ver1 updates: improved the stability when generating multiple characters __________________________________________________________________________________________ 生成多人图时推荐使用自然语言而不是tag来描述角色特征。step5800的泛化性比较好，step7000有些过拟合不过可以更还原角色特征。 複数キャラを生成する際は、タグよりも自然文でキャラクターの特徴を記述することを推奨します。5800steps版は比較的汎用性が高く、さまざまな構図やシーンで扱いやすいです。7000steps版は少し過学習気味ですが、その代わりキャラクターらしさをより忠実に再現できます。 When generating images with multiple characters, it is recommended to describe each character's features using natural language rather than tags. Model trained for 5800 steps has better generalization and tends to work well across a wider range of compositions and scenarios. Model trained with 7000 steps is somewhat overfitted, but it can reproduce the characters' distinctive features more faithfully.

Files with SHA256: bd742d48b88111055fdc964619397a68bf4d50c4f6bebd24af6e5dd9b574d1be