Romiks – Rafael Albuquerque × Kittyhand | Western Comic Style
I was looking through some old comics when I noticed that Rafael Albuquerque and Kittyhand had certain similarities in their artwork. That got me thinking: What would happen if I mixed their styles together?
So I gave it a try!
The original idea was to create a gritty, old-school Western comic-book style, combining bold shadows, expressive linework, detailed inking, and attractive character rendering.
Well... things didn't go exactly as planned. LOL.
This is my first MOSS-Image LoRA, and also my first attempt at training MOSS using AI Toolkit. I probably overcooked it a little while experimenting with the training settings.
I had a particularly difficult time getting the model to consistently reproduce the fine ink hatching I wanted, especially around facial shadows.
Still, I ended up liking the results!
The final style is something of its own. It has strong outlines, dark graphic shadows, high contrast, and a blend of traditional comic-book aesthetics with cleaner digital rendering.
Although I originally intended it for gritty comics, it works surprisingly well across different genres, including fantasy, science fiction, modern settings, and more.
## Recommended Settings
- Trigger word: romiksMs
- Recommended LoRA weight: 1.0
- Base model: MOSS-Image Base
- Sampling steps: 30–50 for regular generation
- Turbo compatibility: Works great with the 8-step Turbo LoRA using its intended fast-generation settings
Important: This LoRA was trained on the original MOSS-Image Base, not a Turbo or distilled model.
The LoRA generally favors darker colors, strong shadows, and high contrast, but it can also handle brighter scenes and colorful lighting.
Pale Skin Warning
Be careful when prompting for pale skin or very pale skin. The LoRA sometimes interprets these descriptions as extremely pale, grayish skin, giving characters an almost vampiric appearance.
If you want pale but natural-looking skin, combine the description with additional color information, such as:
pale skin, subtle pink undertones, natural skin coloration, soft rosy cheeks
You can also experiment with warm or neutral undertones to avoid the grayish appearance.
Speech Bubbles
The LoRA sometimes generates unwanted speech bubbles, dialogue balloons, or comic text.
Adding speech bubbles, dialogue balloons, text to the negative prompt may help, particularly when using CFG above 1, depending on your generation setup.
## Optional Lighting Boosters
These are not required to use the LoRA. They're just examples of how I like to experiment with lighting and color.
Strong contrast and directional lighting can help make the shadow work and fine ink details more noticeable.
Dark Violet / Magenta Lighting
Low-key cinematic lighting, nearly black ambient shadows, dim violet ambient illumination, intense magenta and purple rim lighting, sparse neon signs casting localized reflections, deep blue-violet shadow tones, minimal frontal fill, selective luminous highlights, dark atmospheric contrast.
Blue Cyberspace Lighting
Low-key cinematic blue cyberspace lighting, glowing cyan circuitry and holographic interfaces as the primary light sources, deep navy shadows, minimal neutral fill light, cool blue color cast across skin, clothing and environment, selective electric-blue highlights, localized light spill, dramatic comic-book contrast.
Warm Orange / Amber Lighting
Warm orange ambient illumination, deep brown and amber shadows, golden highlights, soft reflected light from nearby surfaces, subtle reddish-orange rim lighting, dark atmospheric contrast, selective illuminated areas, rich warm color separation.
Feel free to adjust the colors, intensity, and lighting direction to your liking. These are examples, not requirements.
## Sample Prompts
Fantasy
romiksMs, a mature woman warrior and a rugged man adventurer standing together in a medieval town square, leather armor, cloaks, swords, stone buildings, market stalls, distant castle, medium shot
Science Fiction
romiksMs, a woman with long blue hair wearing a black business suit, standing on a futuristic rooftop, holographic interfaces, floating magical swords, city skyline in the background
Modern Setting
romiksMs, a mature woman with long black hair wearing a leather jacket, sitting at a cafe with a cup of coffee, tables and windows in the background, medium shot
Action
romiksMs, a female mercenary and a rugged male soldier taking cover behind a damaged vehicle, ruined city street, debris, abandoned buildings, dynamic composition
## Training Information
- Training model: MOSS-Image Base
- Trainer: AI Toolkit
- Dataset: 110 images
- Published checkpoint: 5,390 steps
- Network: Rank 32 / Alpha 32
- Resolution: 768
- Optimizer: AdamW8bit
- Learning rate: 0.0001
Training involved experimenting with balanced, high-noise, and low-noise timestep distributions.
This was mostly an experiment to see how MOSS handles mixed comic-book styles. I'm still learning how to train this model properly, so I'll probably release an improved version once I get a better understanding of its training behavior.
For now, enjoy Romiks!
Feedback, suggestions, and example generations are always welcome.