MeMaMoss Flat Anime Style
LoRA for anime magazine art generation
2 min read
A new LoRA (a small add-on for a base model) for image generation based on the MOSS model, recreating the recognizable flat anime aesthetic of Japanese Megami Magazine publications. The model's dataset was manually cleaned of extraneous Japanese text, and images were carefully curated to achieve accurate color reproduction, characteristic line art styling, and detailed eye rendering from the original print editions. This model is useful for digital artists, anime art enthusiasts, and anyone experimenting with high-resolution image generation.
Key features
- Distinctive Megami Magazine anime style from the 2018 to 2026 period.
- Trained on high-resolution images (up to 2048 pixels) for clean detail.
- Version V3 is adapted for the MOSS Image SFT base model for maximum accuracy.
- Meticulously cleaned dataset free of stray text and artifacts.
What you can do
- Generate high-quality 2D anime illustrations with flat fills in high resolution (around 1600×1984 recommended).
- Configure generation with the proper prompt, using minimal tags for the positive prompt (such as high quality) and the author-specified negative list.
Example
Recommended prompt for generation:
Positive: high quality
Negative: best quality, normal quality, low quality, worst quality, photorealistic, realistic, 3d, film grain, bad anatomy, watermark, jpeg artifact, bloom, blurry, blurry background, depth of field, extra legs, too many fingers, too many toes, thick lineart, painterly
Read more
DetailsBase model: MOSS Image / MOSS SFT
Training resolution: Up to 2048×2048 (V1 — 1792×1792)
Dataset size: 68 images (for version V3)
Recommendation: Use with the Moss SFT model and configure Skip Layer Guidance (values 22, 26).










Comments