{
  "title": "Keep alignment of camera inputs with hand inputs",
  "caption": "Author-redrawn Generated Reality control diagram. The visual branch uses channel concatenation of video and skeleton latents. Camera geometry and 3D hand articulation use different encoders. The generator adds their embeddings before chunk generation. New tracking supplies input at the next chunk boundary. This diagram does not show autoregressive-history internals, decoder detail or original result frames. Geometric control alone is not evidence of correct contact or object state.",
  "alt": "Generated Reality flow: the model uses channel concatenation of video and skeleton latents before patchification, then adds camera and hand embeddings.",
  "sources": [
    "https://arxiv.org/abs/2602.18422",
    "https://codeysun.github.io/generated-reality/"
  ],
  "limitations": [
    "The playback is an example. Animation time is not measured runtime.",
    "Chunk names are examples. They are not measured timestamps or tracking-to-display latency.",
    "The three paths show dependencies. They make no claim of encoder execution at the same time."
  ]
}