
Weekend Creative Challenge: VibeVerse — AI Mood-to-Music Story Generator
Discover how to build a serverless, multi-sensory AI application that translates user moods into custom creative stories, color worlds, and music playlists. The project showcases how to orchestrate Amazon Bedrock (Nova Lite), Amazon Polly, and AWS Lambda with zero-dependency frontend hosting on AWS Amplify.
Vision & What the App Does
Have you ever felt a powerful emotion — a rush of joy, a quiet sadness, a restless 3am energy — and wished you could just live inside that feeling for a moment, expressed beautifully?
That's exactly what VibeVerse does.
You type in a mood, a feeling, or a life moment — "I just got rejected from my dream job," "first morning of summer vacation," or "falling in love and everything feels surreal" — and VibeVerse generates a world around that feeling:
- A personalized story or prose poem written specifically for your mood — vivid, emotional, and human
- A curated playlist concept with an evocative title and five track themes that match the vibe
- A color palette of 4 hex codes and a poetic description of what those colors feel
- An audio narration of your story, read aloud in a warm voice
The result is a beautiful "vibe card" — a creative artifact of your emotional moment. You can close your eyes and listen to your story, stare at the colors, and feel understood by a machine for a few minutes. That's the magic of VibeVerse.
How I Built It
The Idea
I wanted to build something that felt genuinely creative — not just a chatbot wrapper. The mood-to-story idea came from a simple question: what if AI could translate emotions into multi-sensory experiences? Words, color, and sound together feel much richer than any one alone.
Development Process
I started with the Lambda backend first. The key challenge was prompt engineering — getting Amazon Bedrock's Nova Lite model to consistently return structured JSON (story + playlist + palette + vibe tag) without markdown formatting breaking the JSON parse. My solution was a strict system prompt that explicitly says "respond with ONLY valid JSON, no markdown," combined with a defensive fence-stripping parser on the Python side.
For the color palette, I found that asking the model to generate hex codes alongside a poetic description worked surprisingly well — the model naturally associates emotional tones with color temperatures (warm ambers for nostalgia, cool blues for melancholy).
The Amazon Polly TTS integration was the smoothest part. The neural "Joanna" voice has a warmth that makes prose feel genuine rather than robotic. I return the audio as a base64-encoded MP3 directly in the API response, so the frontend doesn't need a second round trip or S3.
Challenges
The biggest challenge was JSON consistency from the LLM. Sometimes the model would add a preamble like "Here is the JSON:" before the response. I handled this with a strip-and-split approach in Python. A temperature of 0.92 gave me creative, varied stories while staying coherent.
Another challenge: Polly has a 3,000-character limit per
SynthesizeSpeech call. I added a slice at 2,999 characters to prevent silent failures.
AWS Services Used / Architecture Overview
Browser (Amplify Hosted) │ │ POST / ▼AWS Lambda (Python 3.12, Function URL) │ ├──► Amazon Bedrock (Nova Lite) │ • Mood → story + playlist + palette + vibe tag │ • Structured JSON output via system prompt │ └──► Amazon Polly (Neural TTS) • Story text → MP3 audio • Returned as base64 in API responseService | Role | Free Tier |
|---|---|---|
Amazon Bedrock (Nova Lite) | Story + playlist + palette generation | ~$0.0002/call |
AWS Lambda | Serverless backend, Function URL | 1M requests/month free |
Amazon Polly | Neural text-to-speech narration | 5M chars/month free |
AWS Amplify | Frontend hosting (static HTML/JS) | 15 GB/month free |
No API Gateway, no database, no persistent storage — beautifully simple and cost-effective.
What I Learned
1. Nova Lite is impressively creative. I expected a "lite" model to produce generic output, but with the right system prompt and a high temperature, the stories it generated were genuinely moving. The model understood emotional nuance — "3am restlessness" produced very different output than "peaceful morning."
2. Function URLs are underrated. Skipping API Gateway entirely and using a Lambda Function URL simplified the architecture dramatically. CORS configuration is handled natively without extra infrastructure.
3. Multi-modal output elevates the experience. Combining text + color + audio made VibeVerse feel like a product, not a demo. The audio especially — hearing a story about your own mood read aloud — creates an unexpected emotional resonance.
4. Prompt engineering is engineering. Crafting the system prompt took as long as writing the Python code. Being explicit, structured, and anticipating model failure modes (markdown leaking into JSON, long preambles) is a real skill.
Try It / Source Code
Built for the AWS Weekend Creative Challenge — August 14–17, 2026.
Powered by Amazon Bedrock, Amazon Polly, AWS Lambda, and AWS Amplify.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article