First of all, model is for fun. Make cool images, make Harry Dean Stanton as Superman, make him as a silly clown, whatever. No selling, no gross stuff. Be nice. Be kind. Be smart and be respectful!
Typically works best with prompts starting with "hds1, man, solo, blank blank blank" and then add "<lora:HarryDeanStanton-06:1>" at the end. Lora works best with cyberrealistic, but is pretty versatile. I have all ten epochs but am uploading the one I use most. If you would like the others just comment and I can upload it.
Part of an ongoing to make LORAs of cult actors and more obscure figures for Civitai. Far too many beautiful actresses and waifus! Time for when men where men, life was hard and everyone died from alcoholism at 55!
Harry Dean Stanton (July 14, 1926 – September 15, 2017) was an American actor, musician, and singer.[1] In a career that spanned more than six decades, Stanton played supporting roles in films including Cool Hand Luke (1967), Kelly's Heroes (1970), Dillinger (1973), The Godfather Part II (1974), Alien (1979), Escape from New York (1981), Christine (1983), Repo Man (1984), One Magic Christmas (1985), Pretty in Pink (1986), The Last Temptation of Christ (1988), Wild at Heart (1990), The Straight Story (1999), The Green Mile (1999), The Man Who Cried (2000), Alpha Dog (2006) and Inland Empire (2006). He had rare lead roles in Wim Wenders' Paris, Texas (1984) and in Lucky (2017).
Description
FAQ
Comments (5)
I have a question. You said you had all ten epochs. I notice when I make Loras I get like 10 safetensors in the output folder. Is one better to use than the other for likenesses? The 10th one better? The first one?
I don't know myself, I just try them all and I choose which one I like the best. Sometimes I'll use a different one just for fun sometimes but I think it's up to you. Someone who knows more about this than me can probably answer better!
Well whatever your doing it's great. Your likenesses are all very strong. My problem with Lora and creating likeness is they just don't work as strong as the checkpoint models I have made. The Loras very look close, but not close enough for me to confidently post them. I have to use Colab to create them as my laptop does not have enough gpu power. I would prefer to make better Lora's as they are more versatile and can be used with other models. Don't know why mine are not as powerfully close to the likeness as yours seem to be.
@tglunzer401 I use Colab as well, from the Make your own Loras, easy and free guide here. I collect about 18-20 images from DuckDuckGo image search and make sure they are very big images, all from the "wallpaper" or "large" section. It's also helpful to get images where the face is very clear and not wearing sunglasses, hats, etc
@dajamesbondsuperfan007Thanks for the tip! I used that guide as well. Perhaps my images are not originally high enough resolution. I will keep trying to make them better.