I don't know if it affects other parts of randomization; but setting * in the checkpoint field and 2 batches with a size of 2 or 1 batch with size of 2 causes half of the generated images to be made with the default model that was loaded in the "stable diffusion checkpoint" menu in the top left corner of txt2img.
Instead of loading a new checkpoint each time, half of the time, SD-forge reloads that model. That makes half of the images randomized and the other half not randomized.
Could you check it and fix the extension so it works with SD forge?
I don't know if it affects other parts of randomization; but setting * in the checkpoint field and 2 batches with a size of 2 or 1 batch with size of 2 causes half of the generated images to be made with the default model that was loaded in the "stable diffusion checkpoint" menu in the top left corner of txt2img.
Instead of loading a new checkpoint each time, half of the time, SD-forge reloads that model. That makes half of the images randomized and the other half not randomized.
Could you check it and fix the extension so it works with SD forge?