This is a Krea 2 LoRA trained on 150 high-quality splash art illustrations (16:9 landscape).
🎨 Trigger Word:
Add lorst at the end of your prompt to activate the style.
⚙️ Recommended Settings:
- Base Model: Krea 2
- Sampler: Euler
- Scheduler: Simple
- Steps: 10 - 12
- CFG: 1.0
- Recommended Weight: 0.5 - 0.8 (adjust to your preference)
🖼️ What it does best:
This LoRA captures the highly detailed, cinematic, and dramatic lighting of modern game splash arts. It excels at epic compositions, vibrant color contrast (especially warm and cool tones), and rich textures for characters and environments. It is perfect for generating fantasy warriors, magical scenes, and dynamic wide-angle shots.
⚠️ Notes:
This is a pure style LoRA. It does NOT lock you into specific characters. You can use it with any character prompt, and it will apply the splash art style to them. For best results, keep your prompts in natural language (full sentences). Avoid writing "league of legends splash art" in your prompt, as the trigger word lorst takes care of that.
Negative Prompt: (Optional for Krea 2)
blurry, lowres, bad anatomy, bad hands, text, watermark, ui, jpeg artifacts.
Description
Title: Lorst V3 Update: More Diverse Data, Simpler Prompts, and Better Usability
Welcome to the latest (and likely final) major update of the Lorst style LoRA!
✨ What's New in V3?
200 Carefully Curated Images: I re-filtered the dataset, adding types of images that were missing in the previous two versions, specifically focusing on elderly characters and various non-humanoid creatures.
Experience-Driven Optimization: By summarizing my previous training experiences and discussing them with AI, I identified key issues from past runs and made targeted corrections.
Focus on Dataset Diversity & Prompt Optimization: This version is all about balancing usability and effectiveness.
🚀 The Biggest Breakthrough: Usability
In the first two versions, you often needed highly specific, detailed prompts to trigger the style. Through testing, I realized that short prompts frequently failed to activate the LoRA's style.
In V3, I heavily optimized for this. Now, you can achieve the distinct Lorst style with much simpler, more natural prompts!
🔍 An Important Finding for Users (Please Read)
During my final testing, I noticed that while the style is highly versatile, it has a strong inherent bias: it leans heavily towards "Fantasy" themes.
If your prompt is fantasy-related (e.g., warriors, mages, dragons, ancient ruins), the style will activate beautifully, regardless of the character or creature type.
However, if you try to use it for highly modern, everyday life scenes, the style might struggle to manifest.
Recommendation: For the best results, keep your generations within a fantasy or epic sci-fi context.
💡 Pro Tip: The "Overdrive" Secret
If you find the style isn't triggering as expected (especially in difficult scenes), try cranking the LoRA weight up to 2.0. It forces the model into a heavy stylized mode, giving you a surprising and dramatic brushstroke effect! (Note: It may make the strokes heavier, but the style will definitely pop!)
📌 Final Thoughts & Future Plans
This will likely be the final update for this specific style LoRA, as I feel it has reached a great state of balance. I am already exploring new directions and training new types of LoRAs.
Thank you all for your support and feedback! Stay tuned for my next project!