Trained on Dreamshaper_v8
Recommended weight: 0.8-1
This is the character "Steamboat Willie", a prototype of what would eventually become a behemoth in the world of entertainment.
As of January 1st, 2024, This character is now in public domain.
Training images were pulled from three public domain disney animations. "Steamboat Willie", "The Galloping Gauncho", and "Plane Crazy"
I tried my best given the nature of the old and grainy source film.
This Lora is licensed as CC0. This effectively gives you the right to do whatever you want with it.
Note: Version 1 was removed due to copyright concerns.
Description
FAQ
Comments (18)
he has gloves.. so technically, Disney can claim copyright on this.
He had gloves in "The Mad Doctor", which is public domain (due to a failure in copyright renewal), so following the same logic as Steamboat Willie it should be fine. There are other shorts between the two that are copyrighted, so I assume it is sort of a grey area. I'm not a lawyer.
@dillon101 Literally every article makes a POIGNANT AND CLEAR POINT that Mickey with gloves is a no-go and under copyright.
@lifania I see, I was not aware of this, I'll make a new version then. Thanks by the way.
@dillon101 Grey area with Disney means Disney wins..
The theatrical poster of that time gave Mickey gloves and red shorts, and that's also in the public domain.
@tonytins I'm not necessarily aiming for mickey mouse. Many checkpoints do that well out of the box. I'm doing this for two reasons. One, I hate what Disney did to copyright, and they have a long history of doing things that would be considered mean, rude or unnecessary. If I can have some fun at their expense I consider that a win-win.
Two, I really like the character and design, and this is the art equivalent of bringing dinosaurs back from the dead with their DNA, which is cool as hell.
@dillon101
I'm all for sticking it to Disney. Train on! We all make mistakes hehe
@dillon101 Fair enough, and I hate what Disney did to copyright too.
You should have trained him on SDXL instead of 1.5
I want to say SDXL might have handled it a little better
@qwx Yes, definitely, It will happen.
For @dillon101: The original silent version of "Steamboat Willie", "Plane Crazy" and "The Gallopin' Goucho", released in 1928 as silent cartoons, are in the public domain.
Next year will be an important year,, as there are several other 1929 Mickey cartoons that introduce the gloves, such as:
"The Barnyard Battle" - 1929
"The Karnival Kid" - 1929
"The Plow Boy" - 1929
and many others.
The confusion is because Title Card and Promotional Posters for the sound versions of "Plane Crazy" and "The Gallopin' Goucho" (1929) introduces the gloves. And the copyright year on the title card says 1930 in Roman Numerals, not 1929
So maybe play it safe, include "Steamboat Willie", "Gallopin' Goucho" and "Plane Crazy" the cartoon, but not the promo poster or Title Cards, and not "The Mad Doctor" Mickey.
Thanks.
Got it, right now the main issue I am having is getting reasonable quality images, due to the film grain and shoddy resolution.
There is a good quality version of Steamboat Willie (4k remaster), but I'm having trouble with the other two, I found 480p Gaucho and Plane Crazy, which were good enough for a handful of close up images but that's all.
The dataset is very noisy.
Create a 2nd Lora with a stack of currently trademarked Micky's, particularly those early cartoon with white gloves.
Then use that Lora at a negative value, along with this lora to potentially eliminate the white gloves.
I'm having difficulty at the moment getting reasonable images without the white gloves, whilst keeping the scene generalized.
Negative prompting for 'gloves', 'white gloves', hands, fingers, etc have some effect. Although a lot of images have the white gloves.
Trying to use positives like 'black hands', 'fist', 'blobby hands', etc... also help, however they tend then produce a third ear on the head... which makes sense considering how similar a steamboat willie hand is to a mouse ear in dimensions and colour.
Not to mention using a 'color' prompt word bleeding .. keeping it to black and white helps mitigate this.
Are you still using the V1 model? I don't have this issue with v2.
Maybe something is different. Could you try copying the prompt and model from one of the posted images?