Fooocus
Fooocus copied to clipboard
lllyasviel
【BUG】it seems that seed_random is not effective anymore.
After the support for wildcards was added, it seems that seed_random is not effective anymore. Regardless of whether I check seed_random or not, the seed value keeps increasing based on the image_number. Even if I provide a specific seed value, it still increases according to the image_number. I need to achieve the following functionality: when seed_random is checked, the seed value should not be influenced by the image_number, and it should be a random value each time. When seed_random is not checked, I want to be able to enter a specific seed value to keep it fixed, and only change the wildcard representations of prompt words to generate the desired number of images.
when seed_random is checked, the seed value should not be influenced by the image_number, and it should be a random value each time. When seed_random is not checked, I want to be able to enter a specific seed value to keep it fixed, and only change the wildcard representations of prompt words to generate the desired number of images.
your logic will break the basic thing:
check random, generate 4 images, uncheck random, generate, get the same 4 images.
a proper way to do this is to add an option "only change wildcard when seed is fixed" to the dev tools.
your request is noted and however depend on workloads. you may want to ask several forks if they put wildcard and non-GPT prompt manipulation in higher priority.
also take a look at the artists https://github.com/lllyasviel/Fooocus/blob/main/wildcards/artist.txt
if use use seed iteration, then the results will always be one artist name start with same letter, like starts with "a".
I agree with OP that this isn't expected behaviour.
Scenario with no wildcards: If i select a fixed seed and generate 10 images, I don't expect 10 identical images. It makes sense to ignore the seed selected by the user and increment it for images 2-10.
Scenario with a wildcard present in the prompt: If I select a fixed seed and generate 10 images, I do expect the seed that I've entered to be adhered to and for the wildcards to increment.
In both scenarios, the purpose of a batch with a fixed seed is to control all but one variable and measure it's impact. The current behaviour fails to control all but one variable in scenario 2. Ideally, it would adhere to the seed and increment wildcards one at a time.
