I've been doing some type of SD image creation for about a year, registered here since may, and I have had quite a blast as an hobbyist. One of the things I find annoying is trying to figure out how a model will look once I have it running. I know everyone has at least an example of two of what style, look, and feel the model will give, but with such massive use of LORAs, embedding, paragraphs for prompts, etc. where does the model end and the extra add-ons begin?
I am not trying to telling people how to showcase their models or anything they create, but I do find it hard at times when browsing the hundreds (maybe thousands) of models on this website. For example:
This is an image created only with positive prompts and nothing else. Its a nice image, but the hands are a bit wonky. Good thing I have some negative embedding for that:
Hmmm. The hair changed color and the hands still off... Also I want more details... More embedding and a LORA should help...
Floating popsicle... great... but when you compare the first image, arguably the intended style, they are almost completely different. And if I just use no embeddings or negatives and just the LORA for detail:
Still a difference in style.
I know that everything you put into a prompt can change styles and the look and feel, but how does anybody know if the perceived quality of a model is do to the training/merging of careful intent, or its a combination of LORAs, embeddings, control-nets, etc that are giving that look.
This is just some ramblings of a random internet stranger so take whatever I say with a grain of salt.
(Yes each image had the same prompts, seeds, CFG, etc with the only change of embeddings and LORA. I also have used embeddings in the two models I have made. I think I will have to put some images without those for a more accurate look.)