Tribeca's AI experiment failed — and why custom models are Hollywood's only hope
Tribeca film festival highlights a divide in ai filmmaking: while vanilla model outputs resulted in slop, google and openai's custom trained projects showed technical promise for coherent storytelling.
the tribeca film festival showcased the current state of ai filmmaking, revealing that standard generative models often produce disjointed, unpolished results described as video slop. in contrast, collaborative projects like google deepmind's 'dear upstairs neighbors' succeeded by using custom builds of veo and imagen models trained on specific human-made concept art, ensuring a consistent aesthetic and narrative vision. this approach highlights a significant industry shift: hollywood adoption will likely rely on fine-tuning specialized models rather than using off-the-shelf tools, though projects like 'roar' proved that without this technical rigor, ai films remain visually incoherent.