I tested version 2.7.1 a bit with LM Studio. It generally works well.
As a test, I ran 10 images in a row with Gemma 3b and found that entries were still being created for the keywords that shouldn't be there (date, coordinates).
In a second run, all entries were correct (?)
Now I've started a third run with only one image, using a different AI model each time.
Firstly, the error with the entries doesn't appear everywhere, and it can't be reliably reproduced, but it hangs every time with the zai-org/glm-4.6v-flash model.
Log attached.
lrgenius-server.log