A LoRA to let you dress characters in the jacket from the second Mad Max film, The Road Warrior. Variations of the jacket have become an staple inclusion in future post-apocalyptic media so I decided you should be able to generate characters wearing it. Please show me the pictures you make with this.
Prompt: trwjkt, leather jacket, asymmetrical clothes, asymmetrical sleeves, single pauldron, zipper, zipper pull tab, (V1: Replace trwjkt with Road_Wargear, or just leave it off and let leather jacket do the work)
Optional: Fingerless gloves, black gloves, belt, belt buckle, black shirt, leather pants, black pants, thigh holster
Tagged but don’t expect good results: Adjustable wrench, binoculars
Also tagged, but even less likely to work: gun, holding gun, double-barreled shotgun, sawed-off shotgun
Negative: Road, street (seems to be reading that from the trigger word. Worst case, just drop the main trigger since everything was also tagged leather jacket), head out of frame
Training data was made 9 pieces of fan art, 7 pictures from a couple store listings for replicas, an illustrated movie poster, a screenshot, a promotional still, and a piece of promo art (home video release cover?). Heads were cropped off as needed to avoid this making Mel Gibson.
Edit: V2 added. Changes trigger word to prevent roads from spawning.
Description
Changed trigger
FAQ
Comments (2)
So you are allowed to make LoRAs based off screenshots from live-action movies as long as you crop off the actors’ heads? A while ago, I sent a question to the Civit.ai support team asking about this and how/if I’m allowed to make a LoRA using screenshots from live-action movies, but they never responded.
If this is allowed, that’s wonderful to know, because I’ve been dreaming of making a LoRA based on the fabulous fashions of the Disney Descendants movies. <3
This was made before the total ban on models meant to depict real people so the cropping was just done for functional reasons but paid off long term (note the fanart was of characters wearing the outfit). I have however asked about updating a set of clothing LoRAs that used photos of random models and was told (quote)
"Yes, training with that data is still allowed. The rule is about the output: a resource must not reproduce a real person. A dataset of stock photographs with no faces and no identifiable individuals, varied enough that no particular person could be reproduced from it, does not run into that. That is the same answer you were given earlier and it still stands.
What is not allowed is a resource that reproduces a real person, whether that comes from training on a specific individual or from a live action character whose likeness is the performer who played them."
Among other considerations, you'd still need a variety of adults (and mannequins) wearing whatever outfit you wanted to train. I don't know the movies in question, but if Disney made them there's likely enough cosplayers out there any outfit could be done without even touching the original.