Spiritual Animal
Category
Visual system / Interaction / Implementation
Tools
TouchDesigner (TD), Coding, MediaPipe (hand/face tracking)
/
/
Owls hold symbolic significance across many cultures: in shamanic traditions and on ancient Chinese ritual artefacts, they are often regarded as messengers linking the human world with the spiritual realm. By continuously generating images of owls and incorporating hand and facial interactions, ‘Spiritual Animal’ allows viewers to feel as though they are participating in a digital ritual, rather than merely interacting with a screen.
The Problems
During the development process, the challenges I encountered were not merely about generating the owl, but rather about simultaneously addressing generation quality, real-time stability, and interaction alignment.
First, training the model with over 200 images initially resulted in blurry outputs that resembled X-rays or unidentifiable clusters.
Second, the real-time connection between the local model and TouchDesigner was unstable due to server and network issues.
Third, the visual overlay of gesture and facial tracking suffered from issues such as incorrect scaling, central freezing, facial misalignment, and layer occlusion, all of which affected the final viewing experience.
How I Solved Them
I began by shifting the focus of iteration from quantity to dataset curation. I removed unsuitable images, manually selected a small set of consistent visuals, and removed their backgrounds. After retraining, the generated lines became noticeably clearer and more controllable.
Next, I repeatedly fine-tuned the parameters and mappings for the interactive system, such as adjusting the screen position and aspect ratio when triggered by gestures. I continuously tested the alignment between facial and hand tracking layers, gradually resolving issues with central freezing, occlusion, and misalignment.
Finally, I made necessary trade-offs in the closing stage. When the overlay-and-hold effect proved difficult to maintain consistently, I switched to a more reliable structure based on trails, scene transitions, and drumbeat sound effects, using rhythmic repetitions and layering to enhance the sense of a digital ritual.
Future Development
For future development, I plan to refine the generated owl imagery so the hand-held owl line drawing becomes sharper and more readable, with cleaner edges and higher contrast. I also want to remove or better control the background, using segmentation/matting and prompt constraints so the owl remains the clear focal point rather than blending into noise. Beyond visual clarity, I hope to implement ideas I couldn’t finish during production: more stable real-time tracking, smoother gesture-to-image mapping, and a richer “ritual” feedback loop (sound, lighting, and adaptive pacing) so interaction feels intentional, immersive, and emotionally responsive.
