Revisiting Scene One
After six weeks of experimentation with AI tools and video generation workflows, I revisited the first scene of the film. Looking back at the original version, I realised there was still significant room for improvement, particularly in terms of cinematic storytelling, scene transitions, and the gradual reveal of the environment.
To differentiate it from Scene Two, I redesigned the opening sequence. Instead of using a moving object as a transition, I chose a top-to-bottom camera tilt. The shot begins with a sandstorm filling the frame, and as the dust slowly clears, the post-apocalyptic landscape gradually appears. This creates a stronger sense of mystery and encourages the audience to discover the world alongside the characters.
I then used a low-angle tracking shot following the girl and the honey badger as they walk forward. Rather than revealing the entire environment immediately, the scale of the world unfolds slowly through the characters’ journey, creating a more cinematic sense of exploration.
PROMPT EXAMPLE1:
Character 1 is the taller character. Character 2 is significantly smaller, approximately two-thirds the height of Character 1. Maintain this size relationship throughout the entire sequence.
A cinematic adventure travel shot in a vast post-apocalyptic ruin.
The sequence begins with the camera positioned high above the ancient ruins. The camera slowly tilts downward, revealing a long stone staircase descending into the scene. At the base of the frame, Character 1 and Character 2 appear and begin walking side by side up the staircase toward a mysterious structure hidden deeper within the ruins.
As the camera completes the downward tilt, it smoothly transitions into a low-angle tracking shot from behind the characters. The camera follows closely as they continue climbing the long staircase and moving deeper into the ancient environment.
A strong desert wind suddenly sweeps through the ruins. Swirling dust and sand roll across the staircase and into the foreground. The characters continue walking forward against the wind. Their coats, scarves, straps, loose fabric, and accessories flutter naturally in the airflow. Fine sand particles drift through the sunlight.
The gust becomes stronger and briefly fills the frame with blowing dust, partially obscuring the lens. As the dust begins to clear, an abandoned ancient temple gradually emerges ahead, revealed through the haze and drifting sand.
The two characters continue walking up the staircase toward the temple. The camera remains low behind them, emphasizing the scale of the journey and the grandeur of the forgotten structure ahead. The movement should feel like the beginning of a long expedition into an unknown world.
Wide-angle lens, smooth cinematic camera movement, tilt-down opening shot, low-angle follow shot, immersive environmental storytelling, strong sense of scale, exploration, discovery, companionship, adventure journey atmosphere.
Low-saturation post-apocalyptic color palette. Muted earth tones, dusty beige, faded ochre, weathered stone, atmospheric haze.
Hand-painted oil painting aesthetic, visible brush strokes, painterly textures, organic brushwork, traditional concept-art rendering, textured paint layers, handcrafted cinematic illustration quality. Strong wind effects, drifting dust, moving fabric, highly immersive atmosphere.
Exploring Prompt Design
While rebuilding the sequence, I found myself thinking less about the final image and more about the process behind it. One question repeatedly emerged: how many reference images should I provide to AI, and at what point do additional references become unhelpful?
Through testing, I noticed that too few references often produced generic results, while too many references sometimes confused the AI, leading to inconsistent compositions and visual styles. Similarly, highly detailed prompts could improve certain aspects of the image but occasionally caused the AI to ignore more important cinematic elements.
Reflection
This week’s work shifted my focus from image generation to communication. I began to realise that AI generation is not simply about describing what I want to see. Instead, it is about deciding which information is essential and how that information should be prioritised.
This raises an interesting question: is the creative challenge in AI filmmaking creating images, or designing instructions? At this stage, I believe the quality of the result depends less on the AI itself and more on the creator’s ability to organise references, define priorities, and communicate visual intentions clearly. In this sense, prompting has become a creative skill of its own.