If you feel like showing some love, awesome.

But the best kind of support?

Use the model. 🛠️

Post your results. 🖼️

Let the art speak. 🎯

🧙‍♂️ RazbolicX - Semi-Realistic Fantasy LoRA

RazbolicX - Semi-Realistic Fantasy LoRA is a community co-created fantasy LoRA made with DiabolicX, the winner of my community co-creation event.

The original winning idea was Real Fantasy, but after training and testing, I think Semi-Realistic Fantasy is the more honest and accurate description.

The model can produce very photoreal fantasy images, but depending on the prompt, base model, and LoRA strength, it can sometimes lean into a more polished CGI / 3D render texture. So instead of calling it strictly “real fantasy,” I prefer to describe it as a semi-realistic fantasy LoRA with strong photoreal potential.

🤝 About the Community Co-Creation Event

RazbolicX came from my first community co-creation event.

Instead of just picking the next model idea myself, I asked the community to submit LoRA concepts. People shared their ideas, explained the kind of model they wanted to see, and then the community voted on which one should be made.

DiabolicX won the event with his idea: Real Fantasy.

That idea became the starting point for this model. During training and testing, the final result ended up being better described as Semi-Realistic Fantasy instead of pure “Real Fantasy,” because while it can go very photoreal, it can also sometimes lean into a polished render / CGI texture depending on the prompt and LoRA strength.

So this model is here because the community helped choose it, and because DiabolicX brought the idea that shaped its direction.

I’ll link the original articles here if you want to see the full event, the submitted ideas, and the final vote.

Links to previous articles related to this event:
Community Co-Creation Event! [CLOSED]
Community Event Reminder [CLOSED]
Community Co-Creation - Poll Time [CLOSED]

I really enjoyed seeing the ideas people came up with, so there’s a good chance I’ll eventually do another co-creation event in the future.

⚔️ What it does

RazbolicX was built to help create mixed fantasy imagery.

It works well for:

• Classic fantasy characters

• Weird fantasy concepts

• Fantasy / sci-fi hybrids

• Anachronistic character ideas

• Mixed-era worldbuilding

• Anthropomorphic characters

• Unusual fantasy occupations

• Creatures, warriors, villagers, monsters, and oddball fantasy personalities

The dataset was not built around one narrow fantasy trope.

It includes a mix of grounded fantasy, strange fantasy, sci-fi touches, and mashed-up concepts... things like crusaders in fairy villages, monks with chainsaw staffs, orc office workers, unusual fantasy races, and characters that feel like they belong in worlds where genres collide.

🔥 The key idea

RazbolicX is not meant to force one single fantasy look.

It is meant to help generate fantasy images that feel more alive, more varied, and more unexpected.

The model was created with DiabolicX’s type of image ideas in mind: not just standard heroic fantasy, but mixed concepts, weird combinations, and characters that feel like they come from strange little stories.

A knight does not always need to stand in a castle.

An orc does not always need to be a warrior.

A monk can carry a chainsaw staff.

A crusader can wander through a fairy village.

An anthropomorphic character can sit naturally beside classic fantasy races.

That is the kind of creative space this model was trained for.

🧬 Dataset direction

The dataset contains a mix of:

• Fantasy characters

• Sci-fi fantasy characters

• Mixed-era outfits and settings

• Anthropomorphic characters

• Male and female characters

• Creature-like characters

• Classic fantasy races

• Strange occupation / race combinations

• Grounded fantasy portraits

• More unusual worldbuilding concepts

And for once, I tried not to bias the dataset only toward women. 😆

A lot of my models naturally lean into beautiful female characters because, well... that is often what I enjoy making and testing. But DiabolicX’s main complaint about some of my other models was that they were very obviously woman-focused.

So with RazbolicX, I made a real effort to include more men, creatures, fantasy races, anthropomorphic characters, and mixed character types.

It can still make attractive fantasy women very well, of course, but it is not meant to be only that.

⚙️ Recommended Settings

These are the settings I tested most and had the best results with:

Sampler + Scheduler: dpmpp_2m_sde + beta
ModelSamplingAuraFlow: 7.00
Steps: 12
CFG: 1.00
LoRA Strength: 0.65

Strength Notes

0.65 is my personal favorite and the setting I recommend starting with.

The model is usable from 0.20 to 1.00:

0.20 - 0.40 = lighter influence, keeps more of the base model behavior, good for subtle fantasy enhancement

0.50 - 0.70 = best balance of fantasy influence, realism, and stability

0.65 = my favorite tested strength

0.75 - 0.90 = stronger fantasy identity, more visible LoRA influence, but more chance of render-like texture

1.00 = very strong influence, can push the CGI / 3D render look a lot

At 1.00, the model can still create strong images, but it becomes much less subtle. It may give more dramatic fantasy designs, but also a more polished rendered texture depending on the prompt.

For the cleanest semi-realistic / photoreal fantasy results, I recommend starting around 0.60 - 0.70.

For weirder and more stylized fantasy concepts, you can push higher.

⚠️ Notes

This is not a strict photoreal LoRA.

It can produce photoreal-looking fantasy images, but some prompts may create a more CGI / 3D render feel.

Higher LoRA strength increases the chance of rendered textures.

The model works best when you give it an actual concept, not just “fantasy woman standing in forest.”

It was trained for variety, so outputs may shift between classic fantasy, weird fantasy, and mixed-genre fantasy depending on prompt.

The dataset includes more gender and character variety than most of my previous models.

Anthropomorphic characters are supported, but not the only focus.