AI-Enhanced Media Generation: How Photozig Scaled Image and Video Workflows with Avahi and AWS

Project Overview

Photozig, a digital tech company focused on health and media applications, aimed to streamline and scale their AI-driven media generation processes. They partnered with Avahi to build a production-ready system on AWS that enhanced user prompts using advanced LLMs, integrated cutting-edge diffusion models for image generation, and enabled short-form video creation. This 8-week project successfully delivered a scalable, APIdriven platform backed by AWS-native tools, improving content quality and delivery speed while reducing operational complexity

About the Customer

Photozig, Inc. is a California-based technology firm specializing in digital media solutions for health, wellness, and creative industries. The company focuses on leveraging artificial intelligence to enhance image and video generation for end-users and enterprise customers alike.

The Problem

Photozig needed a modern, efficient pipeline to process user-generated prompts and turn them into highquality images and videos. Their existing systems lacked prompt optimization, and media generation relied on outdated or isolated model workflows. Without addressing these gaps, Photozig risked slow turnaround times, inconsistent outputs, and lost market competitiveness in AI-powered creative tools. Furthermore, the inability to handle large volumes of content or support modern video features limited their ability to scale production.

Why AWS

AWS provided the secure, scalable, and GPU-optimized infrastructure required for generative AI workloads. Services such as Amazon Bedrock enabled prompt enhancement via powerful LLMs, while Amazon EC2 with NVIDIA GPUs supported high-performance model inference. AWS Lambda, S3, CloudWatch, and SageMaker simplified orchestration, storage, monitoring, and deployment across the platform.

Why Photozig Chose Avahi

As a Premier-tier AWS Partner, Avahi brought deep experience in building AI-centric cloud solutions, including hands-on knowledge of model evaluation, workflow optimization, and serverless architecture. Avahi’s consultative approach ensured rapid execution through weekly milestones, flexible integration with existing Photozig assets, and tailored use of state-of-the-art models like Nova Canvas and AnimatedDiff. The team’s focus on production-readiness and AWS-native implementation was key to delivering a robust system within 8 weeks.

Solution

Avahi implemented a multi-component solution across image and video workflows:

• Data ETL: Ingested, cleaned, and stored existing user prompts using custom pipelines into Amazon S3.
• Prompt Enhancement: Built a production-grade API powered by Amazon Bedrock’s Nova Pro model to refine user inputs for higher-quality generation outcomes.
• Image Generation: Integrated diffusion models like Nova Canvas, SD3.5, and Flux with background removal and post-processing, optimizing for multiple resolutions.
• Video Generation: Deployed short-video pipelines using LTX, Wan 2.1, and AnimatedDiff with advanced features like zoom, pan, face swap, and product swap.
• Infrastructure: Provisioned GPU-accelerated EC2 instances (g6e.8xlarge with NVIDIA L40S), API Gateway, Lambda, and monitoring via CloudWatch for performance and reliability.
• ComfyUI Platform: Integrated a node-based workflow engine to allow intuitive management and batch processing of AI pipelines.

Key Deliverables

• Production-ready prompt enhancement API
• ETL pipelines for user prompts
• Integrated diffusion models for image generation
• Short video generation pipelines
• AWS deployment scripts and automation
• Documentation and knowledge transfer sessions
• Monitoring via AWS CloudWatch

Project Impact

The deployment enabled Photozig to operate a high-quality, scalable AI content generation platform with seamless integration into their existing systems. The customer now supports advanced use cases across image and video with faster output times, streamlined workflows, and multi-resolution compatibility.

Key Metrics

• 100% AWS-native deployment with no third-party SaaS dependency
• Support for 3+ image formats and resolutions (1024×1024, 864×1536, 1536×864)
• GPU optimization via NVIDIA L40S (46GB VRAM) on g6e.8xlarge instances
• 6+ AI models integrated including Nova Pro, Flux, LTX, AnimatedDiff, and Hunyuan