Introduction

When a team needs to optimize and deliver images at scale, the question is often asked the wrong way: Do we use a SaaS or build it ourselves? The right question is different: What level of control, flexibility, and responsibility are we willing to assume? An image in production is no longer just a static file. It is performance, cost, user experience, and architecture.

Background / The pain points

In many projects, the image system starts simple: store in S3 and serve from a CDN. But growth brings new problems:

• The need for dynamic resizing.
• Optimization by format (WebP, AVIF).
• Automatic compression based on headers.
• Smart cache handling.
• Cost control during traffic spikes.
• This is where the dilemma appears:

SaaS: fast and simple, but with variable costs and external dependency. Self-managed infrastructure: greater control, but greater technical responsibility.

Core idea

Building an infrastructure based on Dynamic Image Transformation for Amazon CloudFront means accepting complexity in exchange for autonomy. It’s not just about processing images with Lambda and the sharp library. It’s about designing a complete system that:

• Leverages serverless architecture.
• Scales on demand.
• Minimizes costs when there’s no traffic.
• Allows advanced customization.
• Maintains high availability under AWS SLAs.

Development

Architecture: full control vs SaaS simplicity

The CloudFront + S3 Object Lambda + Lambda solution enables:

• Dynamic resizing via URL.
• Automatic format conversion (AUTO_WEBP).
• On-demand quality adjustment.
• Intelligent edge caching.
• Integration with multiple internal services.

But it requires:

• Deep AWS expertise.
• Runtime and update management.
• Understanding limits like concurrency and payload.
• Continuous maintenance.
• SaaS removes operational friction, but limits customization and can create structural dependency.

Scalability and cost model

One of the biggest draws of the serverless approach is the pay-per-request model. When there’s no processing, there’s no Lambda execution cost. According to official estimates (US East — Northern Virginia):

• 100,000 new uncached images ≈ $4.98 per month.
• 1,000,000 new images ≈ $19.26.
• 1,000,000 cached images ≈ $7.65.

Edge caching in CloudFront radically changes the cost structure. However:

• Poor cache configuration can spike consumption.
• Excessive customizations can increase execution time.
• Higher traffic increases monitoring complexity.

Customization vs maintainability

The infrastructure allows you to:

• Modify Lambda code.
• Add watermarks.
• Adjust transformation logic.
• Integrate validations or advanced security.

But each customization:

• Increases technical debt.
• Complicates future upgrades.
• Can break compatibility with new service versions.

Observability and operations

A self-managed system forces you to handle:

• CloudWatch Logs.
• Execution metrics.
• Lambda concurrency.
• S3 events.
• Secrets management for Signed URLs (e.g., Secrets Manager).

It’s not just “works or doesn’t work.” It’s understanding the path each request follows: Client → CloudFront → normalization → cache → S3 Object Lambda → processing → response → edge cache.

Recommendations

• First evaluate expected volume and traffic patterns before choosing architecture.
• Design a caching strategy from day one; it impacts costs more than the infrastructure itself.
• Avoid over-customizing Lambda unless strictly necessary.
• Calculate costs considering uncached images and peak scenarios.
• If the team lacks AWS mastery, include learning costs in the budget.

Conclusions

Building a custom image optimization infrastructure on AWS turns a simple problem into a deep architectural decision. It’s not just about delivering images it’s deciding how much control you want over performance, costs, and future evolution. Serverless architecture offers scalability, flexibility, and efficiency, but requires technical judgment and ongoing governance. The true trade-off isn’t about technology; it’s about strategy.

Glossary

• Serverless: A model where the provider manages infrastructure and you pay per execution.
• S3 Object Lambda: A service that allows objects in S3 to be modified in real time during read operations.
• Edge caching: Storing content in locations close to end users.
• Payload limit: The maximum data size a function can process.
• High availability: The system’s ability to remain operational under normal and failure conditions.