" Abandoned, but somehow beautiful. "
Background
Text-to-image synthesis are neural networks that create images from text captions. They are capable of producing a wide range of objects and scenery and are also able to visualize abstract and intangible scenes. With the right text prompts, these neural networks can combine unrelated concepts in convincing and plausible ways, somewhat similar to human artists.
Goal
Generate AI-imagined scenes of abandoned, but somehow beautiful spaces. Use a soundtrack and VFX to augment the scenes to give a more moody vibe, natural lighting and atmosphere.
Project details
Independent video project and accompanying soundtrack.
Skills
  • Motion design
  • VFX
  • Machine learning
  • Music production
  • Musical composition
Project takeaway

This project was a good way for me to improve upon my video post production skills. The text-to-image technology that I used gives the video a distinct visual aesthetic and matches the curious and moody vibe that I was aiming for. After finishing the video, I composed a soundtrack in Ableton Live using string and percussion samples.

This is the first project where I used Adobe After Effects to add VFX to a video. By adding volumetric lighting and dust particles I was able to breathe life into an otherwise flat, 2D scene. I also used 3D photo in-painting to create depth maps of each scene to make the camera zoom-ins more realistic.

A computer generated image of a dark cellar with morning light rays being cast on the walls. The room looks dusty, tattered and abandoned.
abandoned basement