{
  "video": "video-0dd3ac51.mp4",
  "description": "This video, titled **\"TokenDial: Continuous Attribute Control in Text-to-Video via Spatiotemporal Token Offsets,\"** is a demonstration of a generative AI model that allows users to control the visual attributes of video generation using text prompts and continuous sliders.\n\nThe video showcases the model's capability to modify specific aspects of an animated scene dynamically across time.\n\nHere is a detailed breakdown of what is happening:\n\n**Overall Concept:**\nThe video presents three distinct visual examples, each demonstrating control over a different attribute (fire intensity, aurora brightness, and snowflake density/size) using a continuous \"Strength\" slider.\n\n**Demonstration 1: Fire Control**\n*   **Scene:** The initial frame shows a vibrant, burning fire.\n*   **Control:** A slider labeled **\"Strength\"** is present below the image.\n*   **Action:** As the slider is manipulated (implied to be moving from left to right across the timeline), the **intensity and size of the fire** in the animation change continuously. The animation transitions from a less intense or smaller fire to a more intense, larger blaze, demonstrating fine-grained control over a physical attribute.\n\n**Demonstration 2: Aurora Control**\n*   **Scene:** The scene features a colorful display of the Aurora Borealis (Northern Lights) over a darker background.\n*   **Control:** A slider labeled **\"Strength\"** is present.\n*   **Action:** Moving the slider controls the **brightness of the aurora**. The animation shows a transition where the auroral glow becomes gradually brighter or dimmer, allowing the user to dial in the desired atmospheric intensity.\n\n**Demonstration 3: Snowfall Control**\n*   **Scene:** The scene depicts a night or twilight setting with snowfall.\n*   **Control:** A slider labeled **\"Strength\"** is present.\n*   **Action:** Manipulating this slider controls the **size and density of the snowflakes**. The animation illustrates a change in the precipitation, where the snow can appear as smaller, sparser flakes to larger, denser clusters, providing control over the texture and visual weight of the falling snow.\n\n**In Summary:**\nThe video serves as a compelling visual proof-of-concept for **TokenDial**, showcasing its ability to exert **continuous, spatiotemporal control** over visual elements within generated video clips based on numerical or continuous inputs (sliders), going beyond simple on/off text prompts.",
  "codec": "av1",
  "transcoded": true,
  "elapsed_s": 10.5
}