JINST JINST CaloClouds3: Ultra-fast geometry-independent highly-granular calorimeter simulation Buss Thorsten DESY Hamburg U. Day-Hall Henry DESY Gaede Frank DESY Kasieczka Gregor Hamburg U. Krüger Katja DESY Korol Anatolii DESY Madlener Thomas DESY McKeown Peter CERN Mozzanica Martina Hamburg U. Valente Lorenzo Hamburg U. 2026 21 We present CaloClouds3, a model for the fastsimulation of photon showers in the barrel of a high granularitydetector. This iteration demonstrates for the first time how apointcloud model can employ angular conditioning to replicatephotons at all incident angles. Showers produced by this model canbe used across the whole detector barrel, due to specially producedposition agnostic training data. With this flexibility, the modelis usable in a full simulation and reconstruction chain, whichoffers a further handle for evaluating physics performance of themodel. As inference time is a crucial consideration for agenerative model, the pre-processing and hyperparameters areaggressively optimised, achieving a speed up factor of two orders ofmagnitude over Geant4 at inference. research-article