We current a novel strategy to predict molecular conformers by a easy formulation that sidesteps most of the heuristics of prior works and achieves state-of-the-art outcomes by utilizing some great benefits of scale. By coaching a diffusion generative mannequin instantly on 3D atomic positions with out making assumptions concerning the specific construction of molecules (e.g. modeling torsional angles) we’re in a position to radically simplify construction studying, and make it trivial to scale up the mannequin sizes. This mannequin, referred to as Molecular Conformer Fields (MCF), works by parameterizing conformer buildings as features that map parts from a molecular graph on to their 3D location in house. This formulation permits us to boil down the essence of construction prediction to studying a distribution over features. Experimental outcomes present that scaling up the mannequin capability results in massive beneficial properties in generalization efficiency with out implementing inductive biases like rotational equivariance. MCF represents an advance in extending diffusion fashions to deal with complicated scientific issues in a conceptually easy, scalable and efficient method.