Spotify Clears 50 Million AI Slop Tracks: A Backlash for Human Music?

Spotify's Two-Pronged Strategy: Clean Up Slop, But Don't Force Labeling

Spotify dropped a significant number this week: over the past year, it removed more than 50 million AI-generated slop tracks from user recommendations, while adjusting its systems to give more human-made music exposure. That sounds like a show of support for human creators, but Spotify's stance on AI music is far more complex than this number suggests.

According to The Verge's Vergecast podcast, the 50 million tracks Spotify removed mainly came from its recommendation system, not its entire catalog. This means Spotify's cleanup is closer to "noise reduction at the recommendation algorithm level" rather than a systematic ban on AI music.

More intriguing is how Spotify differs from its peers. YouTube Music and Deezer have taken stricter stances on AI music: labeling AI songs and actively filtering out AI slop. Spotify, however, does not mandate that music be labeled as AI-generated, and has even partnered with Universal Music Group to launch a tool for creating AI music remixes.

This contradictory posture reflects the platform's dilemma: on one hand, AI-generated slop dilutes recommendation quality and harms user experience; on the other, AI music creation itself is a huge commercial opportunity, and the platform is reluctant to set limits too early. Cleaning up recommendation results without forcing source labeling is essentially a "pragmatic middle path"—not giving up the commercial potential of AI tools while trying to maintain the basic quality of recommendation feeds.

Why "Not Labeling" Is a Key Choice

In AI content governance, labeling is one of the most basic and effective means. Deezer and YouTube Music choosing to label effectively hands judgment to users; Spotify not forcing labeling keeps judgment in the hands of platform algorithms. The former offers greater transparency, while the latter allows more regulatory room but weakens users' right to know.

From an SEO and recommendation system perspective, Spotify's approach has its logic: if AI music were forcibly labeled, many AI tracks could be downranked in search results, potentially affecting the platform's partnership ecosystem with AI music creation tools. But this also means users can hardly identify on their own whether what they're hearing is AI-generated, and the platform bears greater responsibility as a "content gatekeeper."

AI content labeling typically refers to platforms adding visible identifiers (such as tags, watermarks) or metadata markers when displaying AI-generated content, enabling users to identify the source. Deezer employs a system that actively detects and marks AI-generated tracks, while YouTube requires creators to disclose AI-generated content and label it in video descriptions. The core value of labeling mechanisms lies in safeguarding users' right to know, but they may also trigger a "labeling stigma" effect—labeled content may automatically receive lower user trust and algorithm weight. Spotify's stance of not mandating labeling effectively shifts identification responsibility from users to platform algorithms, creating potential tension with regulatory trends such as the EU AI Act's requirement that deep synthesis content must be disclosed.

Adobe Lightroom: AI Helps You "Open Your Eyes"

If Spotify's news is about the content ecosystem, Adobe's news is about everyone's photos. Adobe announced that mobile Lightroom can now use generative AI to open the eyes of people with closed eyes in photos.

The technical details of this feature are noteworthy: it not only matches the original image's lighting and pose when generating replacement eyes, but also references other photos of the same person in your Lightroom library to generate more accurate eyes. This means AI isn't "drawing" a pair of eyes out of thin air, but performing personalized reconstruction based on the person's facial features.

Vergecast hosts immediately recalled Wesley Snipes' classic meme from Blade—when producers reportedly used CGI to add eyes because he wouldn't open them. This analogy precisely captures the awkwardness and potential of generative AI in image editing: it can fix a moment of closed eyes, but may also create a jarring sense that "that's not really them."

Where Are the Boundaries of Generative AI Photo Editing?

The practical scenarios for Lightroom's feature are clear: someone blinks in a group photo, or an expression isn't perfect in a candid shot—previously you'd have to reshoot or give up, but now you can fix it in post. But the risks are equally clear—when AI generates eyes based on other photos of the same person in your library, how natural the result looks depends on the quality and quantity of reference photos. For people with only a few photos in the library, results may suffer significantly.

This raises a broader question: as generative AI intervenes more deeply in photo editing—from denoising and outpainting to now "eye-opening"—users' definition of "real photos" is being renegotiated. Adobe choosing to launch this feature in mobile Lightroom also shows that this type of AI photo editing is moving from professional tools to everyday use.

Generative Fill is a feature powered by Adobe's Firefly model that allows users to select an area in an image and have AI generate matching content based on context. In Lightroom's "eye-opening" feature, the technical process roughly involves: first locating the eye area through face detection, then retrieving other photos of the same person from the user's library as references, extracting eye features for the generative model to synthesize new eyes, and finally matching lighting and pose. The advantage of this "personalized reconstruction" over generic generative models is maintaining personal identity consistency, but the disadvantage is heavy reliance on the quantity and quality of reference photos—if the library lacks clear frontal photos of the same person, generation quality drops significantly. This also raises a privacy concern: the feature requires access to biometric data in users' libraries, and how Adobe stores and processes this facial information deserves attention.

Nintendo Switch 2 and Sports Resort: A Hybrid of Motion and Mouse Controls

Vergecast finally turned its lens to gaming. Nintendo Switch 2 is now on sale, and Nintendo Switch Sports Resort launched the same day, featuring 12 sports including two classics from the original Wii Sports: tennis and bowling.

The game's use of Switch 2's new hardware is noteworthy: it uses Joy-Con-like controllers to simulate jet ski handlebars, mouse controls for skateboarding, and offers an open world and sightseeing mode letting players explore the island by land, sea, and air. Vergecast hosts joked "now it's GTA 6's turn to make a move," hinting at the motion game's ambitions in open-world exploration.

From a product strategy perspective, Nintendo continues its tradition of "hardware features driving gameplay": Switch 2's mouse controls and improved motion sensing are directly translated into skateboard and jet ski controls. Whether this design philosophy can convince players to buy a Switch 2 will be a key indicator for the new console's early sales.

Joy-Con mouse mode is a new hardware feature of Switch 2: when a Joy-Con is placed sideways and slid across a desk, its optical sensor tracks movement like a computer mouse, enabling precise cursor control and sliding operations. This design gives Switch 2 PC-like pointing precision while retaining motion gameplay, suitable for game types requiring fine control (such as shooting, strategy, skateboard maneuvering). The skateboarding event in Sports Resort leverages this feature by mapping mouse sliding to skateboard steering and trick moves, while jet skiing continues to use motion gyroscopes to simulate handlebar tilting. This "motion + mouse" hybrid input scheme is Nintendo's direct showcase of Switch 2's differentiating selling point, and also distinguishes it from Sony and Microsoft's traditional controller designs centered on joysticks and triggers.

The Common Thread Behind All Three Stories

Spotify cleaning up AI slop, Adobe using AI for photo editing, Nintendo using new hardware for motion gaming—these may seem unrelated, but all point to the same trend: AI is moving from "generating content" to "managing content" and "enhancing experiences."

Spotify's challenge is managing the flood of AI-generated content, Adobe's challenge is using AI to enhance rather than fake reality, and Nintendo's challenge is using hardware innovation to carry entertainment experiences in the AI era. What all three share is that mere "generation" is no longer enough—how to filter, label, control, and enhance is where the real competitiveness of platforms and products lies.

For content creators and ordinary users, Spotify's 50 million cleanup number is a reminder: the scale of AI-generated content is already large enough to require proactive platform governance, and the choice of governance approach—label or not, clean recommendations or clean catalog—will directly affect what we hear, see, and play every day.

One-line Summary

Spotify clears 50 million AI slop tracks but refuses mandatory labeling of AI music, while Adobe and Nintendo show how AI and hardware can enhance experiences.

Paragraph Summary

Over the past year, Spotify removed over 50 million AI-generated slop tracks from its recommendation system, yet refuses to mandate AI music labeling and has partnered with Universal Music to launch an AI remix tool—a two-pronged strategy of "cleaning up slop without setting limits." In contrast, YouTube Music and Deezer take tougher stances with labeling and filtering. Adobe's mobile Lightroom adds a generative AI eye-opening feature that matches lighting and pose and references photos of the same person in the library, raising discussions about authenticity and privacy. Nintendo Switch 2 launches, with Sports Resort offering 12 sports and open-world exploration via hybrid motion and mouse controls. All three stories point to AI shifting from content generation to content management and experience enhancement.