1. OpenAI’s video generation tool Sora surprised the AI community with realistic video, but lacks full control over elements like color and timing.
2. Shy Kids, a Toronto-based digital production team, created short films using Sora for OpenAI promotional purposes, emphasizing the limitations and laborious process involved.
3. Sora has restrictions on generating copyrighted content and lacks precise control over movements and timing, making it useful but not capable of creating films independently.
OpenAI’s video generation tool Sora impressed the AI community with its realistic videos, but details were missing from its debut. Filmmakers from Shy Kids were given early access to Sora to create a short film, revealing the extensive work involved in the production process, including storyboarding, editing, and color correction.
Despite the impressive capabilities of Sora, it still lacks full control over consistency which requires elaborate workarounds for simple tasks like choosing the color of a character’s clothing. Additionally, elements like unwanted faces on balloons or strings required manual removal in post-production.
Precise timing and movements are also a challenge with Sora, making it difficult to achieve specific actions like gestures or camera movements accurately. Filmmakers had to render shots composed in portrait orientation and perform post-production adjustments to achieve their desired results.
Sora’s limitations also extend to copyright recognition, as it refused to generate clips resembling copyrighted content like “Star Wars” or specific directorial styles like “Hitchcock zoom.” The model’s training data and capabilities remain a mystery, with OpenAI maintaining secrecy around the information.
Overall, Sora is a powerful tool for certain applications in filmmaking, but its current limitations prevent it from fully creating films independently. As the technology evolves, it may eventually achieve greater control and precision, improving its utility in the industry.