A character lora of our favorite MILF, Helen Parr aka Elastigirl/Mrs Incredible from the movie The Incredibles for Krea2, trained with 38 images using Movie Still screenshots and 3D renders. All workflows are embedded in the images. Simply drag and drop!

Trigger word: h3len and Helen Parr , they can be used independently or together. Use it however you like depending on your prompting style! If you like what you see, feel free to buy me a coffee :D

Other important Trigger Words!

The Lora is also trained with different art style and costumes.

• Use 3D animated character render or photorealistic to change the art style between the movie accurate and photorealistic version of her. Regular Krea2 Turbo model will skew towards photorealistic by default if you don't specify the art style. Other models would differ.

• Use redsuit or silversuit to change her costumes. Not specifying her clothing will generate her either nude or in generic clothing.

• Use wearing a black domino mask to toggle her face mask. By default, it won't generate her wearing it. However, sometimes using 3D animated character render AND redsuit/silversuit would also generate her face mask. Use without her domino mask or use the lora at lower strength if you want her without it.

Recommended Settings

Lora strength;

• 1.0 - 0.9 : very consistent character designs even in long complex prompts. However, if you have her wearing clothes with the color red or silver, it may mistaken it for her superhero suit instead. I found that having very detailed prompt describing her clothes overcomes this.

• 0.7 : Best for flexibility. However, if you generate her wearing her superhero costume, either her redsuit or silversuit, it will mess up the logo on her chest. Recommended if you're not generating her costume.

• 0.55 : Absolute minimum value the lora can go before it starts generating other people's faces instead of Helen Parr.

Steps : 10

Scheduler : Euler/Simple

Note

There's a weird image artifact that generates "streaky" backgrounds that often happens when generating using the base FP8 Krea2 Turbo model. I found that lowering the lora strength helps. You could also keep it at 1.0 but also use other Lora such as "Mystic XXX" at 0.5 strength instead to minimize the artifact. Using different fine tuned model may generate different results.


I suspect there's something wrong with the dataset images. The current version is good enough, but I may release a ver2 if the lora get's enough attention.