Want to know more? If you are in London in May, I'll be speaking as part King's AI festival to launch the book. Register here 👉🏽 https://kcl.ac.uk/events/cinema-and-machine-vision #AI #creativeAI #GenAI #books #London

There are new forms of pleasure, truth and power, that come with this large-scale reorganisation of visual culture. A longer and broader view allows us to trace a technical and cultural lineage that runs from photography to cinema to machine vision, all the way to #GenAI and text-to-video

It is quite telling, for example, that the clips from this TED video, created using Open AI's text-to-video model, had to be edited together professionally to create a meaningful montage 👉🏽 https://twitter.com/TEDTalks/status/1781351036877156452
Models such as this have no strong internal representations of "shots" or "cuts" (yet). They unroll time and motion instead of recording it

It is hard to know. Partly by design. Partly because we don't have the right tools of knowledge to answer these questions yet. But we do know that AI is very rapidly becoming media, an we now have a very good idea of how 👇🏽

These systems are trained to watch on our behalf and reproduce what they see guided by text prompts. But whose vision is encoded in them and how? 👀 Is this a new type of cinematic engine? or a way for tech companies to extract value from filmmakers and creators? or both?

Moving images have gone from celluloid🎞️ to digital🖥️, from recordings 🎥 to data streams 💽 to calculations 🖩. This has made online video platforms and film archives a very valuable source of training material for AI models ― as long as computers can taught how to watch!