So, the long-awaited LoRA for Illustrious and NoobAI-based models is finally here!
I know I took my time with it, but I had to make sure it does what it's supposed to and doesn't break the site's new rules. The second part is fine: this model focuses on the character's general look, not the specific actresses or cosplayers. It's a nice blend of all of them, but you can still spot the classic Mystique features from the original X‑Men trilogy.
But the first condition is not so perfect. I messed around with tons of training runs and datasets, but I still can't get the scale placement exactly like the original Mystique without some help from Inpaint and Img2img. I'm talking about the ???? areas of skin on the thighs, which shouldn't have been so large.
But delaying the release further just because of this felt wrong, especially when the next generation of checkpoints like Z‑Image is on its way. So I decided to upload the most stable version as-is to stop keeping you waiting.
Recommended weight: 0.8 - 1.0
Helpfull tags:
Prompt: scales on [body part], covered [body part], no ???????, dot pupils, dark nails, ...
Negative prompt: slit pupils, ?????????, swimsuit, ...
Using ADetailer , Inpaint or Hires fix is also recommended.
It wasn’t easy, but after several attempts, I managed to train pony-like models to create images with this character.
However, due to the scales on her body, even these models sometimes can’t tell SFW from ???? or draw her with something like a swimsuit or torn jeans. So, I recommend adding (no ???????) to the prompt and including ???? tags and clothing-related tags in the negative prompt to accurately reproduce her look from the movie.
Other tips:
1. Recommended weight - 0.8 and below for anime and 2D styles; 1.0 and above for realistic and semi-realistic images. Personally, I prefer 1.0 or 0.9.
2. Add 3D, realistic to the negative prompt, if you are making anime, cartoon or comics -style image.
3. Some realistic models make her skin too pale. I had the same issue with my Neytiri from Avatar model. So, I recommend using Pony XL and Autism Mix as base models.
4. If she starts to look like a dinosaur, add source_furry to the negative prompt.”
5. Works better with Adetailer and Hires fix
P.S. I also want to thank the model testers who explained the difference between actual person and movie character to me. Now I can confidently remove the restrictions for this model.
There are a lot of Mystique LoRAs on the internet, but almost none of them are based on her appearance in the movies. So I decided to fill that gap because her designs from the movies deserve more attention. The model is a bit raw, but still works fine with realistic, semi-realistic and comic book models.
I recommend these trigger words: blue skin, red hair, yellow eyes and scales (or ????????). Also, the weight value should be around 0.6 - 0.7, but this is highly dependent on the model used. And also I highly recommend to use Hires fix for better experience.
This version can be slightly realistic regardless of style. I recommend adding artist names or specific work titles to your prompt to get the desired result.
1. The rights to reposted models belong to original creators.
2. Original creators should contact SeaArt.AI staff through official channels to claim their models. We are committed to protecting every creator's rights. Click to Claim
