
Weekend Creative Challenge: Bollywood-ify My Day
What if your ordinary day could become a Bollywood blockbuster?... I built Bollywood-ify My Day during the AWS Weekend Creative Challenge, an AI-powered app that turns everyday moments into dramatic movie trailers, posters, stories, and narrated experiences using AWS services.
Series: Weekend Challange (2 articles)
- 1Weekend Creative Challenge: Bollywood-ify My Day This article
What if you could take the most ordinary description of your day and turn it into a Bollywood movie?
"Woke up late, missed the bus, somehow aced my viva."
Normally, that's just a chaotic Tuesday.
But what if AI turned it into a movie trailer, complete with a dramatic title, tagline, poster, narration, and a little Bollywood-level exaggeration?
That was the idea behind Bollywood-ify My Day, the app I built for the AWS Weekend Creative Challenge.
The idea :
Bollywood-ify My Day takes a simple description of your day and transforms it into a mini Bollywood movie experience.
You enter something that happened to you, and the app generates:
- A dramatic movie title and tagline
- A 4-6 line Bollywood-style trailer script
- A cartoon-style movie poster based on the actual scene
- Optional voice narration using Amazon Polly
- Karaoke-style highlighting as the narration plays
- Genre remixes in Horror, Romance, Comedy, Action, or Thriller
- An optional longer story expanding your day into a mini Bollywood plot
One thing I really wanted was for the generated poster to feel connected to the user's story, rather than being a random dramatic image.
So if your input is about running after a bus, the poster
should actually show someone running after a bus. That's where one of the interesting parts of the architecture came in.
How I built it
I started by defining the requirements and architecture before jumping straight into coding. I built and tested the layers independently and then connected them together.
The frontend is built with React, Vite and TypeScript, with a playful purple-pink cinematic theme, animations, and a movie-poster-style presentation.

For the backend, I decided to keep things simple rather than creating multiple microservices. A single AWS Lambda function handles different operations through an
action field such as trailer , remix , story , narrate , and gallery .For text generation, I used Amazon Bedrock with Nova Lite. The model doesn't just generate the trailer. It also returns a structured
visual_scene field describing the key visual moment from the story.That description is then used to generate the poster through Pollinations.ai Flux.
This was actually a workaround I discovered during development.
I initially wanted to use Nova Canvas through Bedrock, but after testing it, I ran into legacy-model restrictions on my account. I also checked the other available image-generation options, but some required marketplace subscriptions or additional payment setup.
So I changed direction and used Pollinations.ai Flux from Lambda through HTTPS. It worked without needing another API key, and more importantly, it let me keep the project within the scope of a weekend demo.
I also learned that** image prompting matters a lot. My first prompts produced images that were much more photorealistic or generic than I wanted.
Changing the prompt structure to describe the specific person and action first, followed by the visual style, produced much better results.

AWS architecture :

Architecture overview: The React frontend sends requests to Amazon API Gateway, which invokes a single AWS Lambda function. The Lambda function routes the request based on the requested action. Amazon Bedrock Nova Lite handles the creative text generation, while Pollinations.ai Flux generates the movie poster from the visual scene produced by Nova Lite. Amazon Polly creates the optional trailer narration and speech-mark metadata used for synchronized highlighting in the frontend. Generated posters and audio are stored in Amazon S3, while DynamoDB stores the generated content needed for the gallery. AWS IAM controls the permissions between these services, and Amazon CloudWatch provides logging for debugging and monitoring.
AWS Service | Purpose | How I Used It |
|---|---|---|
Amazon Bedrock (Nova Lite) | AI text generation | Generated movie titles, taglines, trailer scripts, genre remixes, stories, and the visual_scene used for poster generation |
Amazon Polly | Text-to-speech | Created the dramatic trailer narration using the Matthew neural voice and generated speech-mark timing data |
AWS Lambda | Serverless backend | Handled the application's core logic and routed requests for trailer, remix, story, narration, and gallery actions |
Amazon API Gateway | API layer | Provided HTTP endpoints for communication between the React frontend and Lambda backend |
Amazon S3 | Object storage | Stored generated movie posters and narration audio |
Amazon DynamoDB | NoSQL database | Stored generated content and data used to populate the gallery |
AWS IAM | Access control | Managed permissions between Lambda and services such as Bedrock, S3, Polly, and DynamoDB |
Amazon CloudWatch | Monitoring & logging | Used Lambda logs for debugging, monitoring requests, and troubleshooting issues during development |
I also used S3 prefixes to expose only the generated
posters/* and audio/* objects while keeping the rest of the bucket private.For the narration feature, Polly generates both the MP3 and speech-mark metadata. The frontend uses that timing information to highlight each trailer line as the audio plays.
Cost Overview
Service | Approx. cost per generation | At 1,000 generations |
|---|---|---|
Amazon Bedrock – Nova Lite | ~$0.00014 | ~$0.14 |
Amazon Polly – Neural | ~$0.000008 | ~$0.008 |
AWS Lambda + API Gateway | ~$0* | ~$0* |
Amazon S3 + DynamoDB | Negligible | ~$0.001 |
Pollinations.ai Flux | $0 | $0 |
Vercel | $0 | $0 |
Estimated total | ~$0.00015+ | ~$0.15+ |
A few challenges along the way :
The project definitely wasn't a straight line from idea to finished app.
One unexpected issue was Polly SSML. I initially used
<emphasis> tags with the Matthew neural voice, only to get an InvalidSsmlException . After testing the tags individually, I found that neural voices support a more limited SSML subset. I removed the unsupported tags and kept supported controls such as prosody , break , and p .I also ran into an IAM explicit Deny that was blocking Lambda and Bedrock calls during development. Instead of removing the cost-lockdown policy, I scoped the permissions to only the specific Lambda, Bedrock models, and DynamoDB table I needed.
That turned out to be a useful AWS lesson in itself:
An Allow doesn't override an explicit Deny.
What I learned :
This challenge taught me much more than just how to connect AWS services.
I learned that model availability in Bedrock can depend on the account, so something appearing as available in the console doesn't necessarily mean an actual API call will work.
I learned that prompt structure can matter more than simply choosing a different model. Generating a dedicated
visual_scene alongside the trailer gave the image generator much better context without needing another AI model.I also got a much better understanding of Polly speech marks, SSML limitations, IAM policies, S3 permissions, and how to build a small serverless application without over-engineering it.
Most importantly, I was reminded that building something creative is a really fun way to learn cloud technologies.
This started as a weekend challenge and ended up becoming a tiny Bollywood production pipeline powered by AWS.
Try it / Explore the code
If your day was a movie, what would you call it?
Series: Weekend Challange (2 articles)
- 1Weekend Creative Challenge: Bollywood-ify My Day This article
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article