DEV Community: zhouhua xiao

How I Evaluate AI Tools: A Small, Repeatable Test

zhouhua xiao — Thu, 10 Sep 2026 06:19:19 +0000

I used to evaluate AI products by their first demo. If the answer sounded fluent, the interface looked polished, and the generation felt fast, I was impressed. After a few days, however, the real problems usually appeared in the details. Could the tool follow a fixed format? Could I trace an answer back to its sources? Would it admit uncertainty? Would the free allowance cover one complete job rather than one impressive preview?

My current process has four steps. First, I define a real task with a clear boundary. Second, I prepare fixed inputs so the test is comparable. Third, I record time to completion, rework, error types, and final usability. Fourth, I keep product claims, personal impressions, and checkable evidence in separate notes. The result is not a vague statement that a tool is “amazing” or “bad.” It is a practical description of where the tool fits and where it stops fitting.

I put the findings into four boxes: usefulness, traceability, limits, and cost. This is also why the Maker Jury approach makes sense to me. Its public methodology asks reviewers to define the user, job, inputs, comparison basis, product version, account tier, and stopping conditions before testing. New public work does not force everything into a single numeric total; it records the task, observations, measurements, limitations, and disclosures.

The historical records on the homepage are especially useful as examples. The NotebookLM record emphasizes bounded source work and traceability. The Lovable record describes a successful publication of a bounded internal tool, while also noting a long first run and unclear authorization waiting. The Framer record describes a polished, editable, responsive site, alongside cautions about a recoverable connection error and confident claims. That is closer to the answer I need than a leaderboard: what can it do, and where will it make me do the work again?

I No Longer Look for the “Most Powerful” Tool: How I Make Software Decisions

zhouhua xiao — Thu, 10 Sep 2026 06:16:28 +0000

Most software pages promise to be faster, smarter, and more complete. Yet when I am the person paying, I usually want one concrete problem solved: organize a body of research, ship a small website, or remove a repetitive task from my week. As a user who regularly tests new products, I no longer begin by asking which tool is “the best.” I write down the job, the inputs, the expected output, and the limits I cannot accept.

Then I run a small comparison. I give each product the same input, the same task, and roughly the same time budget. I look beyond features: learning curve, export options, reliability, pricing, and whether I can understand a failure when something goes wrong. This changed my definition of software selection. It is not collecting attractive links; it is reducing the number of expensive wrong turns.

That is why I have started using Maker Jury near the beginning of the process. The site organizes products around areas such as Research, AI Coding, Marketing, and Design, while separating basic product information from hands-on test records. Its public historical records show pass, mixed, and fail observations instead of flattening everything into a flattering score. Its methodology also distinguishes Listed, Verified, Jury Reviewed, and Jury Pick, and clearly states that rankings are not pay-to-win.

For me, Maker Jury is not a substitute for judgment. It is a faster way to form a shortlist: understand the job first, inspect what a public test actually observed, and then verify the fit in my own workflow. As the number of tools keeps growing, a trustworthy filtering method is becoming a productivity tool in its own right.

Developers Should Check Hidden Information in Screenshots and Documents

zhouhua xiao — Thu, 10 Sep 2026 05:18:23 +0000

As a developer, I often share screenshots, bug reports, PDFs, presentations, and test files. I once believed that manually covering visible sensitive text was enough. Then I realized that the file itself could contain additional information, including the creator’s name, computer username, local file path, editing software, modification time, comments, and version details.

This information may not appear on the page, but it can become visible when someone downloads the file and checks its properties. The risk is easy to miss when submitting a GitHub Issue, publishing a tutorial, sending a support ticket, or sharing debugging material with an external team.

My sharing workflow is now more structured. I place the file in a temporary folder, remove unnecessary metadata, review the visible content, and check for API keys, internal domains, usernames, local paths, and customer information. When working with images or supported document types, I use MetadataRemover.ai to prepare a cleaner copy without installing a separate utility on every device.

Metadata removal is not a replacement for a complete security review. A visible API key, email address, database URL, or customer name still needs to be manually removed or blurred. Source code, configuration files, and confidential logs should never be uploaded casually to an online tool. I also check whether the file type is supported and inspect the cleaned result before sharing it.

Developer tools are designed to improve efficiency, but efficiency should not come at the cost of information leakage. By adding MetadataRemover.ai to my file-sharing checklist, I can make technical communication, issue reporting, and client collaboration more careful and reliable.

My Photography Workflow Now Includes Metadata Removal

zhouhua xiao — Thu, 10 Sep 2026 05:16:10 +0000

Photography has always been about preserving details. A camera records the scene, but it may also record the time, lens, aperture, shutter speed, device model, and other technical information. Some phones and cameras can also save precise GPS coordinates.

These details are useful for my own archive. They help me remember how a photo was made and make it easier to organize a large collection. However, I do not always want to publish every technical detail when I share a photo on social media, a portfolio website, or a photography forum.

I now keep two versions of important images. The original RAW file remains in my private archive with its complete metadata. The JPEG version prepared for public use goes through a metadata-cleaning step first. This gives me the best of both worlds: a detailed personal archive and a safer file for public sharing.

Before processing a large group of photos, I usually test one image. I use MetadataRemover.ai to create a cleaned copy, then inspect the file properties or use an EXIF viewer to confirm the result. I also open the final image to make sure the visual quality and dimensions are correct.

This is especially useful for travel photography. I may want to share the landscape, but I may not want to reveal the exact location of my hotel, home, or private route. When delivering photos to clients, I can also decide whether copyright and technical information should remain included.

Photography taught me to preserve details. Privacy taught me to share them selectively. Adding MetadataRemover.ai to my publishing routine helps me show my work while maintaining better control over personal information.

I Made Privacy Checking Part of My Sharing Routine

zhouhua xiao — Thu, 10 Sep 2026 05:00:21 +0000

I used to focus only on whether a photo looked good before sharing it. I checked the composition, lighting, and resolution, but rarely thought about the information stored inside the file. Later, I learned that image metadata, especially EXIF data, may include the capture time, camera or phone model, editing software, and sometimes GPS coordinates.

That information is easy to overlook, but it can reveal more than expected. A photo taken inside my home, at my office, or during a private trip may contain a location that I do not want to publish. The image itself may look harmless, while the file quietly carries additional personal details.

My current routine is simple. I keep the original file safely on my computer, make a copy for sharing, remove unnecessary metadata, and then upload the cleaned version. For quick processing, I use MetadataRemover.ai. It gives me a convenient way to prepare a cleaner copy without installing complicated desktop software.

I use this process for family photos, interior images, document scans, order screenshots, and work files. However, metadata removal does not hide visible information inside the image. If a photo contains a name, address, phone number, or order ID, I still need to blur or cover it manually.

Privacy protection does not always require a complicated workflow. Keeping the original separate, sharing a cleaned copy, and using MetadataRemover.ai as a final check helps me reduce unnecessary exposure while keeping the process fast and practical.

I’d Use an AI Bathroom Concept as the Start of a Better Contractor Conversation

zhouhua xiao — Thu, 10 Sep 2026 04:52:42 +0000

When I describe a bathroom renovation, I can easily say, “I want it modern, but warm.” The sentence makes perfect sense in my head and arrives at the job site without an acceptance test.

I would use AI Bathroom Designs to explore one visual direction from my bathroom photo, then sort my notes into three columns:

Keep: the window, the approximate shower position, and storage I cannot lose.
Change: wall color, tile mood, vanity finish, and lighting atmosphere.
Confirm: measurements, door clearance, drainage, plumbing, electrical work, ventilation, waterproofing, and material availability.

That split keeps a concept image from quietly turning into a construction drawing. A photo shows surfaces; it cannot reveal pipes behind a wall or confirm that a product can actually be sourced.

I would put the source photo and concept side by side and ask a few small questions: Did a door or window move? Does the mirror reflection make sense? Is the vanity proportion plausible? Which changes did I request, and which ones did the AI invent to fill a gap?

The bathroom remodel from photo guide makes a useful next-step checklist. The concept can express color and atmosphere; measuring, material selection, and construction decisions still belong to the real room and the relevant professionals.

The best outcome is not asking a contractor to copy the image exactly. It is being able to say, “I like this part, this part must stay, and please verify the rest against the site conditions.”

Would you bring an AI concept image to your first conversation with a designer or contractor?

Before Buying Tiles, I Let My Bathroom Try on Three Palettes

zhouhua xiao — Thu, 10 Sep 2026 04:23:31 +0000

The easiest part of a bathroom renovation is saving inspiration photos. The hard part is choosing a color.

White comes in cool white, creamy white, and gray-white. Wood comes in pale oak and walnut. Muted green looks lovely online and can behave very differently in a small real room.

I would keep one bathroom photo fixed and let the room try on three “outfits” in AI Bathroom Designs: warm minimal, soft sage, and a more natural spa direction. The homepage also shows Quiet classic, Modern luxury, and Coastal calm, which gives me enough options for a first preference check.

The warm palette asks whether pale walls, a wood-tone vanity, and warmer light make the room feel calmer. The sage palette asks whether color can add personality while the floor and vanity stay quiet. The darker spa direction asks whether the room needs more light to keep the contrast comfortable.

I would keep a small comparison table: Is there enough daylight? Does the tile scale suit the room? Does the vanity work with the floor? Do the hardware and mirror finishes agree? Can I find real materials in a similar finish? A green that looks perfect on a screen still deserves a physical sample in daylight and under the room’s lights.

The bathroom design examples are a useful place to browse before uploading a photo. Any free credits, prices, or limits should be checked on the current pricing page.

An AI concept helps me explain “warm, but not yellow.” It cannot confirm dimensions, budget, cleaning effort, or installation conditions. Let the bathroom try on a few looks, then measure the room and check real materials.

Which renovation choice slows you down most: wall color, tile, or the vanity?

My First Bathroom Photo-to-AI Plan Started With a Better Photo

zhouhua xiao — Thu, 10 Sep 2026 04:15:24 +0000

The first time I prepared to hand my bathroom to an AI design tool, I blamed the prompt for being too short. Then I looked at the photo: a towel covered the vanity, only a corner of the floor was visible, and the mirror reflected half the apartment. Even I had to guess what I was looking at.

AI-generated conceptual illustration, not a website screenshot.

I would start with the AI Bathroom Designs bathroom photography guide. Shoot from the doorway or another useful position, keep the floor, vanity, shower, window, and wall boundaries in frame, turn on enough light, move anything that hides a fixture, and check the mirror for people, addresses, or other private details.

The goal is not a real-estate listing. It is a photo that explains the room. The site currently supports JPG, PNG, and WebP images, so I would make a supported-format copy and keep the original untouched.

Then I would open AI Bathroom Designs, keep that source photo fixed, and ask one question at a time: warm or cool walls? Wood-tone or painted vanity? Bright task lighting or a small spa mood?

AI-generated conceptual illustration, not a real renovation before-and-after.

If I change the camera angle, tile, lighting, and vanity in one round, I get a pretty image with no clear explanation. A fixed input and one design question make it easier to understand which change I actually prefer.

The output is still a visual direction. It cannot measure the room, reveal hidden pipes, or approve construction conditions. The useful shift is not “AI finished my bathroom.” It is being able to say what I want to change, what I want to keep, and what needs a professional check next.

Before you upload a room photo to an image-based AI tool, what do you check first?

Before You Rewrite the Prompt, Check the Photo

zhouhua xiao — Tue, 08 Sep 2026 12:07:04 +0000

Picture this: you want AI to give your bathroom a new look. Your prompt reads like a dissertation, but the result seems to come from a parallel universe.
Before adding an eighteenth “premium, elegant, realistic,” look at the input photo. Towels hide the vanity, only a corner of the floor is visible, and the mirror is trying to reflect the entire world.
It’s a little like filing a bug report that says “it’s broken.” Even a capable assistant needs enough context to work with.

AI-generated conceptual illustration of an input problem.
Give the model a photo that explains the room
AI Bathroom Designs uses bathroom photos to generate visual concepts in different styles, helping you compare tile, vanity finishes, and color palettes. With tools like this, the photo is part of the brief.
Before taking it, try four small adjustments:

Show the space. Shoot from a suitable position, such as the doorway, and try to include the floor, vanity, shower, and wall boundaries.
Make the light useful. Avoid a bright window turning the rest of the room into a silhouette. Keep the main areas visible.
Reduce obstructions. Move towels and clutter that cover fixtures. Check the mirror for people or personal items, too.
Watch the perspective. Keep the camera reasonably level and avoid obvious distortion just to fit everything into the shot. You don’t need a real-estate advertisement. You need a picture that helps someone understand the room. Treat redesigning as a small experiment Once you have a suitable photo, keep it fixed and ask one clear design question at a time. For example: “Do I prefer warm walls or cooler tones?” Focus on color in that round and materials in the next. Generated concepts may still alter several details, so keep the original alongside each result and check whether the changes match your intention. If you change the camera angle, style, floor, and lighting all at once, it becomes hard to explain which change you actually liked. It’s like adjusting ten variables while debugging: you have an outcome, but you’ve lost the explanation.

AI-generated conceptual illustration of a photo-taking approach.
A small UX idea for AI product developers
A specific sentence beside the upload field may be worth testing before you write a long prompting tutorial.
Consider a clear preview, a reminder to include major fixtures and the floor, and a way to state what should change and what should stay. These are interaction ideas to evaluate, not claims that the website implements every suggestion.
A successful upload tells you the transfer worked. Whether the input provides enough information is a separate question.
Next time, start with another photo
For a practical starting point, read the bathroom photography guide, then use your own photo to compare design directions.
A clearer photo still doesn’t guarantee accurate dimensions or a perfectly preserved layout. It gives the process a clearer input; the output still needs checking.
If you build image-based AI features, what one reminder would you put beside the upload button?

Your Tool Collection Is Growing. Your Project Folder Is Still Empty

zhouhua xiao — Tue, 08 Sep 2026 04:43:30 +0000

Picture this: you need one handy AI tool to save some time on your weekend project.
Soon, your browser is throwing a party. One tool promises to write your code. Another will build your website. A third sounds ready to take your job on Monday. Your coffee is cold, your bookmarks are full, and your project folder is still empty.
When did choosing tools become another job?

Start with the job you need done
You don’t need a dragon-slaying sword to chop an onion. Software shopping works the same way.
Are you fixing an error, sorting through a few documents, or building a landing page that actually opens? A specific task makes it easier to separate useful features from landing-page confetti.
MakerJury is a software and AI tool discovery site with evidence-led reviews. It brings together product information, test procedures, and limitations so you can judge whether a tool fits your work.
It also distinguishes listings from tested products: being listed doesn’t mean a tool has been reviewed, and scores appear only after a documented test.
Give your shortlist a tiny audition
You don’t have to start with “build me a billion-dollar company.” Ask three smaller questions:

What can it finish? Give it a concrete task, such as fixing a reproducible bug.
How can I check the result? Does the code run? Can I find the cited source? Does the exported file open?
How much cleanup is left for me? What needs manual editing? Is there a paywall? Does it disrupt my existing workflow? The demo gets you excited. The audition tells you whether you can work together.

Read the test, not just the score
For example, MakerJury’s three-task Cursor review separates bug fixing, feature work, and refactoring. It identifies the small public codebase used for testing and explains why those results cannot establish performance on production projects.
That gives you more to work with than “Amazing. Must buy.” You can ask: does this resemble my problem, and what do I still need to verify?
Give yourself a stopping rule, too. If the tool completes the task, the result is checkable, and the cost works for you, start using it. Don’t hand your entire weekend to a comparison spreadsheet while searching for the perfect option.
Before your next software shopping spree, browse MakerJury’s reviews, then try a candidate on your own task.
Your bookmark collection can keep growing. Your project deserves a turn.
What’s one tool you bookmarked thinking “I’ll use this immediately” and then never opened again?

How to Remove GPS Metadata from Photos and Verify the Result

zhouhua xiao — Tue, 08 Sep 2026 03:24:59 +0000

A photo can reveal more than what appears in the frame.
Depending on the device, camera settings, and export process, the file may contain GPS coordinates, a capture timestamp, camera details, or editing information. None of those fields need to be visible when you open the image.
Before publishing a photo in a blog post, attaching it to a public issue, or adding it to a marketplace listing, it is worth checking the actual file you are about to share.
Disclosure: I run MetadataRemover.ai. This article explains the workflow and includes an independent way to inspect the result.
What should you check?
GPS coordinates are commonly stored inside EXIF metadata, alongside fields such as camera make, model, and capture time.
However, EXIF is only one part of the picture. Depending on the format and the applications that touched the file, other information may appear in XMP, IPTC, or image text chunks.
Useful fields to review include:

GPS coordinates and altitude.
Original capture date and time.
Camera model and device identifiers, where present.
Author, description, and editing-software fields. Not every image contains these fields. Inspect the file instead of assuming they are present—or already gone. A browser workflow: inspect, remove, verify For a local browser workflow, you can use the EXIF remover. The current image tool supports JPG, PNG, and WebP files up to 25 MB each. Inspection, cleaning, and output verification happen locally in the browser; the image file does not need to be uploaded to the service. The steps are:
Keep an untouched original for your own archive.
Select the final exported image you intend to share.
Review the detected metadata groups.
Choose Privacy Safe, or use Custom to select supported groups for removal.
Download the cleaned copy and inspect it before publishing. Check the final export, not just the original photo. An editor or conversion step can preserve existing metadata or add new fields. Also, make sure you share the downloaded copy rather than accidentally selecting the original again. Verify the downloaded file with ExifTool The tool checks its generated output, but a second metadata reader provides an additional check. If you have ExifTool installed, run these read-only commands with your own filenames: exiftool -a -G1 -s original.jpg exiftool -a -G1 -s cleaned.jpg The options show duplicate tags, include group names, and use short tag names, making the output easier to compare. Review the fields you intended to remove. Pay attention to GPS entries and any relevant EXIF, XMP, or IPTC fields. Do not expect the cleaned file to produce empty output. ExifTool can still report file-system information, image dimensions, compression details, and other properties. A file modification timestamp is different from an embedded original capture timestamp. The useful question is whether the targeted embedded fields remain within the reader’s supported coverage. Open the image as well. Metadata inspection does not tell you whether its appearance still meets your needs. Two details developers should watch EXIF Orientation affects how an image displays. Some photos store pixels in one orientation and rely on an EXIF value to display them correctly. Removing that value without handling the pixel orientation can produce a sideways image. Normalizing the orientation may require re-encoding, so “remove metadata” should not automatically be described as “always lossless.” ICC profiles affect color rendering. They have a different purpose from GPS coordinates or author fields. Preserving a color profile does not mean privacy cleanup failed. These are reasons to use a deliberate cleanup policy and check rotation, color, and dimensions afterward. Know the limits Metadata removal does not hide an address visible on a package, a reflection, or a recognizable landmark. It also does not remove pixel-level invisible watermarks or erase copies previously uploaded elsewhere. Some platforms strip metadata during processing, but behavior can differ between upload methods and downloaded versions. Checking before upload gives you control over the file that leaves your device. For an image-sharing workflow, a useful sequence is: Export → inspect → remove selected metadata → verify → publish. If you work on an upload or publishing pipeline, where do you handle metadata checks: in the browser, on the server, or both?