The Shoot

We hosted a media day photo shoot for a local youth baseball organization that needed large-format team banners for game days, along with individual portraits of every player. The session took place over several hours at our studio in Chesapeake, Virginia, where we built a controlled environment using a custom RGB lighting setup and rolls of artificial turf to recreate the atmosphere of the field.

The shoot produced nearly 1,500 high-resolution photographs that required post-production. Under a traditional editing workflow, processing this volume of images would have required a significant amount of manual work. Instead, we used the project as an opportunity to develop and validate an AI-assisted editing pipeline capable of accelerating production while maintaining consistent, professional-quality results.

The Pipeline

Building a dependable workflow required extensive testing and iteration. We chose Claude Code to orchestrate the automation process and DALL·E for AI image generation. A custom Python application was developed to bridge the two systems, allowing them to communicate and exchange assets automatically.

As production progressed, the pipeline evolved continuously. New validation rules, processing logic, and prompt refinements were added throughout development to improve consistency, eliminate edge cases, and reduce manual intervention. Every iteration made the system more reliable while increasing the quality of the final output.

Input vs Output

The pipeline was expanded to include facial recognition, allowing photographs to be automatically grouped by player. Images were sorted into individual folders before entering a quality control stage.

Each photograph was reviewed against a series of technical criteria, including focus accuracy, sharpness, and exposure. Images that failed to meet our standards were excluded, while the strongest captures advanced through the editing pipeline. This automated curation process ensured that AI processing was performed only on high-quality source material.

Subject Isolation

The selected photographs were processed using automated subject segmentation to isolate each player from the background. Because the studio lighting had been designed around a solid black backdrop, the extraction process produced exceptionally clean masks with minimal manual cleanup.

Isolating the subject created a flexible compositing workflow, allowing each player to be placed into custom AI-generated environments while preserving fine details such as uniforms, gloves, bats, helmets, and hair. This non-destructive approach provided complete creative flexibility throughout post-production.

Prompt Engineering

Generating consistent backgrounds required far more than a single prompt. We developed a structured prompt framework that controlled composition, perspective, lighting direction, atmosphere, and subject placement.

Early iterations occasionally introduced inconsistencies, such as modifying clothing colors or producing environments with incorrect scale or perspective. These issues were addressed through continuous prompt refinement and additional processing rules until the generated scenes consistently matched the original photography.

Once the core workflow had stabilized, controlled variation was introduced to produce unique backgrounds while maintaining a consistent visual style across the entire project. The result was a collection of images that shared a unified aesthetic without appearing repetitive.

Training the AI

Developing the visual style required extensive research and experimentation. A curated collection of reference images from Pinterest was provided to Claude along with detailed explanations describing the lighting, composition, atmosphere, and visual characteristics that defined our desired aesthetic.

Those references informed the prompts used throughout the pipeline, allowing realistic composites to be generated with a consistent cinematic style. When necessary, the subject was re-lit during compositing to better match the lighting direction and mood of the generated environment.

Additional rules were introduced to preserve visual consistency throughout the project. For example, smoke effects appearing on the left side of an image were always generated in white, while smoke on the right remained purple to complement the RGB lighting captured during the original photo shoot. High-quality outputs were continuously fed back into the workflow as reference material, creating an iterative process that steadily improved the overall quality of the generated images. Image uploads and processing were fully automated.

Final delivery

The project extended beyond image editing into the development of a complete client delivery platform. We built a custom WordPress plugin that automatically organized completed photographs into secure, player-specific galleries.

Each gallery included watermark protection, individual image downloads, full-gallery ZIP downloads, and secure sharing options for family members. By developing both the AI editing pipeline and the supporting delivery infrastructure, we created a scalable system that can be reused for future media day productions.

Becoming AI-native

Artificial intelligence is most valuable when it becomes part of an organization's everyday workflow rather than an isolated tool.

Made In The Dark™ helps forward-thinking companies integrate AI into creative and operational processes through custom automation, workflow optimization, and intelligent software solutions. Whether streamlining repetitive production tasks, accelerating content creation, or building entirely new AI-powered workflows, we help organizations reduce manual effort, improve efficiency, and scale with confidence.

Partner with us to discover how practical AI integration can save time, reduce costs, and transform the way your business operates.