[10:33 Sat,27.September 2025 by Thomas Richter] |
Runway&s collaboration with Hollywood film studio Lionsgate was (grandly announced about a year ago) as the first cooperation between an established film production company (Lionsgate is one of the largest global entertainment groups, which not only produces but also distributes films and TV series) with a video AI provider. Runway wanted to train its own AI model based on Lionsgate&s comprehensive film catalog, which includes 20,000 film and television titles (such as "John Wick", "Twilight", "Hunger Games", "Rambo"). With its help, filmmakers, directors and other creatives at Lionsgate Studios were to be supported in their work and costs saved. The new AI model was to generate videos in the right look, which would then be further processed with Runway&s editing tools.
With effusive words, Lionsgate Vice Chair Michael Burns announced the project: "Several of our filmmakers are already excited about the potential applications in their pre- and post-production processes. We see AI as a great tool to expand, improve and complement our current workflows." ;) Runway und Lionsgate But since then, things have been quite quiet around the project. PetaPixel is now reporting on the difficulties of the project based on an exclusive article, unfortunately behind a paywall, from the online trade magazine for the entertainment and media industry "The Wrap". Thus, the time since the announcement has been relatively unproductive, for the simple reason that Lionsgate&s film catalog is not sufficient to train its own video model - not even Disney&s much larger archive would have been enough, according to an insider. This means that the basis for an "own" video AI is missing, which - similar to Adobe&s Firefly - would not have any potential problems with legally questionable origin of the training material only when using exclusively own material for training - a basic requirement for the commercial use that Lionsgate has planned. ;) John Wick The training data hunger of AIs is notorious - they need enormous amounts of text, image and video data (ideally chained) in order to be able to derive multimodal patterns from them, i.e. to learn to recognize and generate objects including their properties and names. It is speculated, for example, that OpenAI used large amounts of copyrighted video content (e.g. from YouTube, TikTok, Twitch or other platforms) to train its video AI Sora, and Google does not deny that it trained its AI using the enormous mass of videos from its video platform YouTube - which is probably one reason why Google&s Veo 3 model is currently the best. Current lawsuits (e.g. Strike 3) also accuse Meta of downloading pornographic videos and other pirated video files via BitTorrent in order to use them as visual training data. Another reason is also of a legal nature: What rights do actors have if their image appears in an AI-generated clip - and how are they compensated? The article does not say how things will proceed from here. It is conceivable that Runway&s use will initially be limited to pre-production tasks such as storyboarding or pre-visualization, and that the generation of films by AI will be postponed until the current problems are solved.  more infos at bei petapixel.comdeutsche Version dieser Seite: Lionsgate und Runway - die KI-Revolution in Hollywood stockt |
 |
|
|