Cover Image
Hong Kong held the “Victoria Harbour Maritime Carnival,” showcasing four large inflatable puppet sculptures along the Victoria Harbour shoreline. This is a photo of the inflatable puppets being transported. (via)
Several Creative Uses for Nano Banana
Last month, Google released the image model Gemini 2.5 Flash Image (project name Nano Banana).
Google calls it the “most advanced image generation and editing model” currently available.
After trying it out, I feel it is indeed very powerful and is free to use. You can start using it by opening the official website (image below).
(Note: If you cannot access the official website, there are also third-party websites in the weekly discussion section that connect to the official API, but most require a fee.)
Netizens have discovered various amazing uses for this model, and some have even collected them into an Awesome repository.
I have selected a few very practical examples from this repository to share with everyone. It should be noted that I think other image models can also do these things, and everyone is welcome to try them out.
(1) Portrait Processing
The most common task for image models is certainly portrait processing. Let’s first upload a personal photo.
Then, we ask the model to convert it into a passport photo with the following prompt:
Please generate a 1-inch ID photo for the person in the photo, requiring a white background, professional attire, open eyes, and a smile.
This effect is somewhat astonishing. It means that the person’s expression, hairstyle, makeup, clothing, and posture can all be changed.
Below is an example of changing the person’s expression to have them turn their side face towards the camera and smile.
Changing the person’s posture, “Change the person in the second image below to the posture in the first image.”
Photo studios might be in danger in the future, as portrait photos, travel photos, and group photos can all be entrusted to AI.
(2) Architecture Processing
Another use for image models is for home decoration. If you want to see home decoration renderings, just let the AI generate them. Changing decor colors and furniture is a small case.
Here is a more challenging example: uploading a floor plan and asking it to become a 3D model rendering.
Extracting an architectural model from a photo is also quite magical.
(3) Packaging Processing
Next, we ask the model to change the packaging of an item: “Affix the comic image from Figure 2 onto the packaging box from Figure 1, and generate a professional product photo.”
Book covers and software packaging boxes can be generated in the same way.
(4) Map Processing
Another large market for image models is map applications (geographic information), but a way to monetize it has not yet been found. Below is an innovative use case.
Upload a map with an arrow marking your selected location, and ask the model to “Generate the scene seen along the red arrow.”
It can even generate a real-world scene at the red arrow location from a topographic contour map.
Technology Trends
An American company has launched a supersonic chef’s knife.
It has a button on the handle, which, when pressed, puts the blade into ultrasonic mode.
According to the description, when ultrasonic waves are turned on, the blade vibrates more than 40,000 times per second, making the knife much sharper than it actually is and saving up to 50% of the effort required for cutting vegetables.
In some cases, placing it on food allows it to automatically cut the food using the waves generated by the vibrations.
This knife has a built-in battery, so it comes with a matching knife charger.
Circuit boards are the foundation of electronic products.
A foreign netizen, in order to demonstrate that circuit boards are not highly complex products, specially made a clay circuit board.
They posted photos of the entire process online, first collecting mud, then flattening it.
Carving out the circuit on it, and then firing it.
Finally, after installing copper wires and electronic components, the circuit board is complete.
According to the AI report by renowned analyst Mary Meeker, if AI-related positions are excluded from the IT industry, the number of employed people in the US IT industry has been flat or declining for many years.
In the image above, the blue line represents the total employment in the IT industry, and the green line represents the employment with AI positions excluded. The peak in the middle occurred during the pandemic.
This means that while the IT industry itself has been expanding, all job growth has occurred in the AI sector.
Articles
- Beyond Sandbox (English)
How to safely run third-party code on a webpage? Google proposes a brand new solution: SafeContentFrame.
It is a JS library that loads third-party code onto a separate domain, googleusercontent.com, and then inserts it into the current webpage using an iframe, thereby providing maximum isolation.
Offline use functionality has not yet become popular. The author believes that offline use is equivalent to building a distributed system, which faces complex synchronization issues and is difficult to implement correctly.
- Elasticsearch Was Never a Database (English)
Elasticsearch is currently a mainstream search service. Can it be used as a primary database? This article tells you that it cannot, as it was not designed for use as a database.
A simple introductory tutorial, guiding you step-by-step to write a Python script to extract a transcript from audio using the Whisper model.
- Avoid using @ts-ignore (English)
TypeScript’s @ts-ignore annotation is used to suppress all errors on the next line. The author argues it should not be used, preferring to switch to the @ts-expect-error annotation or the any type instead.
The author discovered that Apple has added an unpublicized CSS property to the Safari browser, allowing web elements to display a “liquid glass” effect.
A beginner’s tutorial teaching you 5 tips to shorten boot time by adjusting systemd settings.
Tools
A tool to display GPU running information, with a built-in Web management panel, supporting various brands including Nvidia/AMD/Intel/Apple.
There is also an online GPU performance testing website Volume Shader BM. (@BOS1980 submission)
A terminal tool for monitoring network traffic, which displays detailed connection information, cross-platform.
A self-hosted dashboard that lists local ports occupied by various services. When combined with a Compose file, it can start/stop Docker containers. See the introduction article.
A Docker container that scans the current network and graphically displays network node information.
A terminal-based file manager, supporting Linux and Mac.
There is another similar terminal file manager called Yazi.
A client-side auxiliary Bash script for the inner network penetration tool frp, simplifying the creation and management of tunnels. (@openapphub submission)
Open-source task management software, supporting Web/mobile/desktop platforms, and can be deployed with Docker. (@CaryTrivett submission)
A PostgreSQL/MySQL database backup tool written by a netizen, which can automatically backup, encrypt, and compress databases, and upload the backup files to Tencent Cloud COS or Alibaba Cloud OSS. (@iKeepLearn submission)
A command-line tool collection that enables 1000+ CLI tools with one click, cross-platform, and supports AI features. (@Zhengqbbb submission)
AI Related
A personal voice companion that generates a podcast-like “daily briefing” for listening, with content including daily news, interests, personal calendar, and emails.
It is the startup product from the core developers of NotebookLM after they left Google. It is currently free to use. See the introduction article.
An open-source Chrome extension previously introduced in the weekly, whose functionality has now been extended to perform browser automation using AI. (@buttercannfly submission)
A Python script that automatically changes the Windows wallpaper to the Bing daily wallpaper and adds a “word of the day” to the wallpaper, with the AI generating the word explanation and example sentences. (@klemperer submission)
A fine-tuned model based on the open-source Index-TTS speech synthesis model from Bilibili, which improves the prosody and naturalness of the voice. (@asr-pub submission)
A terminal-based smart coding assistant (Code Agent), which can be seen as an open-source Claude Code. (@xierenyuan submission)
A Web-based AI video subtitle editing tool that can automatically convert video speech to text and generate subtitles. Try the Demo. (@x007xyz submission)
A management tool for MCP services that connects to various MCP servers and has a built-in Web management panel. (@whillhill submission)
Resources
An online English book that features 99 selected small physics experiments from Dutch secondary schools, covering various fields (force, light, magnetism, waves, etc.).
A series of npm package poisoning incidents have recently been exposed. This repository collects various npm security measures, divided into two main parts: for users and for publishers.
Images
- Electric Vehicle Principle
An image circulating online illustrating the principle of an electric vehicle.
There is a small circle inside a square. What is the relationship between the radius of the small circle and the side length of the square?
This problem seems quite difficult. The answer is 4/33 of the side length of the square.
Excerpts
Based on my observations, the way senior and junior programmers in a company use AI is different.
Senior programmers do not fully trust the output of AI; they only use AI to accelerate projects. They generally review and refactor the code generated by AI, and they approach AI’s architectural decisions with skepticism.
Junior programmers are more inclined to skip reviewing and refactoring, fully accepting the AI’s output, which leads to “house of cards” code: it seems to work, but it collapses as soon as it’s put into use.
I don’t know if AI will replace programmers in the future, but at the current stage, AI coding cannot solve 100% of software problems, but it can solve 70%. This is equivalent to AI being able to reduce the workload of senior programmers by 70%.
The remaining 30% still requires the programmer’s experience and expertise, and it is precisely this 30% that junior programmers lack.
Therefore, it may sound counter-intuitive: AI is more helpful to senior programmers than to junior programmers, and is more likely to generate work results.
AI at the current stage is more like a very motivated junior programmer in the team, capable of writing code quickly, but requiring constant supervision and correction. The more you know, the better you can guide it.
Therefore, the correct use of AI is for senior programmers to use it to accelerate what they already know how to do, and for junior programmers to use it to learn what they should be doing.
Quotes
AI will expand until most of the sun’s energy is used for computation.
– Elon Musk’s latest interview
I think Mars can be self-sufficient within 30 years. Every two years, the planets align, and you can depart for Mars. So, there are about 10 to 15 Mars departure windows within 30 years.
With each launch, the tonnage of cargo transported to Mars will increase exponentially. Thus, we can make Mars self-sufficient within 30 years.
– Elon Musk’s latest interview
A new job is quietly emerging in the software industry called “Vibe Coding cleanup,” which specializes in solving the problems caused by “Vibe Coding.” This is the greatest irony of the AI era: humans are hired to clean up AI’s trash.
– “Vibe Coding Cleanup as a Service”
An AI bubble is very possible, but for a company like Meta, the greater risk is hesitation.
If we end up wasting hundreds of billions on AI, it’s obviously very unfortunate, but I actually think the risk of missing out on AI is higher. For us, the risk is not being too aggressive, but not being aggressive enough.
Today’s computers are responders: you ask it to do something, and it does it. The next phase of computers is the agent. It’s like a little person in a box who starts predicting what you want. It doesn’t just help you; it guides you through vast amounts of information, like having a little partner in the box.
– Steve Jobs, 1984 interview