All Disney Princesses LoRA Model - Ralph Breaks the Internet
If You Like This Model, Give It a ❤️
This LoRA model is trained on screen capture images featuring beloved Disney princesses from the movie Ralph Breaks the Internet, including Rapunzel, Snow White, Ariel, Aurora, Belle, Cinderella, Elsa, Anna, Jasmine, Mulan, Merida, Tiana, Moana, and Pocahontas. It's designed to be an All-in-One Model for your creative endeavors.
Check out SDXL Version Here: https://civarchive.com/models/212532
Bonus Characters
Additionally, this model includes representations of other characters like Vanellope, Shank, Raya, Namaari, Asha, and Rapunzel with short hair. While these extras are included, the quality may vary.
Quick Tips
Direct Name Usage: You can summon any princess by directly invoking their names, such as
Anna,Ariel,Aurora,Belle,Cinderella,Elsa,Asha,Jasmine,Merida,Moana,Mulan,Namaari,Pocahontas,Rapunzel,Raya,Shank,Snow White,Tiana,Vanellope.LoRA Scale: Optimal LoRA scales range between 0.6-0.7 for close-up character generation and 0.3-0.5 for full-body generation. Combining these scales using the ADetailer plugin can yield enhanced results. Ensure an adequate number of sampling steps for better output.
Base Models: Although various base models can function, models with animation styles like DreamShaper, RealCartoon, RevAnime, or AnimeMerge are preferred for optimal performance.
Enjoy exploring the LoRA!
Description
This is a 1st version.
FAQ
Comments (19)
Nice! How do you prevent the charcters from blending into each other?
For the 1st version, no prevention. Just use 1 character at a time
@titansteng I meant during training - I tried to train 2 characters into 1 Lora, but the result was more of a mix
@_1_ I don't have any prevention method during training. I also notice the problem, may be I'll improve it in the next version. If your training have the mix problem, may be increase the rank and/or train it longer..
Best way to do this is probably with an inpanting mod and some revisits. Prompt x number of girls, but only use the one name. Then mask over the others in turn, changing the name each time.
If that doesn't work, try breaking down your normal image size into equally sized slivers, generating each individually, putting these all side by side via image manipulation software (ala Photoshop), then take that unified image back in. Prompt all the characters THIS time, and set the denoise between 0.4 and 0.7 or so.
Edit: I just read your clarifying comment. Well, leaving this here in case anyone gets the same idea I did lol.
@Anistasiata thanks, but again I mean during training. SD 1.5 knows thousands of celebs that don't blend into each other, but if I train just 2 ppl into a lora I get a mix, no matter how I tag them. I need to try harder... My dream would be a single Lora that has all 47 or so of my charcters :D
@_1_ I have multiple lora and that can remember multiple characters. For example, https://civitai.com/models/193225/sprite-art-from-jump-superstars-and-jump-ultimate-stars-or-pixelart-ai-model this one can remember over 100. I believe your dream is passible.
@titansteng That's awesome 😱
Poor Jazmin... She is the only one that doesn't fit the crowd
Okay, open ended question I know, but what could I be doing wrong such that when I use this lora in A1111, I get very garbled results? The XL version seems to work fine when I use it with XL models. But this 1.5 version seems to give me really messy junky looking images when I use it with 1.5 models. Any common mistake that might lead to that?
What's the base model that you use? Have you tried a different base model? If the problem still occurs, maybe it's just a normal divergence problem. This problem can be fixed by increasing the number of steps, reducing cfg scale, reducing the lora alpha, or doing everything together. For this model, I only used lora alpha = 0.4 for the sample images.
@titansteng Yeah I just figured out and came back to say, it's because I was using the lora at 1.0, I changed it to lower values and it worked much better.
Update. 🙏
This project have no update because I move my focus to SDXL and Pony Diffusion. You can checkout the similar version of this model here: https://civitai.com/models/212532/all-disney-princess-xl-lora-model-from-ralph-breaks-the-internet
titansteng Hi. Would you mind if I use your dataset to train SD1.5 lora?
Cum_Miser no problem
titansteng Thank you.
Has anybody found any combination of this LoRA and ANY model that can produce a character that looks like Rapunzel? I've lost count of how many models, embeddings, variations on positive and negative inputs, and never once had anything come out even close. Half the time the face is ... disturbingly messed up. Not even trying to get anything fancy, just some sort of base line.
Very nice
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.