AI Video Presents Possibilities, Raises Questions

After Dall·E2, Stable Diffusion, and Mid journey took the Internet by storm by doling out remarkable still images generated by an AI fed with text prompts, many were left wondering just how rapidly this technology would evolve. Questions of artistic ethics, ownership, and the notion of work all fed into the hype surrounding text-to-image generators. Now, motion has come into play courtesy of a handful of powerful AI video generators, posing even more questions about this often-astounding technology.

The Phenomenon of Phenaki

Google’s Phenaki has already made waves with its AI text-to-video creations. Most notably, Phenaki went further than many expected and proved itself capable of producing video running not just seconds but minutes. From near-photorealism to cartoonish styles, Phenaki’s output offered up surprisingly lengthy sequences that adapted to a user’s text prompts.

Variations on “An astronaut riding a horse in the park at sunrise,” generated by Phenaki. [Image: Phenaki]

Not to be outdone, Meta also pushed its generative video creator into the public discussion. Make-A-Video may not come with an especially imaginative name, but its capabilities allow for creative work to be aided by a robust dataset and learning model. Claiming Make-A-Video lets users “bring [their] imagination to life by generating whimsical, one-of-a-kind videos with just a few words or lines of text,” Meta clearly has light-hearted content as its focus. There are, however, more serious issues arising from the development of generative imagery.

Privacy, Identity, and Authorship

Just who is responsible for the imagery outputted by AI generators? With AI art now winning top prizes at art fairs and leading some to question their validity as human artistic creations, and with “deepfakes” still quite fresh in the cultural memory, generative video may have a tough time becoming roundly accepted in all fields of media.

A first-place award to an AI-wielding artist at the Colorado State Fair has created yet another firestorm of debate over the use of image synthesis in art—and whether AI art is art at all. https://t.co/Sb1cWPhTU5

— Ars Technica (@arstechnica) September 3, 2022

What fed the dataset that led to the creation of the imagery? Are celebrities entitled to limit the use of their likeness and exclude themselves from AI-generated content? Is there a case to be made for more ethical production of sexually explicit content starring AI-generated performers? There are no easy answers here, and, it seems, the only way we’ll find out is by watching the tech evolve and seeing where complications (as in the above case of the Colorado State Fair contest) arise.