Shy Kids has constructed two short films for OpenAI where it uses “VFX+Sora” with accompanying breakdowns of the source Sora output, and what they did to make the video clips “work” in their montage narrative. If you’re doing this much work to prop up the output, and it still looks bad, then what the hell is the point? To say nothing of how tone deaf it is to showcase this during a huge VFX industry downturn. Bananas that they’re proud of this.
The black levels don’t match, the lighting doesn’t match, the part where the balloon joins the neck changes shot to shot. Clothing details like his collar change. The balloon itself is even replaced by a human in several shots seemingly because it’s too shiny for continuity. The Sora output has to be stabilized and cut out showing fluctuating edges and highlights because Sora doesn’t have an alpha channel and can’t actually composite anything, just mush images together.
If this was a real pipeline you’d cast a guy for the role who would walk and talk with the reporter - what a concept! Then you’d paint out his head and replace it with a CG balloon asset rendered to match his acting and blended to the skin on his neck the same way every time. This is like making a pipeline where your main cast is stock footage.
@joesteel I was just about to ask this - their end result would be way easier to achieve by starting with two real actors and just replacing the head with a balloon.
Add comment