Rendered via text-to-image in stable diffusion.
This was rather complicated to produce. The basic steps:
- Crop and tag art from the Wetware series
- Train hypernetwork
- Render frames using hypernetwork and multiple control nets, sourced from a video if Victoria Macreatis
- Basic temporal smoothing filter in VirtualDub to reduce flicker
- Use segmentation map to mask out the subject
- Multiple passes of RIFE to smooth out flickering (render in-between frames, delete original frames, re-render in-between frames, repeat)
- Filter the background heavily, since it was flickering nonsense
- Recomposite
