3 features that Google still hasn’t launched: Photos, Gboard

Google's best features often never see the light of day. The company has a well-documented history of showcasing ambitious software capabilities at developer conferences, only to quietly shelve them or let them languish in perma-beta. While Apple typically ships what it announces within months, Google sometimes lets years pass between a product demo and any meaningful rollout - if it ever arrives. This gap between promise and delivery isn't just frustrating for users; it erodes developer confidence and fragments the Android ecosystem.

Nowhere is this more visible than in Google Photos and Gboard, two of the company's most popular apps. Both have received teasers for features that remain unreleased, even as competitors ship similar functionality. In this article, we explore into three specific capabilities - two from photos, one from Gboard - that Google has shown off but not yet launched. We'll examine why they matter, what technical hurdles may exist. And what the delays tell us about Google's internal product strategy.

The Curious Case of Google's Feature Teases

At Google I/O 2023, the company unveiled "Ask Photos," a conversational AI layer for its photos app. The demos looked polished: you could say "show me my favorite hikes from last summer" and get a curated album. Yet as of early 2025, the feature remains in Limited testing for Pixel 8 users only, with no public launch timeline. This pattern repeats across product lines. According to internal documents leaked to 9to5Google, at least 30% of features shown at I/O between 2019 and 2023 were never deployed to stable channels. The disconnect appears systemic, tied to Google's cultural aversion to committing to single-product roadmaps and its tendency to parallel‑develop competing solutions internally.

Developers who build on top of Google's APIs have learned to treat pre‑launch announcements with healthy skepticism. When a feature like "Smart Cropping" is shown in a keynote but then stalled, third‑party apps that began integrating it waste engineering cycles. The lack of clear communication around feature discontinuation (the so‑called "Google Graveyard") compounds the problem. For Photos and Gboard - apps that serve over a billion users each - the stakes are particularly high.

1. Google Photos - AI-Generated Video Highlights with Custom Music

In 2020, Google demonstrated a feature Inside Photos called "Movie Studio. " It promised to automatically select frames from your library, synchronize them to a choice of licensed music. And produce shareable highlight reels. The demo used on-device machine learning to avoid uploading content to the cloud. More than four years later, Movie Studio has never been released. Instead, Google launched a watered-down "Memories" feature that auto‑generates slideshows but offers zero customisation - no music selection, no manual trimming, and no export control.

The technical challenge here is twofold. First, copyright‑cleared music licensing at a global scale is a legal minefield. Google would need agreements with every major label (Sony, Universal, Warner) in all regions, a cost that may not have been justified by engagement metrics from early prototypes. Second, the on-device neural network required for scene detection and music synchronisation is computationally heavy; running it on mid‑range Android phones without draining battery or causing lag is non‑trivial. A former Google Photos engineer recounted on Hacker News that the team "ran into performance variances across a hundred device models, and the QA cycle stretched from two months to over a year. "

Meanwhile, competitors like CapCut (ByteDance) and Adobe Premiere Rush ship automatic video highlights with custom music support. Google's delay has ceded the casual video‑editing market entirely. A recent official Google Photos blog post teased new "AI video suggestions" but again provided no shipping date. This is a classic case of a feature being "Google‑ready" (i e., working on test fleets) but not "launch‑ready" (cost‑effective and reliable at scale).

2. Gboard - Universal Inline Translation for Any App

Back in 2019, Gboard's product lead showed off a keyboard‑level translation feature: you could type in any language, and the keyboard would automatically translate your text into the app's target language before sending. Unlike Google Translate. Which requires copying text between apps, this would work inside WhatsApp, Twitter. Or any text field without context switching. The demo used TensorFlow Lite for on‑device translation, meaning it could function offline.

Six years later, the feature has never shipped. Instead, Gboard offers a manual "Translate" button that opens a pop‑up window - essentially the same workflow as switching to Google Translate. The inline translation model was likely evaluated during A/B tests and found to have high error rates on short, casual messages where context is minimal. Translating "Yeah, sure" in isolation can produce wildly different outputs depending on tone. And on‑device models at the time struggled with nuance. Additionally, Google may have been concerned about user confusion: if a keyboard auto‑translates your typed message, who takes the blame when a miscommunication occurs?

• Why it matters: Real‑time cross‑language communication is the holy grail for messaging apps. Apple has announced a similar feature for iOS 19 (translation inside the keyboard), threatening to make Gboard irrelevant in multilingual markets.
• Technical barrier: Achieving low‑latency, high‑quality translation across 100+ languages, each with different tokenization schemes (such as Chinese or Arabic), remains a deep learning challenge. Google's own Universal Dependencies research highlights that morphologically rich languages have 15-20% higher error rates in such settings.

3. Gboard - Emoji Mashup and Sticker Creator (Full Version)

The third feature is perhaps the most mysterious. In an APK teardown from April 2023, 9to5Google revealed a hidden "Emoji Mashup" tool that let you blend two emojis to produce a custom sticker - for example, combining 😂 and 🚀 to make a laughing rocket. The feature appeared fully functional in debug builds, with a dedicated UI and on‑device rendering. Yet over 18 months later, it remains behind a server‑side flag that Google has never flipped for a wide audience. Only a handful of beta testers with the right account IDs have seen it.

Why hasn't Google released this, and a likely reason is moderationWhen users can combine any two Unicode characters, the potential for offensive or system‑generated content skyrockets. Google would need to filter combinations against a dynamic blocklist (e. And g, fire + heart could be interpreted as "burning love," which might be safe. But fire + police car could be problematic in certain contexts). Automating that filter at the scale of billions of combinations is expensive. Another possibility: the feature overlaps with Gboard's existing "Emoji Kitchen" (which launched in 2021 with pre‑designed mashups) and Google doesn't want to confuse users with two competing sticker generators.

From a developer perspective, the Emoji Mashup tool represents a missed opportunity for platform differentiation. Apple's Memoji are tightly integrated with Messages, and Meta has advanced AR sticker creation for WhatsApp. Google's inability to ship a year‑old feature gives the impression that Gboard is an endless beta - a reputation that damages adoption on iOS. Where the default keyboard is already heavily ingrained.

Why Does Google Abandon Features?

The common thread across these three cases isn't technical impossibility but organisational friction. Google employs a "build and iterate" culture where teams compete internally for resources. Features are often greenlit by a product manager eager to impress at I/O. But if the data from early internal testing doesn't show instant engagement, the project is deprioritised. Unlike Apple. Where a VP can force a feature through to a global launch, Google's flat hierarchy allows team leads to kill projects with a single quarterly review.

Moreover, Google has a higher bar for "launch quality" than many assume. The company's internal document, "Launch Checklist v7. 0," requires features to pass automated scalability tests that simulate 10x expected load. For on‑device features like translation, that means verifying performance on every supported device model - a matrix that now exceeds 2000 Android devices. This thoroughness prevents the kind of viral bugs that plague smaller competitors. But it also slows shipping to a crawl.

Finally, there's the "rivalry effect. " When Sundar Pichai announced in 2023 that "every team must justify AI investment," product managers started adding AI polish to existing features to secure headcount. Soon after, Google Photos' Memory creation and Gboard's translation both got AI surface updates, diverting engineering resources away from the original core features. The result: three promising ideas remain perpetually "under consideration. "

The Impact on Developer Trust

For developers building on Google's platform, these delays are a liability. Consider an Android startup that relies on Gboard's inline translation for a travel app. The startup prototypes integration with the unreleased API (exposed in an alpha SDK), then watches as the feature fails to materialise. They must either hack a workaround using Google Translate's old API - which is being deprecated - or abandon the idea. This churn breeds distrust. A 2024 survey by Appfigures found that 40% of Android developers now prefer delaying adoption of new Google APIs until they're in stable release for at least six months.

Furthermore, Google's pattern of showing features then cancelling them gives competitors a blueprint. Apple can observe Gboard's emoji mashup UI in the APK teardown and replicate it in iOS with a faster timeline because they control both the keyboard and the OS. Google's open platform constraints mean that any new Gboard feature must work across OEM skins. Which introduces additional fragmentation. Developers pay the price of this complexity.

What We Can Learn from Google's Feature Pipeline

Despite the frustration, there are lessons for engineers and product managers. First, on‑device machine learning features require a robust A/B testing framework that accounts for device heterogeneity - not just OS