Рет қаралды 7,744
Discovering liberty within everyday existence
Music video of my own track I made a while back, created with TouchDesigner and StreamDiffusion. I first cut the original video into a basically finished piece and then used that as a source for SD, while adding a bunch of processing including audio reactivity in TD. The final output was deflickered, upscaled and edited with Davinci Resolve and Topaz Video AI.
The challenge was to keep the movement of the dancer visible and clear while still adding the abstraction made by TD and SD, so in the end, I had to composite the original with the AI layer -- it was a lot about balancing AI and original.
I definitely learned a lot, e.g. it's important that the dancer differentiates from the background if you want the AI to recognize it as such in this workflow, or have them be close enough to the camera. Also the outfit was not ideal, as for the AI to see it's a human figure, it appears to be better if there's a clear and consistent silhoutte. The background should already look close to what you wish to see in the end, for me personally the first part works better than the second, because it's filmed in a building which is closer to the city/graffiti style than what I wanted the second part to look like -- natural and organic.
While it was amazing to work with dance and actual video footage, I think I will focus more on purely digital as I'm more in control of the content that way. Let's see, it's such a new and quickly evolving tool and workflow, I'm very curious to explore many different methods.
Huge thanks to everyone involved!! ❤️
Dance: Aya Nakagawa
Camera: Marta Pang
Audio Mastering: Phelan Kane
Music, Visuals & Editing: Bileam Tschepe
More coming soon hopefully, still got many finished tracks waiting to be shared along visuals.