AWS Builder Center
MoodCanvas: A Self-Improving Daily Art & Poetry Agent Using Amazon Nova Lite

MoodCanvas: A Self-Improving Daily Art & Poetry Agent Using Amazon Nova Lite

An AI-powered daily creative agent that generates unique poems and art concepts themed to the weather, date, and community prompts — and evolves its artistic style over time based on user feedback. Built 100% serverless with Amazon Nova Lite on Bedrock.

MoodCanvas is an AI-powered daily creative agent that generates a unique poem and visual art concept every day — themed to the weather, date, and community prompts. The twist? It learns from user feedback and evolves its artistic style over time. All generation is powered by Amazon Nova Lite (us.amazon.nova-lite-v1:0) on Amazon Bedrock via the Converse API.
PROBLEM STATEMENT
Most generative AI tools produce random, context-free output. They don't know what day it is, what the weather feels like, or what the user's taste is. The output yesterday looks the same as today. Can we build a creative agent that produces output uniquely tied to this moment — and that genuinely improves over time?
SOLUTION
MoodCanvas is a daily creative agent that:
  • Knows what day it is (Monday's fresh energy vs. Friday's celebration)
  • Feels the weather (rainy = melancholic poetry, sunny = vibrant)
  • Listens to the community (crowd-sourced creative prompts)
  • Self-improves based on user feedback (thumbs up/thumbs down)
HOW IT WORKS
An EventBridge rule triggers the Daily Generator Lambda at 8:00 AM UTC every day. The Lambda gathers context (date, season, day mood, live weather from OpenWeatherMap, a random community prompt, and the current Style DNA), then invokes Amazon Nova Lite via the Bedrock Converse API to generate a poem (4-12 lines), an art concept (detailed visual description), a text-to-image prompt, a color palette (5 hex codes), and mood tags.
KEY INNOVATION: SELF-IMPROVING STYLE DNA
The agent maintains a Style DNA — a JSON file that defines its artistic personality including preferred tones, color tendencies, poetry style, art style, complexity level, and abstraction level. When users give feedback, the system tracks patterns. Every 5 interactions, it triggers the Style Evolver — which uses Nova Lite itself to analyze what worked vs. what didn't, and proposes small evolutionary changes. The agent uses AI to improve AI.
ARCHITECTURE
The entire application is 100% serverless:
  • Amazon Bedrock (Nova Lite) — All text generation + style evolution
  • AWS Lambda — 4 functions (generator, feedback, prompts, evolver)
  • DynamoDB — Store creations + community prompts
  • S3 — Style DNA storage + frontend hosting
  • EventBridge — Daily 8 AM trigger
  • API Gateway — REST endpoints
Estimated monthly cost: less than $1.
AMAZON NOVA LITE INTEGRATION
Nova Lite is invoked in two distinct contexts:
  1. Daily Generation — creative output (poems, art concepts) with temperature 0.8 for creativity
  1. Style Evolution — analytical introspection on feedback patterns with temperature 0.4 for consistency
This dual-use of the same model for both creation and self-reflection is a key architectural decision.
CHALLENGE CRITERIA MET
  • Daily-art or daily-poem agent: Generates BOTH daily art concepts + poems
  • Themes output to the day or weather: Uses date context + live OpenWeatherMap API
  • Remixes community prompts: Users submit prompts; agent blends one into each creation
  • Improves style over time: Style DNA evolves with feedback via Nova Lite self-analysis
  • Uses Amazon Nova Lite (Bedrock): ALL generation via us.amazon.nova-lite-v1:0 Converse API
LINKS
CONCLUSION
MoodCanvas demonstrates that generative AI can go beyond random output — it can be contextual, collaborative, and self-improving. By combining Amazon Nova Lite's fast creative generation with a feedback-driven evolution loop, we get an AI agent that genuinely grows with its users. The entire project is open source, serverless, costs under $1/month, and deploys in a single SAM command.
"Every day is a canvas. MoodCanvas just helps you see it."
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article