{
  "video": "video-f2b0a3bf.mp4",
  "description": "This video is a presentation or demonstration of a research paper titled **\"TokenDial: Continuous Attribute Control in Text-to-Video via Spatiotemporal Token Offsets.\"**\n\nThe core of the video appears to be showcasing the capabilities of a system called TokenDial. This system allows users to control and manipulate visual attributes within generated videos (Text-to-Video generation) in a continuous and spatially controlled manner by providing text prompts.\n\nHere is a detailed breakdown of what is happening:\n\n**1. Title and Authorship (Introduction):**\n* The video opens with the title slide, identifying the research: \"TokenDial: Continuous Attribute Control in Text-to-Video via Spatiotemporal Token Offsets.\"\n* It lists the authors: Zhixuan Liu, Peter Schaldenbrand, Yijun Li, Long Mai, and Aniruddha Mahapatra, affiliated with Adobe Research and Carnegie Mellon University.\n* Options are provided to download the paper (PDF, arXiv, Code, BibTeX).\n\n**2. Demonstration Slides (Visual Examples):**\nThe main body of the video consists of a series of demonstration slides, each illustrating a specific attribute manipulation capability of TokenDial. Each slide features:\n* **An input image/concept** (or a description/prompt).\n* **A visual output** demonstrating the change.\n* **A slider bar labeled \"Strength,\"** which suggests that the user can control *how strongly* the requested attribute is applied to the video generation.\n\n**Observed Examples:**\n\n* **Character/Object Modification:**\n    * \"Make the cat more kitten-like\": Shows a transformation of a cat's appearance.\n    * \"Make the person older\": Shows an aging effect applied to a person in the video.\n    * \"Make the dog furrier\": Shows a change in the texture/coat of a dog.\n* **Environmental/Scene Modification:**\n    * \"Make the fire larger\": Demonstrates scaling or intensity control over a fire.\n    * \"Make the aurora brighter\": Shows control over the luminescence of an aurora borealis.\n    * \"Make the snowflake larger and denser\": Demonstrates manipulating the size and density of snowfall.\n\n**Conclusion:**\nIn summary, the video functions as a highly visual abstract or demonstration of a state-of-the-art generative AI model (TokenDial) that excels at **fine-grained, controllable editing of video content** based on text prompts, allowing for smooth, continuous adjustments of visual properties like size, texture, age, and intensity.",
  "codec": "av1",
  "transcoded": true,
  "elapsed_s": 12.8
}