Comparing neuronal activity across individuals poses experimental challenges, despite stereotypical behaviors within the same species. Functional MRI allows coarse-grained brain activity comparison but lacks the resolution for understanding the emergence of whole-brain dynamics. With Zebrafish larvae, we can perform whole-brain single-neuron imaging, yet current cross-individual comparison methods are limited to sensory-driven paradigms. Thus, analyzing spontaneous brain activity across zebrafish larvae remains challenging. Presenting a novel approach, we employ a single Restricted Boltzmann Machine (RBM) to model neuronal statistics from multiple zebrafish larvae, overcoming challenges in comparing spontaneous, neuron-level, whole-brain activity. Our method extends previous work using RBMs to model whole-brain neuronal statistics in single zebrafish larvae. In this novel approach, a single RBM captures neuronal activity of multiple larvae, projecting it to a shared latent space. We introduce two methods: one using voxelized brain activity, revealing conserved pairwise correlations between hidden units despite variable voxel-voxel correlations across individuals; the other, at the single-neuron scale, involves training an RBM on one fish and adapting it to others, enabling a neuron-to-cell-assemblies scale transition common to multiple individuals. Adding a classification layer allows assigning brain states to each recording time point. Markovian transition probabilities between states are partially conserved between individuals, indicating our method's ability to capture stereotypical zebrafish brain dynamics. Our approach combines spontaneous neuronal activity from multiple zebrafish larvae into a shared and interpretable latent space, paving the way for a generic model of zebrafish neuronal dynamics and shortening RBM training time for practical online experiments.