Moescape Diffusion is an advanced tag-based AI image generation model specialized in creating high-quality anime-style artwork. Moescape Diffusion 1.0, its latest flagship version, is built on the robust SDXL architecture with a massive dataset of 8M diverse anime-style images and trained for approximately 2,500 GPU hours with an extraordinary exposure to over 80M training images, this exclusive model implements well-known tag ordering systems to deliver exceptional quality and control in anime art generation.
The model offers two specialized variants:
- Full Version: designed for power users requiring granular control and flexible outputs
- Aesthetic(sft) Version: streamlined for end users seeking quick, beautiful generations
Description
Recommended settings:
- For optimal results, it's recommended to follow the structured prompt template: `1girl/1boy, character name, from which series, by which artists, everything else in any order.`
- Put `masterpiece, best quality, very aesthetic, absurdres, safe` at the end of the prompt.
- Sampler steps should be around 25 to 30; 28 is the sweet spot.
- CFG scale should be around 5 to 7; 10 is fried, >12 is deep-fried.
- Euler Ancestral is highly recommended.
Quality tags: masterpiece, best quality, great quality, normal quality, low quality, worst quality
Rating tags: safe, sensitive, nsfw, explicit
Year tags: year 2020, year 2021, ... year n, etc
Aesthetic tags: very aesthetic, aesthetic, displeasing, very displeasing