🌧️ The Monsoon doesn’t strike — it seeps in, softly and relentlessly.
Monsoon is the latest entry in my "storm" ecosystem (joining Typhoon and Tornado), but unlike its turbulent siblings, Monsoon isn’t here to smash through prompts. It drifts in — slow, stylized, and quietly overwhelming.
This is version 0.1, an early but promising stage. While checkpoint training and a few LoRA merges were block-weighted into the base, the model remains almost raw. It hasn’t been overfit, distilled, or polished — and it’s nowhere near full capability yet, which is exactly why it’s not v1.0. There’s still plenty it can’t do, but what it does do? Stunning.
My personal target? Stylized ethereal portraits and scenery, and it’s already exceeding expectations there.
💡 A direct homage to Lumina Image 2.0
At its core, Monsoon is a love letter to Lumina Image 2.0 — one of the most beautiful, flow-based transformer models out there. Lumina is known for its soft cinematic tone, artistic depth, and painterly control over light. Monsoon borrows heavily from that aesthetic and transplants it into the SDXL diffusion transformer space.
The result? A model that often feels like Lumina, while keeping the freedom, openness, and tooling compatibility of SDXL.
🌅 What is Monsoon good at?
Stylized landscape shots
Dreamlike seasonal art (spring blossoms, winter haze, soft sunsets)
Painterly portraits with glow and ambience
Emotional, cinematic lighting
Strong silhouette and shape language
🔍 What is it not good at?
Hyperrealism — not the goal here
Clean photographic detail
Fast results on the wrong scheduler
Hands, like every SDXL base ever 😅
⚙️ Sampler Suggestions
These are the samplers Monsoon responds best to, based on extensive testing:
DPM++ 2M — Clean and detailed, works great right out the gate
DPM++ 2M SDE — Needs more steps, but rewards with a soft, painterly glow
Euler A — Smooth, creamy, expressive — especially at lower step counts
Euler — Similar to Euler A, but slightly sharper and more defined
DPM Adaptive — Slow, but ultra-refined and moody — worth the extra time
DDIM — Surprisingly balanced; great for stylized and consistent output
DDPM — Dreamy and soft, like Euler A but with a floatier finish
This model was tested head-to-head against Lumina Image 2.0, using matched prompts and seeds where possible. While Lumina relies on flow-based schedulers (like res_multistep) due to its transformer-flow architecture, Monsoon (being SDXL-based) uses diffusion. So, the outputs won't be identical — but vibes? Shockingly close.
📌 Notes
This is v0.1 — it will evolve. Some images will blow your mind. Others might leave you tweaking. That’s fine. It’s a Monsoon — not a lightning strike.
Test it, bend it, break it. I'll update the model as the season rolls on.
📜 Usage Disclaimer:
This model is strictly for personal, non-commercial use only.
You are not allowed to repost, sell, merge, or otherwise profit from this model in any form.
It is hosted on CivitAI for personal use only. Unauthorized redistribution or monetization is prohibited.
🌩️ ☕ Support the Storm
If you like what I’m building — Typhoon, Tornado, the tools, the chaos — and want to help keep it all spinning, consider supporting me on Ko-fi:
https://ko-fi.com/raxephion
Every bit helps cover compute costs, caffeine, and the occasional “why is this broken?” meltdown.
Thanks for keeping the storm alive.