Scene Production
This week, we started producing individual shots for the film. Since our story follows the post-apocalyptic journey of a young girl and a honey badger, many different environments are needed to create a sense of exploration.
I was responsible for the environmental shots. As I had little experience using AI in production, I first planned a simple workflow: finding suitable reference images, using AI to transform them into our chosen style, placing the characters into the scene, and creating storyboard frames.
Finding reference images was easy, but I quickly ran into problems during the style transformation stage.
Challenges with AI
Although the AI followed my prompts, the generated images did not match the visual style we had established. Instead of the painterly post-apocalyptic anime realism style we wanted, many images looked generic and lacked the atmosphere and texture I was aiming for.
This made me realise that AI could recognise keywords, but it did not necessarily understand the artistic intention behind them.
Finding a Solution
To solve this issue, I changed my approach. I selected an image that closely matched our target style and asked AI to analyse it and generate a detailed prompt.
I then reused the extracted keywords together with my own reference images. The results improved significantly and were much closer to the visual direction of the project.
Reflection
This experience changed my understanding of AI-assisted creation. I initially thought AI could understand style in a similar way to human artists. However, I found that successful results depended on learning how to communicate with AI using a language it could interpret effectively.
This also raised an important question: if strong results rely on carefully designed prompts and human selection, where does creativity come from?
At this stage, I believe AI is more of a creative assistant than an independent creator. Human judgement is still essential in guiding, selecting, and refining the final outcome.