Reflection #2

As I begin working on the Production Automation Workflow (PAW) project with Boie and Sean, I have been thinking about what it means to incorporate generative AI into the production of educational media. The project explores how generative AI and workflow automation can support different stages of media production, from script writing and metadata generation to creating and assembling media assets. Our current focus is narrower (finally!): learning to work with ComfyUI, exploring LoRA training, and eventually connecting the outputs to n8n.

This project really places me in the position of both a learner and a potential user of AI-powered creative tools. I have experience with photography, video-making, and visual storytelling, but working with generative AI introduces a different set of questions. Rather than directly producing every visual element myself, I need to understand how models, workflows, prompts, and technical configurations shape the final output. This process is much more complicated than I thought, since there is so much both useful and useless information online, and each of them is so accessible, which makes the process even harder.

• Learning Through Technical Obstacles

My initial exploration of ComfyUI has made me realize that using AI tools is not necessarily as simple as entering a prompt and receiving an image. In fact, we spent over half a month getting through all the technical difficulties and software installations. Setting up models, understanding nodes, managing computational resources, and troubleshooting errors are all part of the process. These technical requirements influence what I can create and how I approach experimentation.

For me, this project is an opportunity to develop a more practical understanding of AI systems rather than treating them as black boxes. I want to understand not only how to make a workflow function, but also why it functions in a particular way and what decisions remain in the hands of the person using it.

• Automation Does Not Eliminate Human Decisions

So far, one aspect of PAW that I find particularly interesting is Sean's emphasis on keeping a human in the loop. Automation can make a workflow more efficient, but efficiency alone does not necessarily guarantee a meaningful or effective educational media product.

For example, generating a series of images with a consistent visual style involves more than obtaining technically successful outputs. Someone still needs to consider whether the images communicate the intended ideas, whether they fit the larger narrative, and whether they serve the project's educational purpose. Similarly, connecting ComfyUI to n8n may streamline the movement of assets between stages, but it does not automatically resolve questions about visual quality, interpretation, or design.

I am beginning to see workflow automation not simply as a way to replace repetitive tasks, but as a way to reorganize the relationship between technical execution and human judgment. A successful workflow should make certain processes easier while preserving opportunities for people to evaluate, revise, and make decisions.

As the project develops, I want to explore several questions:

• How can we design AI-assisted workflows that support creative experimentation without making users overly dependent on automated decisions?

• What knowledge should users develop to understand and adapt an AI workflow rather than simply operate it?

• How should we evaluate the success of an AI-assisted educational media workflow beyond its technical functionality?

As we move toward LoRA training and integration with n8n, I hope to develop a more grounded understanding of how AI tools can become part of a thoughtful, human-centered creative process.