Fldgen provides a method for emulating earth system models (ESMs), including the internal variability typically found in ESM output. Output from fldgen is spatially resolved and produced at annual time resolution. Care is taken to preserve the output variables' correlation properties over space and time. The emulation incorporates stochastic elements so that many independent realizations of a scenario can be generated. This differs from existing climate model emulators, which normally provide only a deterministic mean climate response. The algorithms in fldgen are designed to preserve the correlation structure of the climate variables in space and time. For example, if the temperature for a year is above average in one grid cell (roughly the size of a US county in the data we are using), then it stands to reason that neighboring cells should also have above average temperatures that year. This effect gets generally weaker with distance, but some surprisingly long-range connections, such as the El Niño phenomenon, are known to occur. Likewise, much of the time variation is periodic or quasi-periodic, and this must also be taken into account. Fldgen takes as input a collection of outputs from ESMs and analyzes them to learn the patterns of correlation described above. This process takes roughly 10-15 minutes on a midrange workstation, depending on the size of the input data. Once this training process is complete, the software may be used to generate outputs for any climate scenario desired (climate scenarios are specified by the future trajectory of global mean temperature), as many times as desired. Output generation takes just a few seconds for each realization.