Investigating AI Inputs
This week, I decided to explore a question that had emerged from my previous experiments: what makes an effective AI input? More specifically, how many reference images are needed, what kind of references work best, and how detailed should a prompt be?
To answer these questions, I reviewed online tutorials, research articles, case studies, and discussed my findings with classmates. I also conducted a series of practical tests using different combinations of reference images and prompt structures.
Through these experiments, I found that more references do not necessarily produce better results. Instead, the quality and relevance of the references are more important than the quantity. For example, when generating a shot involving one environment and two characters, three carefully selected images were often enough: one storyboard reference for the environment and one turnaround sheet for each character.
Understanding Prompt Design
I also explored how prompt length and structure affect the output. My findings suggest that, within the model’s limits, detailed prompts generally lead to more controllable results.
The most effective prompts included information about atmosphere, lighting, camera movement, character behaviour, and even the purpose of specific shots at particular moments in the sequence. Rather than simply describing what should appear on screen, I found it useful to explain why a shot exists and what emotion it should communicate.
At the same time, I learned that not every detail needs to be controlled. Leaving some creative space often allowed AI to generate unexpected visual ideas that improved the sequence.
Reflection
This week’s research reinforced a key conclusion of the project: the most valuable part of the creative process is still the human idea behind it.
AI can generate images and videos quickly, but it frequently misunderstands important instructions. One recurring issue was character scale. Despite repeated prompts, the AI would sometimes ignore size relationships or even reverse the proportions between characters.
This highlights a gap between generating images and understanding visual intention. AI is a useful assistant, but it is not a reliable creative decision-maker. The stronger my artistic direction became, the better the results were. Therefore, I now see AI not as a replacement for creativity, but as a tool that depends on human ideas, judgement, and critical thinking.