Luis Royo Style - FLUX.2 Klein LoRA
What does this model do & What is it for? This LoRA brings the iconic dark fantasy, ethereal, and sci-fi aesthetic of the legendary illustrator Luis Royo to the FLUX.2 Klein architecture. It transforms standard concepts into masterful airbrush paintings characterized by melancholic moods, stark contrasts, and highly tactile environments.
What is this model good at?
Extreme Material Contrast: It excels at juxtaposing rough, weathered materials (rusted iron, rigid leather, heavy furs) with delicate elements (sheer lace, glowing silk, smooth lifelike skin).
Chiaroscuro Lighting: Generating dramatic, theatrical lighting with deep midnight shadows and glowing ethereal or stark white highlights.
Atmospheric Backgrounds: Producing moody, desaturated environments (fog, ruins, deep space) that make the vivid, rich colors of the main subject pop perfectly.
Versatility: While trained on dark fantasy, it beautifully adapts to cyberpunk, post-apocalyptic, and high-fantasy concepts.
What is it bad at & How should it NOT be used?
Bright, cheerful, or sterile scenes: The LoRA naturally injects a moody, somber, or melancholic undertone.
Hyper-realistic modern photography: The model is strongly anchored in a traditional "masterful airbrush painting" style. Trying to force a 50mm camera lens prompt will fight against the LoRA's core training.
Flat anime or cartoon styles: It relies heavily on volumetric shadows and realistic anatomy.
⚙️ Optimal Settings & How to Use
To get the absolute best results, please respect the exact trigger structures used during the rigorous training phase:
LoRA Weight: 0.8 to 1 is the sweet spot.
📝 Example prompt structure to trigger the Lora
Opening: A dark fantasy illustration, masterful airbrush painting, chiaroscuro lighting. A muted desaturated color palette with steel blues and greys. Intricate textures of tattered leather and rusted metal contrasting with smooth pale skin, melancholic and solemn atmosphere
or
A post-apocalyptic sci-fi illustration, masterful airbrush painting, dramatic atmospheric lighting. Highly detailed biomechanical elements, dark cables, intricate metal armor, dark grey and earthy tones
or
A close-up portrait illustration, traditional painting mixed with soft airbrush. Pale skin, highly detailed eyes and lips, delicate gothic lace, soft glowing light emerging from deep shadows, melancholic expression
📝 Example of a prompt structure to trigger the Lora (edition mode)
dark fantasy digital illustration based on the provided input image, masterful airbrush painting style.
The main subject features rich colors, slightly muted skin tones, and striking colors that pop brilliantly against the dark environment. Every surface features masterful airbrush painting textures and visible brushwork.
The background is transformed into an atmospheric, desaturated wash of deep shadows, cold steel blues, and floating misty vapor.
Dramatic chiaroscuro lighting, emphasizing high-contrast glowing highlights on the colorful subject against the pitch-black and midnight blue shadows
Edition example:
🛠️ Behind the Scenes
This LoRA is highly flexible because it wasn't trained on generic tags. It was trained on 235 manually curated, high-quality images using an uncensored, and custom captioning workflow. Every single image was analyzed by a local Qwen 3.5 27B abliterated Model to extract exact anatomical details, texture vocabulary, and lighting setups
Disclaimer: This LoRA inherits the non-commercial license of the base FLUX.2 model. It is created strictly for educational, research, and personal use. Please respect the original artist's legacy and do not use this model to generate commercial products.
Description
Style LoRA for Anima that reproduces the dark fantasy and sci-fi illustration look of Luis Royo: detailed airbrushed rendering, sensual female figures, ornate armor, chrome, leather straps and chains, moody chiaroscuro lighting and apocalyptic scenery. This is my first LoRA trained on Anima, and the first one where I kept a checkpoint I am actually happy with.
Trigger word
lu1sRoy0 - put it at the start of your prompt. It is the only tag you need to activate the style.
What it does
Detailed, airbrushed rendering with smooth skin gradients
Sensual, powerful female figures, often armored or chained
Intricate ornament: dark metal armor, chrome, leather straps, spikes, mechanical wings
Moody chiaroscuro lighting, glowing accents, storm and wasteland skies
Muted palettes with hot accents (deep reds, oranges, glowing cyan)
Tested subjects that come out well: warrior women, demon and succubus characters, cybernetic swordswomen, dragons and mechanical beasts, chained or reclining nudes, ruined gothic architecture, post-apocalyptic landscapes.
Recommended settings
Checkpoint: 5500 steps is the published one. 4500 and 5000 also work, slightly weaker stylistically.
LoRA strength: 1.0 (lower it to around 0.7 if the style takes over your prompt).
Base model: Anima. The samples were generated with Anima-Aesthetic v1.1.
Sampler: Euler.
Scheduler: Simple.
Steps: 30.
CFG: 4.
Resolutions: 1024x1536, 1536x1024, 1536x1536 and 904x1600 all work fine.
Negative prompt used:
worst quality, low quality, score_1, score_2, score_3, blurry, jpeg artifacts, sepia
Helpful prompt companions: score_9, best quality, dark fantasy, oil painting, chiaroscuro, detailed airbrushed rendering, plus your subject (armor, dragon, monster, cathedral, wasteland).
What it was trained on
758 images from Luis Royo artbooks and web scans, catalogued with natural language captions plus Danbooru tags
Base model: Anima-Base v1.0 (about 2B parameters, DiT flow-matching). The Qwen3-0.6B text encoder was frozen, only the DiT was trained.
LoRA rank 64, alpha 64, batch size 4, 6000 steps with a checkpoint every 250
AdamW8bit, learning rate 2e-5, flowmatch scheduler, bf16
Roughly 2.9 epochs at step 5500, about 6h30 of training on a single RTX 3090 24 GB with ai-toolkit
Notes and known issues
Step 6000 was rejected on purpose: past 5500 it starts producing body-horror artifacts - mangled anatomy, extra and distorted limbs. That is why the published checkpoint is 5500. Everything up to it is stable, everything after it degrades.
It is a style LoRA: it applies a look, it does not know any specific character. No specific subject is required in the prompt.
The style is sensual by nature, and it also handles violent and very dark subjects well. Age-gate accordingly.
In wide shots, faces at low resolution can look airbrushed and soft. That is a trait of the source artworks, not a bug.
About
This is a student project and my first LoRA for Anima, trained on 758 illustrations by Luis Royo, the Spanish illustrator known for his dark fantasy and science-fiction art of powerful, sensual women. It is not perfect, but it is a lot of fun to work with.
The LoRA was trained on the artist's work for study and personal use. Please respect the original artist and do not use it to reproduce his actual paintings commercially.