Androidify: Turning selfies into a cute Android bot with AI
Learning Stats
Nível CEFR
Total Words
Unique Words
Difficulty
Legendas (127 segments)
DownloadSPEAKER: We're reimagining the entire OS layer,
putting Gemini right at the core of Android.
CHRISTOPHER KATSAROS: Hi, there.
I'm Christopher from the Android team.
With more devs using AI and with Android being rebuilt with AI
at its core, we thought, why not do the same
with our lovable Android mascot, too?
Just like the operating system, the mascot
has always been something you've loved to play around
with and make your own.
Over the years, there's been lots
of iterations of the Android bot,
including a few years ago when we launched Androidify
to let you make your own bot.
So we asked ourselves, how would we rebuild this experience today
using AI?
And the answer, of course, was with a selfie.
It was a bit more complicated than that.
So I spoke with Avneet to tell us
how the team used AI to translate
that selfie into a bot.
Take a listen.
AVNEET SINGH: We knew we wanted to use a selfie.
But how we go from a selfie to a bot was a fun challenge.
We decided that first, we had to validate that the image could
work, and that it was of a person with enough detail
to generate a bot.
Then we had to turn that image into a descriptive prompt.
And Gemini 2.5 Flash, well, the Firebase AI logic SDK
helped us do that.
It captioned the photo and also created a structured output.
And then there was the actual creation of the bot.
And for that, we used Imagen 3.
There are lots of details to getting the Android
bot recognizable and iconic.
So we did have to train the model.
The Imagen 3 model was fine tuned using Supervised Fine
Tuning.
SFT is a fine-tuning technique that directly updates
all of the model's weights.
This allows for comprehensive changes,
enabling the model to learn and encode new detail patterns
and produce highly specific and visually appealing
images of Android bots.
We ran a fine tuning pipeline on the Imagen 3 model,
with Android bot assets of different color combinations
and different assets for enhanced cuteness and fun.
We generated text captions for the training images.
And the Imagen text pairs were then used to fine tune the model
effectively.
Next, we wanted to give your bot a vibe and make it dance.
To do this, we used Veo 3, Google's latest AI
video generation model, to animate the path.
To simplify the process, we chose 10 different vibes,
which are interest-based backgrounds,
and wrote specific prompts for each one that you can select.
The experience then takes your created bot, combines it
with the vibe you chose, and Veo 3 animates it
into a fun, eight-second video.
CHRISTOPHER KATSAROS: Speaking of getting the model right,
you may have noticed that since the last time we had Androidify,
the Android bot went from 2D to 3D.
Getting that to look just right also took a lot of work.
And for that, I spoke with Robbie on the Brand team.
ROBBIE MCLACHLAN: To train the model,
we needed lots of different versions of the Android 3D bot.
So we made them for ourselves and people in the team.
And in the process, we learned a lot
about what people care about.
Some wanted bots which looked like exact mini
versions of themselves.
And others wanted to just show their personality or interest.
So we made hundreds, each with a variety of different clothing
and hairstyles and accessories, all with different textures
and colors.
- When using this data to train the AI,
we saw a lot of stuff that did work and a lot of stuff that
didn't work.
But once we were happy, it then came
to designing the vibes for the short videos.
As mentioned earlier, the vibes are like visual references
that Veo can use to then create the short animations.
We designed them based on people's hobbies or interests,
but still making them feel very Android--
expressive, colorful, and most importantly, fun.
CHRISTOPHER KATSAROS: Once we were ready,
we needed to put it together for users with an app.
Rebecca Franks on the Android Developer Relations team
helped drive the app.
And she started by building, of course, with Jetpack Compose.
REBECCA FRANKS: The user interface of the Androidify app
is built using Jetpack Composed, our recommended modern Android
UI Toolkit.
The app has many delightful details built into the UI.
For example, the app uses the latest alpha of Material 3,
which unlocks the newest design features.
In this sample, we use many of the new Material 3 expressive
concepts, such as dynamic shape morphing and new componentry,
such as expressive button groups.
We also take advantage of the new motion scheme
animations big variables whenever an animation spec is
required throughout the app.
To create the selfie experience that Avneet spoke about earlier,
the app integrates with the Compose CameraX Jetpack Library
to do so.
This makes camera app development way easier
in Compose.
The app also uses a custom camera layout, Composable,
which supports all the typical composables that a camera
preview screen would include.
For example, a zoom, a capture, and a flip camera button,
all in a reusable layout that can
be reused across form factors.
The app functions seamlessly across a range
of different devices, from candybar phones
to foldables and tablets.
This was achieved by extracting out reusable layouts
and leveraging APIs such as the window-size classes
to determine what layout to display.
CHRISTOPHER KATSAROS: So that's Androidify, a brand new version
so you can make your very own Android with a selfie
and Google's AI tools.
You can try it out for yourself.
Download the app on Google Play, or go to Androidify.com.
And check out the sample code on GitHub to see how we made it.
We can't wait to see your dancing Android bots.
[MUSIC PLAYING]
Key Vocabulary (50)
toward
"Go to school."
belonging
"Cup of tea."
also
"You and me."
inside
"In the house."
specific
"That book."
A third-person singular pronoun used to refer to an object, animal, or situation that has already been mentioned or is clear from context. It is also frequently used as a dummy subject to talk about time, weather, or distance.
Used to show who is intended to have or use something, or to explain the purpose or reason for an action. It is also frequently used to indicate a specific duration of time.
A preposition used to indicate that something is in a position above and supported by a surface. It is also used to indicate a specific day or date, or to show that a device is functioning.
A preposition used to indicate that people or things are together, in the same place, or performing an action together. It can also describe the instrument used to perform an action or a characteristic that someone or something has.
A conjunction used to compare two things that are equal in some way. It is most commonly used in the pattern 'as + adjective/adverb + as' to show similarity.
Used to refer to the person or people that the speaker is addressing. It is the second-person pronoun used for both singular and plural subjects and objects.
Used to identify a specific person, thing, or idea that is physically close to the speaker or has just been mentioned. It can also refer to the present time or a situation that is currently happening.
Used to indicate the starting point, source, or origin of something. It can describe a physical location, a point in time, or the person who sent or gave an item.
The word 'we' is a first-person plural pronoun used to refer to the speaker and one or more other people collectively. It is used as the subject of a sentence or clause.
Used to refer to the whole quantity or amount of something, or to every member of a group. It indicates that nothing has been left out from the total being discussed.
Sign up to unlock full features
Track progress, save vocabulary, and practice exercises
Description
Turn your selfie into a cute Android bot! Discover the story behind the rebuilt Androidify app and how it harnesses the power of Gemini and Jetpack Compose. Learn about the design process and...
Categorias
Vídeos relacionados
Bring your most expressive apps to the wrist: Wear OS 6 and the new Pixel Watch 4
The Future of AI is Open [Advertiser content from IBM]
The Genius Spiders Changing How We Think About Brains
What The Media Got Wrong About Octopus Cities
All Penguins Are From Australia
English with Lucy
