Key takeaways
- The JPEG specification does not define application quality-slider numbers as a universal visual-quality scale.
- Identical numbers can produce different quantization tables, chroma subsampling, optimization choices, and output sizes.
- A fair comparison starts from one lossless source and encodes each candidate only once under controlled dimensions and color conditions.
- Choose settings from visible artifacts and the final delivered file, not from the slider number or local file size alone.
Frequently asked questions
Is JPEG quality 90 the same in every program?
No. Programs can map that number to different quantization tables, subsampling, and encoder options.
Does a larger JPEG always have better visual quality?
No. Metadata, optimization, profiles, and encoder behavior also affect size, so inspect the pixels under controlled conditions.
Can saving an old JPEG at quality 100 restore it?
No. Information lost during the first encoding cannot be recovered, and another JPEG save can introduce additional changes.
What should I test besides photographs?
Include colored small text, UI graphics, gradients, and sharp saturated edges to expose subsampling and ringing differences.
How should I select the final setting?
Upload candidates made from the same lossless master, inspect the service-delivered files, and choose the lowest setting that meets the visual and size requirement for each content type.