I’ve got the basics of modeling and importing anime-styled characters, so now it’s time to give them some environments in which to operate. After a bit of searching, I came up with two apparently fast options:
- This amazing piece of work, showcasing Blender’s image management abilities to effectively turn a real-life HDRI into an anime-esque environment
- AnimeGAN v2, a machine learning project that can render real footage in a variety of directors’ styles
Both approaches have their advantages and disadvantages. The HDRI adaption gives me a full 360-degree environment, but I have to make a lot of manual changes every time I plug in a different HDRI. Plus, free HDRI images are in much shorter supply than free-to-use footage.
AnimeGAN feels more promising in terms of source material, but again its limits are significant. It works on 2D footage, rather than generating 3D environments, limiting how I’m able to stage characters. It also requires a complex (for casual use) setup, running on a Ubuntu VM attached to a GPU. I can set this up, but it’s not a system I’d encourage others to try. It also limits commercial uses of its output.
The solution that addresses the issues with both solutions is actually to model environments and style them appropriately. This is time-consuming compared to just feeding an image into a piece of software and getting the goods after a few minutes, of course.
Resources I’d recommend for environmental modelers:
- Kristof Dedene has created several tutorials for doing Ghibli-style terrain
- Blender NPR is a resource for non-photo-realistic rendering techniques
- A free Manga Shader for Eevee
Image source: https://www.ghibli.jp/works/marnie/#frame
Clocktropolis 