Has the behavior changed when using X/Y/Z script?

[anothertal3]

Until very recently (today?) I was able to do something like the following:

portrait photo of {man|woman} standing at __places/all__ , __head/hairstyles__, {medium long shot|medium shot} <lora:add_detail:0.1>

The wildcard input originates in a yaml-file.

I could then setup a X/Y/Z script with the following parameters:

X = "Prompt S/R" = <lora:add_detail:0.1>,<lora:add_detail:0.5>,<lora:add_detail:0.9>
Y = "Seed" = -1,-1,-1

Upon execution I would receive a 3x3 grid. For each row the value would have been randomly chosen only ONCE at the beginning. This allowed me to compare the influence of different LoRA settings because the person, hair style, place and shot type would not change.

Now, the behavior seems to have changed:

• Inline variables appear to stay the same during one row (a 'man' would always stay a 'man')
• Wildcards appear to change on each and every picture (the 'man' would now be placed at different locations even within one row)

For my purpose it's now pretty much useless.

Has some default behavior changed? How can I reset this to the way it acted before?