NovelAI Diffusion Anime V2

Old and New

Comparisons between NAID 1.0 vs NAID 2.0

Get an impression between our old model and #NAIDiffusionV2 , here are some comparison images. They were generated on the same seed with mostly the same prompts (note: quality tags were changed, depending on model)

https://pbs.twimg.com/media/F857TARX0AEWA9Y.jpg https://pbs.twimg.com/media/F857TB2XcAASNJx.jpg

https://pbs.twimg.com/media/F857TDsW8AAisu4.jpg https://pbs.twimg.com/media/F857TF6XYAAtZA_.jpg

Stable Diffusion 1.5のファインチューニング

トレーニング解像度を512x716→1024x1024に

SMEAを使わない基本的な解像度は832x1216に

Undesired Content Strength

クオリティタグの更新

"masterpiece"は無くなり

best quality

amazing quality

great quality

normal quality

bad quality

worst quality

Aesthetics Tags(美的タグ)

very aesthetic

aesthetic

displeasing

very displeasing

年代タグ

year 20**

NovelAI Diffusion Anime V3を開発中

Clip skipの発明

2025/7/8

https://huggingface.co/NovelAI/nai-anime-v2NovelAI/nai-anime-v2

ライセンス

Creative ML OpenRAIL-M & CC BY-NC-SA 4.0

←　NovelAI Diffusion V1