Developer uses Claude Code to remove Xiaomi TV ads and open-sources CleanYourTV

Once upon a time, in people's minds, the TV was a species that could be watched as soon as it was turned on, and channels could be changed with a couple of presses on the remote.

Now, when you turn on the TV to watch a movie, the first thing that greets you is a boot-up ad. After finally getting to the home page, you are immediately hit by a full screen of poster recommendations and membership recharge pop-ups, and if you press pause midway, a full promotional overlay will pop up.

Image | Beijing Business Today

Giving the TV a "deep clean"

Last week, developer Gathin posted a set of photos on social media of modifying his parents' TV. According to his description, the Xiaomi TV at home even required watching a sponsored ad before opening Netflix.

So, he handed the TV model to the coding assistant Claude Code, had it connect to the device over Wi-Fi, disable 25 system components in one go, and then write a minimalist launcher—the originally cluttered home page instantly became clean and refreshing.

He then open-sourced this set of prompts, the launcher source code, and the notes on pitfalls encountered to GitHub, naming the project CleanYourTV.

In just a few days, the project quickly gained about 13,700 stars—it belongs to the category of "the world has suffered from smart TVs for too long." In the related discussion area on Reddit, it was packed with users asking whether their own models could be adapted, and many also successively posted their own minimalist home pages after modification.

Image | 🔗 github.com/Gathin23/cleanyourtv

The underlying logic of AI "taking over" the TV is not mysterious; the core is still an old tool that has been used in the Android ecosystem for more than a decade: ADB (Android Debug Bridge).

As long as it connects to the same local network and enables the TV's debugging permissions, the AI can use the terminal interface to investigate the hundreds of application components running. Which are boot-up ads, which are telemetry programs secretly reporting viewing data—the AI screens them one by one and issues commands to "freeze" them, and finally replaces it with a clean open-source launcher.

Image | CleanYourTV

Another YouTuber, Ansub Khan, also documented his modification process and open-sourced the project as TV Detox. On his original TV, turning it on would by default automatically play noisy carousel content.

After the modification, the new home page was left with only the clock, commonly used video apps, and input sources.

Image | 🔗 github.com/Ansub/tv-detox

In the video, Ansub measured that the TV's available memory soared from about 370 MB to 600–650 MB; a TCL Beyond user who followed the prompts also said that although there was no precise data, after getting rid of the heavy background residency, the entire TV's scrolling and response were noticeably lighter and faster.

Looking through the discussions in the Reddit community, I noticed a very interesting detail: the "ideal TV" everyone has in mind seems to not be the same thing at all, and people even cannot reach any consensus with one another.

Some people are still used to the factory interface and only want to remove ads and excess recommendations, and then remap the remote control's shortcut buttons to commonly used apps;

Another Sony KD-55A8H user pursues minimalism even more—he changed the clock to 24-hour format, deleted the morning and evening greetings, the date, and all decorative lines, and even changed the system font to the neat and austere IBM Plex Sans.

In fact, people's needs are inherently different from person to person. But in the commercial logic that pursues standardization, users often can only customize within the framework given by manufacturers.

Hardware doesn't make money, so the screen becomes ad space

In recent years, TVs have become cheaper and cheaper. Besides the decline in panel costs and increased productivity, there is also an element of manufacturers "selling at a loss to gain attention." But manufacturers are not charities after all; hardware is responsible for "making friends" at low prices, so the money naturally has to be earned back elsewhere.

As for what else TVs can rely on to make money after they are sold, we can get a glimpse from the financial reports of the overseas streaming giant Roku.

In 2025, its device hardware business had a full-year gross loss of $82.02 million; the platform business composed of digital advertising, subscriptions, and content distribution had a gross profit as high as $2.156 billion.

TCL Electronics also disclosed that in 2025, internet business revenue was HK$3.11 billion, with a gross margin of 56.4%; during the same period, the gross margin of the large-size display business was 16.8%. Obviously, internet operations built around TVs have thicker gross margin space than hardware.

In China, rectification targeting smart TVs has indeed had some effect. According to data disclosed by the National Radio and Television Administration, 93% of models from mainstream brands already support one-click or permanent disabling of boot-up ads.

But this still treats the symptoms, not the root cause. Restrictions on boot countdown ads cannot stop manufacturers' need to monetize; their commercial calculations will only become more concealed.

Image | An X user complaining about the plaster-like ads displayed when YouTube is paused

Take some domestic streaming platforms, for example: pre-roll ads that are often 90 seconds long, full-screen promotional posters when paused, and "big-screen exclusive VIP" pop-ups saying the annual card is unavailable on mobile—these are still stuck there unchanged like psoriasis.

Tools like CleanYourTV can only clean up the TV system's own preinstalled packages and official launcher. It can help you clean the living room door, but it cannot govern the rules inside each App's house.

We bought the hardware—so why can't we decide what to do with it?

The "Vibe-Hacking" trend is not limited to "TVs."

In the camera world, a group of photographers is also trying every way possible to recover features that manufacturers have hidden.

On older Sony bodies, the open-source project Recipe Lab uses the old PlayMemories Camera Apps entry point to let models such as the A6000 and A6300 load custom color recipes, writing color matrices and white balance parameters into the camera's low-level system.

Another study targeting A7 IV firmware went straight into the depths of firmware version 6.02 and dug out the shutter angle and LUT implementation that had been officially blocked. Although for now only the icon can be brought up and dial adjustment remains restricted, the very existence of the code confirms that the hardware capability has been there all along.

Image | 🔗 github.com/leogenot/sony-a7iv-feature-research

The Nikon community's open-source project Wonder Lantern takes personalization even deeper. One developer not only extended the Z30 with 900-second ultra-long exposures, an on-screen histogram, and custom shutter sounds; the domestic reverse-engineering community even directly unpacked and rebuilt Nikon firmware, repackaged and signed it with a private key, and successfully added new settings options to the camera menu.

Image | After modifying the firmware, the developer successfully wrote a "hello world" entry into the Nikon camera's native settings menu (source: Xiaohongshu)

In the Ricoh community, GR Mod fits photographers' own .cube LUT color presets into in-camera style slots that the camera can recognize, and even adds a 65:24 wide aspect ratio, giving this street-shooting camera a film-like style and an XPAN-style field of view.

In the Fujifilm circle, developer justice-jake is trying to break the official firmware moat—reverse-porting film simulations, snapshot focus, and the recipe editor that are exclusive to newer models onto the almost-forgotten older X70 body.

The channel curves of the old processor are hardware-locked, so he uses AI-assisted USB debugging and firmware emulation to painstakingly extract an approximate algorithm implementation from the low-level code.

An even more dramatic exploration has emerged around Hasselblad. The open-source project Hasselblad X Unleashed takes direct aim at the low-level system of some medium-format bodies—this camera runs a deeply customized Android internally, and this hidden "back door" has geeks very excited.

They not only followed the trail to unlock wireless microphones, external monitoring, and pixel shift; someone in the community even successfully ran the classic game Doom directly on a Hasselblad camera worth tens of thousands of yuan.

Image | A Hasselblad X2D 100C running Doom, with touchscreen turning and shutter firing (source: Reddit / rrRadium)

Compared with disabling a few Android apps on a TV, digging into a camera's low-level firmware is obviously an adventure of a completely different order.

But these two groups, who seem to have nothing to do with each other, are messing around according to the same logic: we paid full price for this machine, so why should how its features are used be decided only by the manufacturer's business calculations?

In the past, breaking through this kind of setup required geeks to grind through assembly code and decompilation tools for months or even years.

Now, for the first time, AI has put the code auditing and troubleshooting capabilities that once belonged exclusively to reverse engineers into the hands of more ordinary people. Those features that were sealed deep in black boxes and crippled by market segmentation, and those corners stuffed with ads, have become transparent and visible for the first time.

From a big cleanup of the screen in the living room to firmware unlocking in the photography world, a bottom-up technological rebellion is taking shape in one seemingly small act of tinkering after another.

Commercial companies' profit calculations will not stop, and digital fences will not disappear into thin air. But as the scales of technology quietly tip, those factory settings once taken for granted have finally been called into question again.

Welcome to the "Great Reverse-Engineering Era."

This article is from the WeChat public account "ifanr" (ID: ifanr), author: Zheng Tingxu, editor: Xiao Qinpeng, published by 36Kr with authorization.