🌲 Surprising Success

I spent countless hours training - often hitting dead ends - but every dead end taught me more about the process. That accumulated know-how is what finally turned into the success I wanted.

All images were generated with no LoRA, no upscaling and no detailer. They are posted exactly as generated. Most of them are 1920×1280 or 1280×1920 or 1280x1280. I didn’t touch the output.

I still recommend using smaller resolutions, as larger resolutions may produce some artifacts.

🧠 About the Model

You can prompt freely in full sentences or use tags - the model handles both smoothly.
It interprets emotional tones and scene composition quite naturally, so feel free to experiment.

This model is based on Pony, but you don’t need to use Pony-style prompt syntax. That said, Pony-based LoRAs tend to work best.

• Resolution: 1280×1280 works beautifully for capturing details.

• Sampler: Euler ancestral (or multires-multistep ancestral for bigger resolution)

• Scheduler: Karras

• Denoise: 0.8

• CFG: around 9 (depending on prompt style and intensity)

• Steps: 30 (even 15 can already produce great results)

⚠️ NSFW Notice

This model is NSFW-capable, so use it responsibly.
It’s meant for creative exploration, not explicit output - please be mindful of the content you generate.

NFSW Tools Ecosystem

• AI LoraScope – inspect LoRA training metadata and keyword frequencies.

• AI RapidTagger – manage tags, build prompts, and organize datasets.

• AI RenderScope – explore, analyze, and filter metadata from generated images.

• Prompt Builder (coming soon) – a streamlined tool for composing and testing prompts interactively.