Using Automatic1111, I've run in to this issue. I'd like to say I feel like this has only started happening recently as I surely would have noticed it before, but I can't say for sure...
Let's say I want to test the effect of a LORA by generating the same image with and witout the LORA applied:
What I get however is two identical images both with no LORA effect (i.e strength at 0.0). It appears only the first value in the comparison is being used. This can be confirmed by switching it around to:
This happens both using combinatiorial mode, or standard mode with a batch size of 2.
Generating the images separately without using a dynamic prompt (i.e same seed, one with strength 0.0, one with 1.0) gives the expected results. Also comparing other things other than LORAs with dynamic prompts still works as expected.
Using Automatic1111, I've run in to this issue. I'd like to say I feel like this has only started happening recently as I surely would have noticed it before, but I can't say for sure...
Let's say I want to test the effect of a LORA by generating the same image with and witout the LORA applied:
{ <lora:MyStyleLora:0.0> | <lora:MyStyleLora:1.0> }What I get however is two identical images both with no LORA effect (i.e strength at 0.0). It appears only the first value in the comparison is being used. This can be confirmed by switching it around to:
{ <lora:MyStyleLora:1.0> | <lora:MyStyleLora:0.0> }Now both images have the LORA at 1.0 strength.
This happens both using combinatiorial mode, or standard mode with a batch size of 2. Generating the images separately without using a dynamic prompt (i.e same seed, one with strength 0.0, one with 1.0) gives the expected results. Also comparing other things other than LORAs with dynamic prompts still works as expected.