AWS Builder Center
Weekend Creative Challenge: Memory Postcard

Weekend Creative Challenge: Memory Postcard

Turn your written memories into AI-generated illustrated postcards powered by Amazon Bedrock Nova models on AWS.

Vision & what the app does

Everyone has a handful of memories that feel like they deserve more than a text message retelling — a summer afternoon, a specific smell, a conversation with someone you don't get to see anymore. Memory Postcard takes one of those memories, written in the user's own words, and turns it into a keepsake: a short poetic retelling, an evocative title, a standalone quote line, and a fully original illustration — composed together into a postcard you can download and keep.
The idea wasn't to build "another AI image generator" or "another AI writing tool." Plenty of those exist. The goal was to produce a finished artifact — something with the emotional shape of an actual postcard, not just a raw model output. The user stays in control of the feeling through a few simple inputs — mood (Nostalgic, Happy, Melancholic, Peaceful, Funny), visual style (Vintage, Film, Watercolor, Illustrated, Minimal), plus optional location and time — while the AI handles the creative transformation underneath.

How I built it

I separated the two things a "creative app" actually needs: a functioning generation pipeline, and a presentation layer that makes the output feel like something worth keeping, not just a JSON blob on screen.
Key decisions: Amazon Nova Lite handles the text — title, poem, quote, and critically, the image prompt itself, generated from the memory rather than hand-written. Amazon Nova Canvas then turns that prompt into the illustration. Letting the text model write the image prompt kept the artwork thematically tied to the actual memory instead of a generic style match. For the backend, I used a Lambda Function URL instead of API Gateway — with a hard weekend deadline, fewer moving pieces meant fewer places to fail. Generated images are written to S3 and served back via a time-limited presigned URL, so there's no persistent image hosting to manage.
Challenges: The first real challenge was self-inflicted — I built and wired the entire frontend and backend locally before actually provisioning anything in AWS, so early on the app was quietly rendering placeholder artwork with no real backend behind it. The fix was procedural: deploy the infrastructure first, confirm Bedrock model access was granted, and only then trust what the UI was showing. Once real Lambda invocations were hitting Bedrock, the output stopped repeating and started reflecting each memory individually.

AWS services used / architecture overview

  • Amazon Bedrock (Nova Lite + Nova Canvas) — text and image generation
  • AWS Lambda — serverless backend, orchestrates both model calls per request
  • Amazon S3 — stores generated postcard artwork, served via presigned URLs
  • AWS Amplify — hosts the static frontend
User's memory + mood/style/location/time
│
▼
AWS Lambda (Function URL)
│
┌──────────┴──────────┐
▼ ▼
Bedrock: Nova Lite Bedrock: Nova Canvas
(title, poem, quote, (illustration, using
image prompt) the generated prompt)
│ │
└──────────┬───────────┘
▼
Amazon S3 (image stored,
presigned URL generated)
│
▼
JSON response → Frontend
(hosted on AWS Amplify)
│
▼
Rendered postcard, downloadable

What I learned

Chaining two Bedrock models in one request — one generating the prompt for the other — was new territory for me, and it's a pattern I'll reuse: letting a text model translate unstructured, personal input into a well-formed prompt produces noticeably more coherent results than prompting the image model directly from raw user text. I also got hands-on with Lambda Function URLs for the first time, and a good reminder to always verify a "successful" deploy against real infrastructure, not just how the frontend looks.

Link to app or repo

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