Travigo

SS — Mon, 16 Mar 2026 22:43:41 +0000

Travel as fast as you speak with Gemini ! Where live agents meet immersive storytelling & 3D navigation

This project was created for entering the Gemini Live Agent Challenge

#GeminiLiveAgentChallenge

Travigo is a next-generation AI Agent that utilizes multimodal inputs and outputs, moving far beyond simple text-in/text-out interactions. The project leverages Google's Gen AI SDK, Gemini Live API, Gemini 3, Google Maps API cloud services with the creative power of generative AI and spatial context to solve complex problems and create entirely new, immersive user experiences in 3D navigation and storytelling.

Features & Functionality

Multimodal Interactions: Communicate via voice and text while the AI processes real-time visual context from the interactive Street View and 3D map spatial data.
Dynamic Personas: Choose between Concierge Mode (realistic local guides) and Game Mode (mystical/run-time personas) adapting tone and narrative focus on the fly. There are also local personas spinned up on the fly by Gemini 3 depending on the user AR location, street view.
Real-time Context Processing: Uses a Live Agent Orchestrator to stream dialogue & voice directly tied to user actions and spatial events.
Immersive Storytelling: Generates contextual narratives overlaid seamlessly onto the UI and 3D environment.

Models Used

The project utilizes a multi-model architecture, leveraging different Gemini models google gen AI SDK depending on the task:

gemini-2.5-flash-native-audio-preview: Used by the Live Agent Orchestrator to power real-time, multimodal conversations via voice and audio streaming.
gemini-2.5-flash: Used for rapid "Scout" queries, specifically grounding location searches using the Google Maps tool.
gemini-3.1-pro-preview: Used for complex deep reasoning tasks via High Thinking levels, such as planning logistics, tours, generating fictional personas based on spatial context and performing deep "Strategic Analysis" (e.g., visa planning, historic deep dives) grounded by Google Search.

SONICS.ai 🎬 💥 🎞️ create character-consistent Comics that 'speak' your Style

SS — Sun, 14 Sep 2025 14:58:40 +0000

__This is a submission for the Google AI Studio Multimodal Challenge

💡 Inspiration

I always wanted to make comics that can capture my chaotic imaginations - but the drawing, erasing, starting again is 👀 such a drag!

Also, AI didn't help much - create - frustrate - regenerate - repeat and yet couldn't get my vibe 🌈 ... 👀 even more drag!

Well that was until ✨ Gemini nano banana gemini-2.5-flash-image-preview!

I am so blown by its editing capabilities specially working with multi-image, multi-modal inputs that I couldn't allow my lazy self to procrastinate anymore !

So, here's (quick links)

🪄 What I built

🎥 Demo Video

🧩 Multimodal app architecture

⚡ How I Used Google AI Studio

✨ Multimodal capabilities I implemented

🚀 Specific multimodal features I built for UX

🎉 Acknowledgement

What I built

SONICS.ai 🪄 is a Google-AI ✨ powered creative suite 🧠 🎬 📚 🎞️ that transforms user's simple idea into a fully-realized, multi-sensory, character-consistent comic book experience with podcast playbacks.

It allows users to add their flavours/ vibes 🌈 to every aspect of comic creation - from storyline to characters to scenes to dialogues to text styles - all in natural language.

The best part? You dont need to be good at drawing! AI solves it for you in ⚡ minutes !

You can bring your creativity to life without losing your patience with back-n-forth regeneration to get that perfect shot!

You can use SONICS for a variety of use cases - from bedtime stories podcast to full production-ready comics with playback

Bring your stories to life - your style!
Let your imagination go wild !

Demo

My project in action 🎥

My project in action

0:00 Intro
0:10 🧠 Story Conception
0:20 🎬 Character/ Cast Design
0:53 🎞️ Comic Panel Creation
1:24 📚 Comic preview
1:34 🎧 Audio preview
1:47 🎥 Play the Comic that speaks your Style

▶️ Play on Youtube

Note : Due to billing constraints, I couldnt deploy my app so this is the video demo 👆 showing my project in action.

How I Used Google AI Studio

This app was entirely built on Google AI studio ⚡vibe-coded from scratch
👀 as you could have guessed by now for my lazy vibes !

I started with a simple idea prompt and kept on adding features by guiding the AI through pain-points I have faced when vibe-creating comics with my flavour.

The Multimodal capabilities I implemented ...

Multimodal Capabilities

Input

Output

Models ✨

Features 🚀

Text

Image

gemini-2.5-flash-image-preview

imagen

For quality Character, Scene Background generation

Text editor based updates

Image + Text

Text

gemini-2.5-flash

Automatic character description updates for natural language based character edits

Image (mask) + Image + Text

Image

gemini-2.5-flash-image-preview

For precise edits in characters/ scenes, dialogue corrections, text stylings, positional edits, detail improvement

Multiple Images + Text

A composite image with rendered text

gemini-2.5-flash-image-preview

For comics scene panel generations ensuring character consistencies across scenes, dailogue accuracy, scene quality

Input	Output	Models ✨	Features 🚀
Text	Image	`gemini-2.5-flash-image-preview` `imagen`	For quality Character, Scene Background generation Text editor based updates
Image + Text	Text	`gemini-2.5-flash`	Automatic character description updates for natural language based character edits
Image (mask) + Image + Text	Image	`gemini-2.5-flash-image-preview`	For precise edits in characters/ scenes, dialogue corrections, text stylings, positional edits, detail improvement
Multiple Images + Text	A composite image with rendered text	`gemini-2.5-flash-image-preview`	For comics scene panel generations ensuring character consistencies across scenes, dailogue accuracy, scene quality

Multimodal Features

The specific Multimodal functionalities 🚀 I built and why it enhances the user experience 👤 (UX)...

Composite scene panels 🎞️

✨ imagen gemini-2.5-flash-image-preview gemini-2.5-flash

🚀 The comic panels are created through an intelligent composition logic combining the multimodal capabilities of the models to create final panel images from the inputs - scene background, character images, scripts that were themsleves generated by using either of these.

👤 This ensures character consistency, dialogue accuracy as well as scene quality across comic scenes.

Flavour edits 🌈

✨ gemini-2.5-flash-image-preview gemini-2.5-flash

🚀 It is used for enabling precise surgical edits of scenes, characters, dialogues, styles leveraging masking.
Users can simply explain their edits in natural language for feature changes (with / without masking).
It also handles auto-updating user edit requests for images which must reflect in their respective strategic texts like character description to ensure further consistencies.

👤 This helps users avoid regenerating back-and-forth images from scratch which was really frustrating when we need to make a small style/ error correction. And users can add their vibes/ flavours/ styles to the scene in natural language without worrying about any inconsistency.

🎉

Acknowledgement

Google AI studio ⚡ is phenomenal at vibe-coding. I was able to generate and finish a well-working prototype in less that 6 hrs.
But as you could have guessed 👀 Parkinson's law took most time !

gemini-2.5-flash-image-preview ✨ (Gemini nano-banana) is the star of my whole idea. Due to nano banana, I was able to successfully create a consistent character comic experience, and solve the back-and-forth regeneration & vibe-check problem for vibe-comic enthusiasts.

imagen ✨ helped me create beautiful backgrounds for the comic scenes which were then fully realised using composite logic.

gemini-2.5-flash ✨ has been used for prompt engineering for inputs to other models, for auto-updating descriptions and also for optimising the deliverables.

Thank you!
It was a fun and great experience!

👀

~~What~~ Definitely Not a drag!

DEV Community: SS