Crystal Clear Memories

Automating 2D to 3D Model Creation for Laser Engraving

Key results

    • 85% automation of manufacturing-ready 3D model creation
    • Faster turnaround times for customers
    • Consistent high-quality output with human-in-the-loop oversight
    • Scalable pipeline capable of handling higher order volumes
Automating 2D to 3D Model Creation for Laser Engraving
Services SERVICES
Computer Vision
Generative AI
Industry INDUSTRY
Manufacturing
Retail
Technologies TECHNOLOGIES USED
Mono depth estimation
Image segmentation
Face detection
Inpainting
3D texturing
Team TEAM
4 AI engineers
Location LOCATION
USA
Duration PROJECT DURATION
11 months

“It-Jim helped us integrate computer vision and 3D GenAI into our image processing pipeline, saving thousands in manual design work. Their team delivered precise, scalable solutions that transformed how we create our custom 3D crystals.”

About the client

About the client

Crystal Clear Memories is a U.S.-based company specializing in custom 3D laser-engraved crystals. Customers upload personal photos, often of loved ones or pets, which are transformed into detailed 3D portraits and engraved inside crystal blocks.

Their products are cherished as gifts and memorials, where visual quality is central to the experience.

Before partnering with It-Jim, the creation of 3D models was entirely manual. A small team of skilled artists sculpted each photo into a format suitable for engraving, which made the process slow and difficult to scale.

The Challenges & Goals

Crystal Clear Memories set out to automate the creation of 3D engraving models from customer photos – a task that required solving several complex challenges:

Manual 3D modeling workflow

Skilled artists needed hours to sculpt each photo into an engraving-ready model, creating a production bottleneck.

Limited scalability

The design team ranged from three to six artists, and at times only one person handled all incoming orders.

Consistent premium quality

Automated outputs had to match the craftsmanship of hand-made models to preserve the brand’s reputation.

High operational costs

Relying on expert artists for every model increased expenses and constrained growth.

Long turnaround times

Customers had to wait days for their orders, which was especially critical for gifts and memorials.

Solution Overview

To help Crystal Clear Memories scale production, we built an AI-powered pipeline that transforms 2D customer photos into manufacturing-ready 3D models for laser engraving. The system combines subject isolation, 3D reconstruction, and post-processing to meet high quality standards. Now the automation handles most orders, while human review covers complex cases, enabling faster turnaround, reduced costs, and higher order capacity without compromising craftsmanship.

Stage 1: Defining the Path to Automation

When Crystal Clear Memories approached It-Jim, their objective was to remove the manual bottleneck in creating 3D engraving models from customer photos. The process was clear in theory – take a flat image and convert it into a 3D form – but making it fast, reliable, and manufacturing-ready required rethinking the entire workflow.

Key initial steps by It-Jim:

  • Mapped the production pipeline to identify where automation could bring the most impact.
  • Broke the workflow into two critical tasks: isolating the subject from the background and reconstructing it as a 3D model.
  • Designed a plan to combine custom and off-the-shelf neural networks with post-processing to refine outputs.

This first milestone established a clear roadmap for automation and laid the foundation for building an AI-powered 3D model generation pipeline.

Image

Stage 2: Building the First Automation Prototype

With the workflow defined, we moved into prototyping an AI-powered system that could take customer photos and automatically generate engraving-ready 3D models. The goal was to replace hours of manual sculpting with a process that ran in minutes while still meeting the brand’s high quality standards.

How we approached it:

  • Implemented an ensemble of custom and off-the-shelf neural networks to handle subject isolation and 3D reconstruction.
  • Added post-processing routines to correct imperfections and ensure the models were suitable for laser engraving.
  • Introduced a human-in-the-loop step, where 3D artists reviewed or refined around 10% of the most complex cases.

The prototype proved that large parts of the production process could be automated without sacrificing quality, setting the stage for scaling the solution.

Image

Stage 3: Refining the Pipeline for Real-World Use

Once the prototype showed automation was possible, the next challenge was ensuring the system could handle the variety of real-world photos customers submitted. That meant improving both accuracy and visual quality for laser engraving.

Implementation highlights:

  • Enhanced post-processing of 3D outputs to smooth imperfections and deliver consistent engraving-ready models.
  • Devised a solution for fine details such as cat whiskers, which could not be reliably reconstructed from flat photos. By generating artificial whiskers and combining them with a custom engraving technique, we achieved natural-looking results.
  • Optimized the human-in-the-loop step so that complex or low-quality images could still be refined by artists without slowing down the overall pipeline.

This refinement phase ensured that the system produced consistent, high-quality 3D models across a wide range of inputs, making it reliable enough for everyday production.

Image

Stage 4: Scaling with Cloud Infrastructure

After validating accuracy and refining quality, the next step was to ensure the pipeline could handle production-scale demand. Orders came in from multiple sales channels – Shopify, Amazon, and Etsy – so the system needed to be both scalable and cost-efficient.

Infrastructure highlights:

  • Built a cloud-based architecture on AWS using EC2, SQS, S3, Lambda, EventBridge, and CloudWatch.
  • Implemented a task management queue to distribute workloads efficiently and keep the pipeline responsive.
  • Optimized for cost by selecting the most suitable EC2 instances (T3, G4dn, G4ad) for each task type.
  • Automated infrastructure management with Python and Boto3, enabling seamless scaling up or down based on order volume.
  • Integrated a 3D Viewer UI for operators, ensuring easy validation of outputs when human input was required.

This stage transformed the prototype into a robust, production-ready system capable of handling large order volumes and demand fluctuations without compromising quality or cost efficiency.

Image

Final Outcome: A Scalable AI-Powered Production System

The collaboration resulted in a fully operational automation pipeline that turned customer photos into manufacturing-ready 3D models at scale. By combining advanced neural networks, post-processing, and cloud infrastructure, the solution reshaped how Crystal Clear Memories delivered its products.

Key achievements:

  • 85% automation rate: The majority of orders are now processed without manual modeling.
  • Significant time savings: Model creation that once required several hours by an artist can now be completed in minutes.
  • Support for any input: The system reliably handles both color and black-and-white photos, ensuring flexibility for customers.
  • Faster turnaround: Customers now receive engraved crystals much quicker—critical when products are ordered as gifts or memorials.
  • Consistent premium quality: AI-generated models meet professional standards, with human oversight ensuring flawless results in complex cases.
  • Scalability: A cloud-based architecture and a Unified Order Management System allow the pipeline to manage spikes of up to 10x demand across Shopify, Amazon, and Etsy.
  • Cost efficiency: Optimized AWS infrastructure reduced operational costs while increasing throughput.

The project proved how AI can remove bottlenecks in manufacturing workflows, enabling Crystal Clear Memories to grow their business, improve customer satisfaction, and maintain the craftsmanship their brand is known for.

Image
Image

Let’s Talk About Building AI That Works in the Real World

No vague promises. No pressure. Just a real conversation with our technical leadership to understand your needs and explore the best next steps.

    Tell us about your project
    File upload (optional)

    By submitting this form, you agree with our Privacy Policy

    More Projects Like This

    Kids Speech Therapy Case

    Kids Speech Therapy Case

    A custom speech recognition system for children’s therapy, built with phoneme-level modeling, fine-tuned datasets, and production-ready deployment.

    Pardigm

    Pardigm

    Hormone tracking application powered by Computer Vision with 93%+ accuracy in real-life conditions.