Creating a Flood Awareness PSA with AWS Nova Canvas

I decided to experiment with AWS Nova Canvas to generate the visual components for a a public service announcement about flood risks, As someone without professional animation skills, an AI-powered solution offered an innovative pathway to produce quality visuals.

Technical Setup Challenges

Accessing the tool presented my first hurdle. Since All AWS Nova Foundational Models are currently limited to U.S. users, I initially attempted to work through the AWS Bedrock web console interface playground. This experience proved frustratingly inefficient,simple image generations took upwards of thirty seconds each, and producing a basic six second video test consumed nearly half an hour of processing time. This impractical performance led me to pivot to programmatic API access, which delivered significantly improved response times.

To establish a workable environment, I configured an AWS account with appropriate Bedrock permissions, specifically requesting access to the Nova Canvas model. After setting up the necessary credentials and Python environment, I developed a streamlined local web interface to expedite my prompt testing process.

Find the GitHub repository of the simple web UI i built to test out: WebUI_AWS_NOVA_Canvas

Perfecting the Visual Components

The image generation process demanded precise prompting techniques. My PSA required specific disaster visuals like flooded neighborhoods and emergency scenarios, so I crafted detailed descriptions. For instance, I would request "a street with brown floodwaters rising halfway up parked cars during heavy rainfall." While Nova Canvas successfully produced serviceable stock style images, achieving consistent quality required iteration.

Several recurring challenges emerged during generation:

• Spatial relationship misinterpretations (like rainfall appearing inside rooms rather than through windows)

• Inconsistent lighting across related scenes

• Occasional anatomical anomalies in human figures

• Difficulty maintaining visual continuity across different prompts

  

Final Production Process

For the completed PSA, I sequenced the most successful generated images in Canva, incorporating smooth transitions and text overlays to create narrative cohesion. AWS Polly handled the voiceover component, converting my script into natural sounding narration that complemented the visual pacing. The entire process yielded a professional 1 minute PSA that effectively communicated critical flood safety information.

Conclusion and Insights

Reflecting on this experience, Nova Canvas proved valuable for this rapid prototype project. While the tool has limitations in processing speed and visual precision, it offers a viable solution for creating compelling visual content without extensive design resources. Key takeaways include the importance of meticulously crafted prompts, the substantial advantage of API access over web interfaces for production work, and the necessity of patience when collaborating with generative AI tools.

Lets Connect - Lewis Sawe: LinkedIn
Buy me coffee ☕