Squoosh

上传图片时,最让人纠结的往往是文件大小。 原图太大,网站不让传;压缩得狠一点,照片里的细节没了,截图上的小字也开始发虚。反复导出几次,最后自己都分不清该用哪一张。 Squoosh Squoosh 是

Introduction

When uploading images, the biggest concern is often the file size.

The original image is too big and the website won't allow it to be uploaded; if you compress it more harshly, the details in the photo will be gone, and the small text on the screenshot will start to look blurry. I exported it several times, but in the end I couldn't tell which one to use.

Squoosh

Squoosh is an open source browser image compression tool from Google ChromeLabs. It supports adjusting the file size while viewing the image quality. This is also its unique ability to view the comparison between the compressed image and the original image in real time.

It is also an image compression tool that supports adjusting image size, switching output formats and setting compression quality. It is suitable for checking images one by one before exporting.

Note that the website is in English. Generally, users only need to adjust the slider in the lower left corner to adjust the compression quality. The other parameters are left as default.

Tutorial

The first step is to drag in the image. Don’t rush to change the size yet.

Open the website and drag the image into it, or click on the page to select the file.

After entering the editing page, usually the original image is on the left, the compressed preview is on the right, and there is a draggable dividing line in the middle. The file sizes are displayed below.

It is recommended to keep the original size at first and only adjust the compression settings.

When comparing like this, you can tell that the change in image quality comes mainly from compression. If you reduce the image, reduce the quality, and reduce the color at the same time, it will be difficult to determine which step makes the image blurry.

Picture matching suggestions: Take a screenshot of the editing page and mark the "original image", "compressed preview", "comparison boundary" and "file size".

The second step is to select the format based on the image content.

There is no need to study the formats one by one. First, it depends on what images you want to process and what formats the uploading platform supports.

For ordinary photos, you can try MozJPEG first.

MozJPEG here is a JPEG encoder and exports common JPEG images. If the target platform accepts WebP, you can also switch to WebP to compare and see which file is smaller under similar look and feel.

For screenshots with a lot of small text, try OxiPNG first.

It outputs PNG, which is suitable for preserving text and lines first. When maintaining the original size and not reducing colors, you can first see how much space lossless optimization can save. PNGs that have been optimized may not become noticeably smaller.

If PNG still exceeds the limit, try another format and double-check the fine print.

If you need a picture with a transparent background, don’t choose JPEG.

For example, if you have cut out the background of product images and transparent icons, you should choose PNG or WebP that supports transparency, and confirm the transparency effect after exporting.

Before choosing a format, it’s best to look at the upload requirements. When the platform only accepts JPG and PNG, even smaller WebP cannot be used.

The third step is to adjust the quality and focus on checking three places.

When using lossy formats such as MozJPEG, you can passQualityAdjust quality.

Start with the default value, and gradually adjust it downwards. Check the file size after each adjustment, and then drag the dividing line to compare. Don’t just focus on “percent reduction.”

When checking the image quality, you can fixate on the following three positions.

1. Are the small characters blurred?

Give priority to finding the smallest line of text in the image.

Check whether the strokes are still separated, whether there are blurry spots around the words, and whether the light-colored small words begin to blend with the background.

Especially for tutorial screenshots, the entire picture may look good when zoomed out, but once it is enlarged, the menu text is not easy to read.For images whose text needs to be read clearly, readability should be prioritized over file size.

2. Are the edges dirty?

Check the character's hair, product outlines, icon edges, and where dark and light colors meet.

When compression is excessive, noise or color blocks may appear near edges, or otherwise clear thin lines may become loose. These changes tend to be easier to spot on a flat background.

3. Is the file size sufficient?

Assuming that the upload requirement is no more than 500KB, you can stop when the image is below the limit and the details are acceptable.

There is no need to continue to sacrifice image quality in order to further compress the file to 100KB.

in addition,Setting Quality to 75 does not mean that the file will be reduced by 75%, nor does it retain 75% of the original image quality.. Quality values ​​in different formats cannot be directly compared horizontally.

Picture suggestion: Enlarge the same small text or the edge of an object to show the difference between higher quality and over-compression, while retaining the file size reading.

Step 4: If it is still over the limit, consider reducing the size.

If it is still too big after adjusting the quality, you can openResize, appropriately reduce the image width and height.

For example, a photo that is several thousand pixels wide is only used for small illustrations in an article, and it may not necessarily need to retain its full size. But if it is a densely packed tutorial screenshot, if it is reduced too small, the pixels occupied by the text itself will not be enough, and no matter how high the quality is, it will not be able to make up for it.

It is recommended to read twice during the inspection:

  • See details at 100% zoom, check text, outlines, and compression traces.
  • View it again in actual display size, confirm whether it is clear to readers when browsing normally.

Squoosh's preview supports zooming. Don’t judge image quality only on the basis that “the whole picture just fits on the screen”, as that would be easy to overlook the problem.

Step 5: After downloading, open and check

After confirming the effect, click the download button next to the compression result.

After saving, open the exported image and confirm the format, size, file size and text clarity. If there is a transparent background, check that too.

Remember to keep a copy of the original image.If you need to change the size or re-adjust the quality later, try to start from the original image to avoid repeatedly compressing the file that has been processed in a lossy manner.


Visits:9
Updated:2026-11-05
1
Zlibrary
Zlibrary被称为全球最大的数字图书馆,里面包含 9,826,996 本电子书,84,837,646 篇期刊文章。 从各种知名文学著作,理工学科,人文艺术、到学术论文等应有尽有!支持PDF、epu
2
LinkFox AI
为跨境电商提供的AI工具
3
堆友AI
堆友是Alibaba Design打造的设计师全成长周期服务平台,围绕品质、效率、技能、成就、收入五大用户价值布局平台能力,全力服务设计师,旨在成为设计师的好朋友。 堆友历经大厂设计师团队多轮打磨雕刻,集海量高品质3D素材、实时在线渲染、多元场景功能应用、轻便好学易上手等多重优势于一身的设计神器,更自带免费可商用属性,为专业设计师、运营工友、学生小白、社交达人提供了一个零成本的在线设计站点和资源库。
4
HeyGen
AI数字人视频生成
5
PixPix
AI电商套图与视频生成
6
OptimizerAI
AI音效生成器,基于文本或音频生成无限高质量声音。
7
Riffusion
生成式AI乐器,能够根据文本提示创建、重混和分享高品质歌曲。
8
StockTune
StockTune 提供人工智能生成的免版税音乐,供内容创作者使用,无需署名。
9
PDF Converter
免费在线工具,用于转换和编辑 PDF 文件。
10
高性价比人生指南HowToLiveBetter
手机里还有哪些会员在自动扣费?常用邮箱的密码,有没有和其他网站重复?租房合同里,押金到底什么时候退? 这些事单独拿出来,都不算复杂。但没人提醒的时候,很容易踩坑。今天分享一个 GitHub 60.5k
1
Zlibrary
Zlibrary被称为全球最大的数字图书馆,里面包含 9,826,996 本电子书,84,837,646 篇期刊文章。 从各种知名文学著作,理工学科,人文艺术、到学术论文等应有尽有!支持PDF、epu
2
LinkFox AI
为跨境电商提供的AI工具
3
堆友AI
堆友是Alibaba Design打造的设计师全成长周期服务平台,围绕品质、效率、技能、成就、收入五大用户价值布局平台能力,全力服务设计师,旨在成为设计师的好朋友。 堆友历经大厂设计师团队多轮打磨雕刻,集海量高品质3D素材、实时在线渲染、多元场景功能应用、轻便好学易上手等多重优势于一身的设计神器,更自带免费可商用属性,为专业设计师、运营工友、学生小白、社交达人提供了一个零成本的在线设计站点和资源库。
4
HeyGen
AI数字人视频生成
5
PixPix
AI电商套图与视频生成
6
OptimizerAI
AI音效生成器,基于文本或音频生成无限高质量声音。
7
Riffusion
生成式AI乐器,能够根据文本提示创建、重混和分享高品质歌曲。
8
StockTune
StockTune 提供人工智能生成的免版税音乐,供内容创作者使用,无需署名。
9
PDF Converter
免费在线工具,用于转换和编辑 PDF 文件。
10
高性价比人生指南HowToLiveBetter
手机里还有哪些会员在自动扣费?常用邮箱的密码,有没有和其他网站重复?租房合同里,押金到底什么时候退? 这些事单独拿出来,都不算复杂。但没人提醒的时候,很容易踩坑。今天分享一个 GitHub 60.5k