aws bedrock shell playground

Small bash scripts for experimenting with Amazon Bedrock from the command line.

The project keeps two workflows separate on purpose:

• generate-image.sh for image generation with Amazon Nova Canvas
• generate-text.sh for text generation with Amazon Nova 2 Lite

That split keeps the model contracts explicit and avoids mixing image-generation payloads with text-generation payloads.

• sequential output naming for generated files
• separate scripts for image and text model families
• shell-based tests that do not make live Bedrock calls
• Git Bash support on Windows
• minimal dependencies and no framework setup

• macOS, Linux, or Windows with Git Bash
• bash
• AWS CLI installed
• Bedrock model access enabled in your AWS account
• AWS credentials configured locally
• jq
• base64

1. Install and verify the required tools.

On Windows, use Git Bash.

command -v bash
command -v aws
command -v jq
command -v base64
command -v mktemp

Windows install example:

winget install --id Git.Git -e
winget install --id Amazon.AWSCLI -e
winget install --id jqlang.jq -e

If you use Chocolatey instead of winget:

choco install git awscli jq

2. Configure your AWS credentials if needed.

aws configure

3. Make sure your account can invoke the models used by these scripts.

• amazon.nova-canvas-v1:0
• a Nova 2 Lite inference profile such as us.amazon.nova-2-lite-v1:0

4. Run one of the generators.

./generate-image.sh "A green parrot sitting on a tree branch, tropical jungle, photorealistic, high detail"
./generate-text.sh "Summarize the main differences between REST and GraphQL."

Optional flags:

• --region to override the default AWS region
• --output-dir to choose where generated files are written
• --no-open on the image script to skip opening the output file
• --debug to preserve temp request/response files and print their paths

You can configure the scripts with environment variables instead of passing everything on the command line.

Common:

• AWS_REGION

Image generation:

• MODEL_ID

Text generation:

• TEXT_INFERENCE_PROFILE_ID
• MODEL_ID as an explicit advanced override

Precedence for region:

1. --region
2. AWS_REGION
3. built-in default us-east-1

Example:

export AWS_REGION=us-west-2
export TEXT_INFERENCE_PROFILE_ID=us.amazon.nova-2-lite-v1:0

See .env.example for a minimal template.

.
├── generate-image.sh
├── generate-text.sh
├── images/
├── texts/
└── tests/

Generate an image with Amazon Nova Canvas:

./generate-image.sh "A green parrot sitting on a tree branch, tropical jungle, photorealistic, high detail"
./generate-image.sh --region us-west-2 --output-dir ./tmp-images --no-open "A studio portrait of a red fox"
./generate-image.sh --debug --no-open "A watercolor portrait of a blue jay"

What it does:

• sends the prompt to amazon.nova-canvas-v1:0
• saves the result under images/ as image-0001.png, image-0002.png, and so on
• opens the generated image automatically when a supported OS opener is available

Generate text with Amazon Nova 2 Lite:

./generate-text.sh "Summarize the main differences between REST and GraphQL."
./generate-text.sh --region us-west-2 --output-dir ./tmp-texts "Write a short markdown summary of REST vs GraphQL."
./generate-text.sh --debug "Write a markdown summary of event-driven architecture."

What it does:

• sends the prompt through the configured Nova 2 Lite inference profile
• saves the result under texts/ as response-0001.md, response-0002.md, and so on
• prints the generated response to the terminal

Configuration notes:

• by default the script uses us.amazon.nova-2-lite-v1:0
• you can override it with TEXT_INFERENCE_PROFILE_ID
• the value should be the inference profile ID or ARN used for Nova 2 Lite
• the script still accepts MODEL_ID as an explicit override, but the normal path is to use an inference profile

Image prompt example:

A green parrot sitting on a tree branch, tropical jungle, photorealistic, high detail

Text prompt example:

Summarize the main differences between REST and GraphQL in plain English.

Run the test scripts:

./tests/test-generate-image.sh
./tests/test-generate-text.sh

The tests mock the aws command, so they do not make live Bedrock calls.

Common commands:

make help
make check
make test
make test-image
make test-text

What they do:

• make help shows the available targets
• make check runs shell syntax checks
• make test runs the full test suite
• make test-image runs only the image script tests
• make test-text runs only the text script tests

• Missing Bedrock access: if the AWS CLI is configured but the Bedrock call still fails, verify model access in the Bedrock console for the selected region.
• Inference profile errors: Nova 2 Lite may require an inference profile instead of direct on-demand invocation. Set TEXT_INFERENCE_PROFILE_ID if needed. The script defaults to us.amazon.nova-2-lite-v1:0.
• Headless servers: on EC2 or other non-GUI machines, the image script still saves the PNG and prints its full path even if nothing opens automatically.
• Windows Git Bash: use Git Bash for the current scripts. Verify aws, jq, base64, and mktemp are available there.

• default region is us-east-1
• output folders such as images/ and texts/ are ignored by Git
• each script validates the model ID so image and text requests do not get mixed up
• as of March 25, 2026, Nova 2 Lite may require an inference profile instead of direct on-demand invocation in some Bedrock setups
• Windows support targets Git Bash; PowerShell would require separate scripts

This project is licensed under the MIT License. See LICENSE.

For issues or inquiries, feel free to contact the maintainer:

• Name: Rod Oliveira
• Role: Software Developer
• Email: jrodolfo@gmail.com
• GitHub: https://github.com/jrodolfo
• LinkedIn: https://www.linkedin.com/in/rodoliveira
• Webpage: https://jrodolfo.net