In this work, we explore the possibility of using synthetically generated data for video-based gesture recognition with large pre-trained models. We consider whether these models have sufficiently robust and expressive representation spaces to enable "training-free" classification. Specifically, we utilize various state-of-the-art video encoders to extract features for use in k-nearest neighbors classification, where the training data points are derived from synthetic videos only. We compare these results with another training-free approach -- zero-shot classification using text descriptions oResearch goal: How does the few-shot learning accuracy of large pre-trained models for gesture recognition compare when using CausalMixFT-generated samples versus other synthetic video generation methods, measured by k-nearest neighbors classification accuracy?Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 7.9/10.