I've updated the topic title to better reflect the pressing issue. I'm currently getting inconsistent behavior running this journal file.
The consistent part: I'm running the same journal file with the same case/data file on the same hardware (in some cases, the same exact computer) using ANSYS 19.0. The code runs perfectly fine, exits without errors, and always outputs images.
The inconsistent part: The images will sometimes not have the actual geometry in them.
I've combined results from three different runs: two unsuccessful, one successful. They're designated by their job numbers. They contain journal scripts, transcripts and images from the runs.
6634049 (Drive link here) is from the original posting. The first two images are perfect, but the third is rendered at an odd angle. This was run using "-gu" and "-driver null" flags for fluent
6641572 (Drive link here) is a completely successful example. This has more images than the first, utilizes the "save-view" function in the Fluent TUI, and has the axes. This was done using "-gu" and "-driver x11". The driver was changed as a colleague using the exact same hardware is able to get correct plots using similar methods.
6641591 (Drive link here) is a complete failure. The only difference between this and the previous run (6641572) is a change to taking images every other timestep. This also has the axes shown.
Focusing on 6641591, Note that the "[variable]0.png" images were not taken using the "every timestep" sequence of commands, while the rest were. Notice that the axes in "[variable]0.png" are in the correct position (negative x axis pointing at camera) while all others are not in the correct position (positive z axis pointing at camera). This says to me that the view-state is not being updated correctly for some reason.
Note that these (inconsistent) results have been reproduced dozens of times in both batch mode (no DISPLAY variable set, purely scripted on remote compute nodes) and interactive modes (DISPLAY variable set, fully scripted on interactive virtual machines on remote nodes).