This is a model that helps to create arachnid creatures, both male, female and monsters. Skin and hair colour can be changed. See example images for example prompts.
With the version 1 I had problems with the faces but it might be due to my lack of pony skills.
Description
FAQ
Comments (7)
Great work!
Please name the files properly. The v2 verison has same filename as the v1. This causes confusion when people download it and could accidentally override the old one or think they already downloaded it
I think this is intentional.
I follow the same structure myself. It's an indication that you SHOULD replace the old with the new.
If you want to maintain both, that's up to you.
With this workflow makes it easier to re-use old prompts, as you don't have to fiddle with numbers in the lora activation part of the prompt.
In the end it's a matter of preference and I may be in the minority here. I got similar requests on some of my models.
In the end I kept my method as I think it benefits more users that just wants replicateable prompts. This is why I don't leave the epoch number in there, even if I went for a non-final epoch.
@mnemic I can agree with both approaches. Perhaps a source of potential confusion is that A1111 will only show (and use) one version in its list if they have the same name and are in different folders. It can make it difficult to figure out which one is being used or to not even realize you have two versions. For example I use a "test" folder with new loras to see if they're of interest to me before I sort them into a different file or discard them.
@karetai yeah, totally fair. It would need renaming for A1111 as it combines them like that :(
@mnemic With most browsers this approach doesn't even work as they automatically rename it "file.safetensor" to "file (1).safetensor" on saving, which would cause duplicates for most people.
Also prompts become unreproducable when not having a unique name. A prompt for a specific image could generate a completely different result when you have a different version and its not visible from the prompt which version is used
and this gets even worse, when the creators don't put their name inside the file name and you have two "<character>.safetensor" for the same character but from two different creators, its impossible to figure out by merely looking at the file name and/or size if its the correct one w/o going unreasonable amount of work to temporary save it somewhere else and the find a way or tool to calculate their hashes and compare. This is madness
Thank you @karetai, @mnemic, and @TScitai for your inputs. I am fairly new at training models and generating images so it is much appreciated to get some feedback.
Just a heads-up: Before seeing these comments, I had already created a version 2 of the SDXL model and generated images (will be published today or tomorrow), not using a unique name. But I will give it some serious thoughts about using an unique name before I a create a new version of a model in the future.