NanoFingerprints have been recently presented as a new representation of chemical nanocompounds designed for toxicity prediction. Only one algorithm to compute NanoFingerprints have been reported, which needs the original 3D structure of the nanocompound in XYZ format and it has a computational cost to the power of four with respect to the number of atoms. This paper presents a generative algorithm to compute NanoFingerprints, which does not need the exact nanocompound structure and it has a linear computational cost with respect to the number of atoms. Our practical experiments report a low error while generating NanoFingerprints and thus a huge capacity of properly representing the initial nanocompound. This is a crucial advance since the toxicity of larger nanocompounds can be estimated without the need of generating the huge representation of its 3D structure.