Hey everyone! 👋

Presenting my Marita (Black Mantis) LoRA from JAB Comix's Curse of the Black Mantis!

Another one of those rarer characters whose design I immediately loved at first glance.

This was probably one of the hardest LoRAs I've worked on, simply because there is so little source material available for her, and what exists isn't always in the best quality. I spent a lot of time building and refining an AI-curated dataset to bring her design together. The dataset isn't perfect, but I'm happy to finally share the result!

I'm really interested in seeing your creations and hearing your feedback, so do share what you make! ❤️

Model Information

Base Model

• Anima 1.0 — Base checkpoint

• The LoRA should also work well with other Anima-based checkpoints.

Recommended Weight

0.9

Trigger Word

maritacobm

Style

western comics \(style\), cel shading

Although I tried hard to preserve the comic-book style, some subtle stylistic variation crept in during the AI-assisted dataset creation and refinement process.

If you'd like to bring the results closer to the intended Western comic-book look, I recommend trying the style LoRA below at a strength of 0.55. This worked well in my testing; see my example images for reference.

Suggested Style LoRA: LazMadeFromLazer Style

Appearance

mature female, long wavy black hair, amber eyes, red lipstick, red fingernails, white pendant necklace

Default Outfit

strapless sweetheart blue-purple brocade gown, single high thigh slit, asymmetrical dress

Optional tags for certain poses:

black ankle strap high heels, white panties

Using the Tags

The appearance and outfit tags can be used separately, depending on what you'd like to generate.

For example, you can retain Marita's appearance tags while changing her outfit completely, or combine her appearance with the default outfit tags above.

The LoRA is trained primarily for Marita's character identity and appearance in her comic-book design, while allowing for different outfits and poses.

For full-body generations, add the footwear tags when needed. The optional underwear tag can also be used when relevant to the intended image.

Suggested Generation Settings

My testing was done using a Turbo LoRA alongside the additional style LoRA mentioned above, with the following settings:

LoRA Weight: 0.9
Sampler: Euler a
Scheduler: Simple
Steps: 12
CFG: 2.5

These settings worked well in my testing, but other Anima-compatible checkpoints, LoRAs and settings may require some adjustment.

Training & Examples

I tried to tag the training images as accurately as possible, and the example images were generated using recorded seeds to make the results easier to reproduce.

Feel free to refer to the example images for inspiration, including different poses, outfits and variations.

Hope you enjoy the LoRA, and I look forward to seeing what you create! ❤️