This is a submission for the Built with Google Gemini: Writing Challenge
What I Built with Google Gemini:
My project isn't a traditional "build" in the sense of writing code for a new application. Instead, I built a novel interaction with Google Gemini, specifically through the "Emojify" feature on my Google Pixel 10 keyboard. My "build" was essentially demonstrating and exploring how a widely accessible AI feature can generate entirely new, non-standard visual communication..
The "problem" it solved (or rather, illuminated) is the limitation of pre-set, static emoji sets. While existing emojis are vast, they can't cover every nuanced concept. My interaction with Gemini showcased its ability to dynamically create a custom visual to perfectly fit a specific textual context, transcending the boundaries of Unicode.
Here's what happened:
I typed a simple message: "Mums making banoffee pancakes for My birthday". I then asked my Pixel keyboard's built-in "Emojify" function to enhance the text. To my astonishment, alongside standard emojis like 🥞🎉🎂🥳, it generated a completely unique, black-and-white icon resembling a box with three female figures and a nondescript figure inside. This wasn't a standard emoji; it was a novel graphical creation.
Google Gemini played the role of the generative engine embedded within my keyboard. It analyzed my text, identified a semantic gap not perfectly filled by existing emojis (e.g., "a group of mothers/female attendees"), and then synthesized a brand-new visual to represent that concept on the fly.
This wasn't through a dedicated generative AI interface; it was a seamless, on-device AI capability integrated into a common keyboard function.
I have since managed to get it to generate two different custom emojis, on seperate occassions, further proving its capability. (Find 2 images below showing the unique emojis!).
What I Learned:
This experience has been incredibly eye-opening, both technically and philosophically.
Going Beyond the Standard Emoji Table (and My Reaction to It): My biggest ponderance and a major learning point is how it was able to go beyond the standard emoji table of preset characters (Unicode/ASCII). Emojis are typically strict templates of codes, already preset and preloaded. Gemini's Emojify function went out of its preset and preloaded list to generate something entirely new. When I saw this custom visual, my immediate reaction was confusion. I automatically checked my keyboard's emoji list, particularly the "Family" section, and confirmed that the generated icon was not available there. The standard family emojis (e.g., "Family: Adult, Adult, Child") clearly shows they are distinct and generic. This solidifies that the AI truly generated a novel image, rather than pulling from existing assets, and profoundly changed my understanding of what a keyboard's "emojify" function is capable of.
AI's Gender Inference in Generative Output:
A particularly profound aspect is that the AI's generative feature assumed my gender from a simple statement and correctly depicted it in the image, on two separate occasions. From the phrases "Mums making banoffee pancakes for My birthday," and "Mum made banoffi pancakes for my birthday" the AI not only created a novel icon for "a group of mothers/family" but specifically depicted female figures within that icon. This moves into a critical area of AI's ability to infer user characteristics from seemingly neutral text and then embed those assumptions into its generative outputs. It raises important questions about the AI's training data, its inferential capabilities, and the implications of such automated assumptions.
Beyond Retrieval to Generation in Everyday Tools:
I learned that advanced generative AI is no longer confined to large, cloud-based models or dedicated creative applications. It's actively being deployed and integrated into common, on-device user interfaces like a smartphone keyboard. This completely shifts my understanding of what a "keyboard" can do.
The Power of User Interaction:
My simple sentence, combined with Gemini's Emojify feature, pushed the system to generate something unprecedented. It highlighted that users aren't just consumers of AI, but active co-creators who can unlock unforeseen capabilities through natural language interaction.
Dynamic and Adaptive Interfaces:
Perhaps the most surprising discovery was that after I generated this unique emoji, the keyboard's emoji interface itself seemed to update, presenting other "similar but not the same" black and white symbols. This suggests a live, adaptive learning or dynamic asset loading system, where the AI not only generates but also influences its own presentation based on user input and successful creations. It's like the AI learns from its own output and my engagement.
New IP Questions:
This experience has also deepened my understanding of intellectual property in the age of AI. If an AI generates a unique visual in response to my prompt from within a standard tool, where does the ownership lie? It highlights the crucial role of the human operator in guiding the AI to unique creation. My interaction effectively "forced a creation" that didn't exist before.
The "Uniqueness" of Human-AI Collaboration:
Every user is unique, and my interactions proved that this individuality can lead to truly unique outputs from AI, rather than just generic responses. My personal approach unlocked this feature.
Google Gemini Feedback:
What worked well;
Seamless Integration:
The fact that this generative capability is integrated or how it came to be, directly into the keyboard's "Emojify" function is phenomenal. It makes advanced AI accessible and intuitive for everyday communication.
Contextual Understanding:
Gemini's ability to interpret a nuanced phrase like "Mums making banoffee pancakes for My birthday" and infer the need for a "group of female attendees" emoji visual was incredibly impressive.
Novelty Generation:
The core strength demonstrated here is its capacity to create rather than just retrieve. This is a game-changer for personalized visual communication.
Adaptive Learning/Interface:
The apparent dynamic update of the emoji interface after my generation was an unexpected and exciting feature, suggesting a deeper level of intelligence and responsiveness.
Where I ran into friction or needed more support:
Lack of Transparency/Control:
While the discovery was thrilling, the process felt somewhat opaque. There's no clear indication or control for the user to know when they are prompting for a novel generation versus just retrieving an existing emoji. A toggle or a visual cue indicating "generating custom emoji" could enhance the user experience and understanding.
Understanding the "How" and "Why":
I want to understand why the AI "felt" the need to generate a totally new emoji based off my interaction and input, rather than suggesting an existing one. Furthermore, clarity on how this custom emoji was generated by the Gemini function of the keypad built into the Google Pixel 10 (e.g., on-device vs. cloud processing, the specific generative model used) would be incredibly insightful for power users and developers. This relates directly to the "going beyond the standard table" point.
Ethical Considerations of Inference:
The AI's ability to infer my gender and reflect it in a generated image (even if correct in this instance) raises questions about privacy, assumptions, and potential biases embedded in its training data. Providing users with more insight or control over such inferences in generative outputs would be valuable.
Saving/Sharing Custom Emojis:
Since these aren't standard Unicode, there's no easy way to save, share, or consistently reuse these truly custom creations across platforms or even within the same device once generated. Implementing a way to "pin" or export unique generated graphics would be a huge step forward.
Consistency:
I wonder about the consistency of such custom generations. If I were to try to generate the exact same black and white box with four female figures again, would it produce it, or a variation? Understanding this consistency would be key for users.
This experience has shown me the incredible potential of Gemini and Google AI - to revolutionize how we communicate visually, going far beyond simple text. I believe this capability, refined and made more transparent, will be a cornerstone of future AI-powered interface's. Is this the beginning of Fluid AI?
Thank you.
Megan Lawther 26.02.2026
.
Top comments (0)