We demonstrate action conditioning coherence of GAIA. Here, we start from a fixed video context and predict the future four times with varying action conditioning. In the top left, we demonstrate a slight left turn. In the bottom left, we demonstrate a strong left turn. In the top right, we demonstate a slight right turn. In the bottom right, we demonstrate a strong right turn. We note that in all cases, GAIA perfectly adheres to the requested action conditioning and thus shows the model has learned vehicle dynamics well. Read the blog here: https://wayve.ai/thinking/scaling-gai...