This LoRA is trained on screenshots from The Little Mermaid I, II, and III, capturing the animation style and distinctive characters of the beloved Disney films. The animation, known for its vibrant underwater scenes, detailed character designs, and enchanting musical sequences, brings to life the adventures of Ariel, her friends, and her family. This model can recreate both the magical underwater scenes and the unique, stylized look that define the series. This model is trained on Pony Diffusion V6 XL, so it possesses Pony's ability.
Usage Tips:
Use with Pony Diffusion or a Pony-based model.
Use with a pony prompt.
Trigger words for characters, use character name:
Girls:
ariel,ursula,melody,queen athena,morgana,marina del rey.Boys:
eric,tritor.Animals:
flounder,sebstian,scuttle.
You can also add LoRA for mimicking the style.
Description
This model is trained with screen capture from the little mermaid I, II and III on Pony Diffusion V6.
FAQ
Comments (10)
This was incredible. Could you give me some advice on how to train a Lora with several characters?
It is very similar to training with a single character but you just need to prepare data for several characters. Everything else is pretty much the same.
All very well. ^^ Let me ask you. In training, you define all the details of the characters (such as Arie, long hair, red hair...? or just define the name (for example: Arie). Another point, in training would it be good to put high values of Epochs or focus on Num Repeats? And in a set of 100 images what would be the more or less sufficient value of Steps?
@Matsu22 My training data is downable. You can look at it when I explain. I only add the name of the characters to the prompt the rest like long hair, red hair are generated using w14 tagger. For this dataset, I test as an experiment using drop out some tags randomly. I believe the model can lean to put the feature of long hair and red hair to the name prompt rather than putting it in the prompt every time we create an image.
@Matsu22 I use the number of repeats as a tool to make a model focus more on some set of images. such as focus more on face then the full body. Then, we can adjust the number of epochs. I keep in mind that I always can keep adding the number of epochs if I'm not satisfy with the result.
Thank you very much for the information. It helped me a lot ^^
A few of the examples in the showcase really capture that VHS animation vibe. I wonder if that's something you could focus on in a future lora. Combining a bunch of different Disney movie styles of the era to make an overarching "Disney VHS" style Lora
I always find the Loras that try to do characters+style always lean too heavily on the characters and you tend to have those characters show up unprompted when trying to use the style on its own.
Actually, that is my plan. I want to create Disney's cartoon-style LoRA by slowly collecting each movie one by one and creating a LoRA for them. It’s not actually Disney VHS style, but it is quite similar. I think if I am able to tag every image properly, I might be able to separate styles from different eras. Allowing to capture various animation styles from different periods
@titansteng Good to hear. To quote a a certain galactic emperor, I'll be following your career with great interest.
can you update it to also include ariel's sisters?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.