Hey everyone! 👋

Presenting my first dabble with a style LoRA based on JAB Comix's Curse of the Black Mantis (COBM)!

The Story Behind This LoRA

I recently developed a separate Marita LoRA — check it out here!, also based on Curse of the Black Mantis. Since the original references weren't always ideal, I used AI-generated images to help develop her. However, I found that getting her appearance to look just right also required a separate JAB-style LoRA.

That experience gave me the idea to try training a style LoRA of my own!

This project represents around 15 days of learning, experimenting, tweaking, and getting things right. I'm pretty happy with how it turned out, and I'm excited to see what everyone thinks of it and what you create with it!

What This LoRA Does

The primary goal is to reproduce the JAB Comix art style, but I've also baked in all six major characters from the comic. Their appearances turned out better than I initially expected!

That said, this is a style LoRA first, with character concepts as an added bonus. Depending on the checkpoint, prompt complexity, and other LoRAs you're using, you may need a few attempts to get a character exactly right. You might also notice minor accessory or outfit bleeding, but careful prompting should help keep things under control.

Basic Information

• Base Model: Anima 1.0

• Recommended Weight: 0.9

Main Trigger Word

COBM

Combining with Other LoRAs

This LoRA works well alongside other character LoRAs. Depending on the LoRA you're combining it with and your preferred look, try weights between 0.55 and 0.9.

Personally, I find 0.70 to be a sweet spot for combining it with other LoRAs, although your mileage may vary.

Character Triggers and Tags

All six major characters have been baked into the LoRA. Use their individual triggers when you want to generate them specifically.

The character appearance tags can also be reused with other characters for cosplay-style generations. Check out my reference images to see some examples!

I've included the main appearance and outfit tags below. You don't need to use every tag in every prompt. In particular, only include footwear tags when generating a full-body image or when the pose is likely to show the character's footwear.

1. Marita

Appearance

jabmarita, mature female, long wavy black hair, yellow eyes, red lips, huge breasts, pendant necklace

Outfit

purple strapless dress, thigh slit, black high heels, ankle strap

2. Anita

Appearance

jabanita, black hair, blue streaks, blunt bangs, blue eyes, black lips, large breasts

Outfit 1

black sleeveless crop top, brown fingerless elbow gloves, black miniskirt, thigh boots

Outfit 2

black dress with zipper pulled down, thigh boots

3. Aurelia

Appearance

jabaurelia, blonde hair, ponytail, green eyes, large breasts

Outfit 1

yellow t-shirt, jeans, white baseball cap

Outfit 2

black off shoulder top, red skirt

Note: Aurelia's canonical outfit shouldn't expose her midriff. If the model keeps generating a visible midriff, try adding midriff to your negative prompt or explicitly prompting for the top to cover it.

4. Emma

Appearance

jabemma, red hair, short hair, green eyes, freckles, small breasts

Outfit

maid outfit, black dress, puff sleeves, off shoulder, corset, white apron, short skirt, black thighhighs, high heels, black choker, white maid headpiece, protective glasses

You can remove protective glasses if you want to generate Emma without them.

5. Darius

Appearance

jabdarius, black spiked hair, brown eyes

Outfit

black turtleneck, black long overcoat, black pants

6. Buck

Appearance

jabbuck, red hair, blue eyes

Outfit

red and yellow striped shirt, purple track pants, white shoes

My Generation Settings

These are the settings I used successfully during my testing:

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

These settings worked well for me with Anima 1.0. Other Anima-compatible checkpoints, LoRAs, and generation settings may require some adjustment.

Training and Examples

I tried to tag the training images as accurately as possible, keeping the focus on the comic's visual style while including character-specific details where appropriate.

The example images were generated using recorded seeds to make the results easier to reproduce.

Feel free to explore the examples for inspiration, experiment with different poses and outfits, and see how the style works with other characters and LoRAs.

Remember, the main purpose is the JAB Comix style. The six character concepts are an extra feature, so don't hesitate to experiment beyond the original cast!

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