Google AI Edge Gallery Icons and Settings Explained

1. 1

   Menu

   The three lines at the top left. Opens the side menu with Settings, Models and Notifications.

2. 2

   Try ASR & TTS today

   The sound wave marks what is new. Speech recognition (ASR) turns your voice into text, and text-to-speech (TTS) reads replies aloud. The two cards under it open XR Language Teacher and AI Chat, where you can try both.

1. 1

   AI Chat

   The red chat bubbles. Opens a chat with an AI model, such as Gemma, that runs on your phone. You can type, talk to it, and have replies read aloud.

2. 2

   Agent Skills

   The green rocket. A chat where the model can use skills, like showing a map, looking things up on Wikipedia or setting a reminder.

3. 3

   Instant media search

   The blue photo icon. Searches the photos on your phone by describing them in plain words, or finds photos that look like one you pick, without sending anything to the cloud.

4. 4

   Video Moment Finder

   The yellow magnifier. Finds a moment in a video from a description, then lets you trim that clip and save it.

5. 5

   Ask Image

   The red picture icon. Add a photo or take one with the camera, then ask the model questions about it.

6. 6

   Audio Scribe

   The green microphone. Add a short audio clip and the model writes it out as text, or translates it.

Status symbols. These only show information. They are not buttons.

1. 7

   Model count

   How many AI models you can use for that task. You download the one you want before the task runs.

1. 1

   Prompt Lab

   The blue squares. A one-question, one-answer workspace for testing prompts, with ready-made modes like Rewrite tone and Summarise text.

2. 2

   Tiny Garden

   The yellow flower. A small experimental game where you plant and harvest a virtual garden by typing what you want in plain words.

3. 3

   Mobile Actions

   The red sigma. Lets a small on-device model control parts of your phone from simple commands, with no internet needed.

4. 4

   Scrapbook

   The green icon with scissors. Cuts objects out of your photos so you can use the cutouts to make your own scrapbook.

5. 5

   XR Language Teacher

   The blue graduation cap. Point your camera at something and ask about it; the teacher answers in the language you are learning, reads it aloud and saves new words as vocabulary cards.

1. 1

   Back

   The left arrow. Leaves the chat and takes you back to the task page.

2. 2

   Model picker

   The pill under the task name shows the model you are chatting with. Tap it to switch to another model for this task.

3. 3

   Model settings

   The three sliders. Opens Configure settings, where you change things like max tokens, temperature and GPU or CPU for this model.

4. 4

   Chat history

   The clock with an arrow. Opens your chat history, where you can start a new chat or delete old ones.

5. 5

   Read aloud

   The speaker next to a reply. Tap it to hear the reply read out loud.

Status symbols. These only show information. They are not buttons.

1. 6

   Model on GPU

   The label above each reply. It tells you the reply came from the model, and whether it ran on your phone's GPU or CPU.

1. 1

   About live metrics

   The small i at the top right of the panel. Tap it to read what the live metrics mean.

2. 2

   Scroll to bottom

   The down arrow over the chat. Jumps to the newest message.

Status symbols. These only show information. They are not buttons.

1. 3

   Speed

   How fast the model writes, in tokens per second (tok/s). TTFT is the time to the first token of the reply, and Prefill is how fast it read your prompt.

2. 4

   Tokens

   How many tokens went in (your prompt) and came out (the reply) for the last message, and the total.

3. 5

   RAM

   The most memory the app used while answering, and what share of your phone's total memory that is.

4. 6

   Context

   How much of the chat's token limit you have used. The mark at 75% compress is where the app can shrink the chat so you can keep going.

1. 1

   Prompt box

   Type your question or instruction here. The model reads it on your phone, not in the cloud.

2. 2

   Add content

   The plus. Opens a menu to take a picture, pick a photo, record or pick an audio clip, or reuse an earlier prompt.

3. 3

   Toggle live metrics

   The bar chart in a blue circle. Shows or hides the live metrics panel above the prompt box. Blue means the panel is on.

4. 4

   Voice input

   The microphone. Tap it and speak, and your words are typed into the prompt box by on-device speech recognition.

5. 5

   Send

   The paper plane. Sends your prompt to the model.

1. 1

   Take a picture

   Opens the camera so you can take a photo and ask the model about it.

2. 2

   Pick from album

   Adds a photo from your gallery to the prompt.

3. 3

   Record audio clip

   Records a short clip with your microphone and adds it to the prompt.

4. 4

   Pick wav file

   Adds an audio file in WAV format from your phone.

5. 5

   Input history

   Shows the prompts you typed before, so you can reuse one instead of typing it again.

1. 1

   Close

   The X. Closes the chat history panel.

2. 2

   New chat

   Starts a fresh, empty chat with the same model.

3. 3

   Clear all

   Deletes every chat in the list at once.

4. 4

   Earlier chat

   Each card is an earlier chat, shown by the start of its prompt. The note at the bottom says chat history does not persist across demos.

5. 5

   Delete

   The trash can on a card. Deletes just that one chat.

1. 1

   Drag handle

   The short bar at the top of the sheet. Drag it down, or tap outside the sheet, to close the model list.

Status symbols. These only show information. They are not buttons.

1. 2

   Selected

   The tick marks the model the chat is using now. Tap another model to switch to it.

2. 3

   Downloaded

   The down arrow in a filled circle means the model is already on your phone and ready to use offline.

3. 4

   Not downloaded

   The question mark in a circle means the model is not on your phone yet. You need to download it before you can use it.

4. 5

   Model size

   How much storage the model takes on your phone, and how much you download to get it.

1. 1

   Close

   The X. Closes the settings without saving your changes.

2. 2

   Model configs

   The tab you see first. It holds the settings below, which control how the model reads and writes.

3. 3

   System prompt

   Lets you write standing instructions the model follows for the whole chat, like the tone to use or a role to play.

4. 4

   Audio

   Picks the speech recognition and text-to-speech engines, voices and languages used in this chat.

5. 5

   Max tokens

   The most tokens the chat can hold, your messages and the replies together. A token is a short piece of a word. Higher allows longer chats but uses more memory. The range here is 2,000 to 32,000.

6. 6

   TopK

   When writing each word, the model only picks from this many of its most likely next tokens. A lower number keeps answers safer and more predictable; a higher number allows more variety.

7. 7

   TopP

   Another way to limit the choices. The model only picks from the most likely tokens that add up to this share of probability. 0.95 means it ignores the least likely 5%.

1. 1

   Temperature

   How adventurous the answers are. Lower values, like 0.2, give focused and repeatable answers; higher values give more creative and varied ones that can drift off topic.

2. 2

   Visual token budget

   How many tokens the model spends on each image you add. A higher number lets it see more detail but takes longer and uses more memory.

3. 3

   Accelerator

   Runs the model on your phone's GPU or CPU. The GPU is usually faster. The label above each reply shows which one was used.

4. 4

   Enable thinking

   Turn it on and models that support it work through the problem step by step before they answer. Answers take longer but can be better on hard questions.

5. 5

   Restore default

   Puts every setting back to the model's original values.

6. 6

   Cancel

   Closes the settings without saving.

7. 7

   OK

   Saves your changes and closes the settings.

1. 1

   Back

   The left arrow. Takes you back to the home screen.

2. 2

   API Documentation

   The page icon. Opens Google's developer documentation for the on-device API this task is built on.

3. 3

   Example code

   The angle brackets. Opens example code for this task, for developers who want to build the same thing into an app.

1. 1

   More options

   The three dots on a model card. Opens a menu with Benchmark and Delete.

2. 2

   Learn more and see model licence

   The arrow out of a box. Opens the model's page, where you can read about it and its license.

3. 3

   Try it

   Opens the task with this model.

Status symbols. These only show information. They are not buttons.

1. 4

   Best overall

   The star marks the model the app recommends first for this task.

2. 5

   Downloaded

   The filled circle with a down arrow next to the size means the model is already on your phone.

1. 1

   Benchmark

   Opens the benchmark tool, which tests how fast the model runs on your phone. A run takes several minutes, and you can compare runs.

2. 2

   Delete

   Deletes the downloaded model from your phone to free up space. You can download it again later.

1. 1

   Add image

   The plus. Adds a photo from the camera or your gallery. You can add up to 10 images in one chat, then type your question.

Status symbols. These only show information. They are not buttons.

1. 2

   Task name

   The icon and name at the top show which task you are in. The model pill and the send button take the same color as that task's tile.

1. 1

   Add audio clip

   The plus. Adds one audio clip, up to 30 seconds long, by recording it or picking a WAV file. Then ask the model to transcribe or translate it.

2. 2

   Try live metrics!

   Turns on the live metrics panel, which shows how fast the model answers and how much memory it uses.

3. 3

   Voice input

   The microphone dictates your typed prompt, like what to do with the clip. It does not add the clip itself; use the plus for that.

1. 1

   Interactive Map

   A sample prompt. Tap it to try a prompt that asks the model to show the Googleplex on a map, using the interactive-map skill.

2. 2

   Schedule Reminder

   A sample prompt. Tap it to try a prompt that asks the model to set a daily 9 am reminder. Scheduled reminders show up under Notifications in the side menu.

3. 3

   Skills

   Opens Manage skills. The number is how many skills are turned on right now.

4. 4

   MCP

   Opens the MCP server manager. MCP (Model Context Protocol) servers give the model extra tools; the number is how many you have added.

1. 1

   Close

   The X. Closes Manage skills and takes you back to the chat.

2. 2

   Search for a skill

   The magnifier. Type to find a skill in the list by name.

3. 3

   Add

   The plus in a blue circle. Adds a new skill: from a featured list, from a URL, from a folder on your phone, or one you build in the app's skill creator.

4. 4

   Turn on all

   Turns on every skill in the list.

5. 5

   Turn off all

   Turns off every skill, so the model works as a plain chat.

1. 1

   Collapse

   The up arrow next to Built-in Skills. Hides the list of built-in skills; tap again to show it.

2. 2

   Skill switch

   Turns a single skill on or off. A filled blue switch means the model can use that skill in chat.

3. 3

   View

   The eye. Opens the skill so you can see what it does.

1. 1

   Settings

   The blue gear. Opens the app settings: theme, speech recognition engine, Hugging Face token and licenses.

2. 2

   Models

   The green list. Opens a list of all models, where you can download them, try them and run benchmarks.

3. 3

   Notifications

   The red bell. Shows the notifications scheduled from the app, like reminders set in Agent Skills, and lets you delete them.

1. 1

   Theme

   Picks the app's look. Auto follows your phone's light or dark mode; Light and Dark keep it fixed.

2. 2

   Speech Recognition Engine

   Picks the on-device engine that turns your voice into text. Engines your phone cannot use show Unavailable.

3. 3

   Clear

   Removes the saved Hugging Face access token. The app gets one automatically when you download a gated model, one that asks you to accept its terms first.

4. 4

   Enter token manually

   Lets you paste your own Hugging Face access token instead.

5. 5

   View licences

   Shows the licenses of the open-source libraries the app uses.

6. 6

   Close

   Closes Settings.