Usability of Animagine XL V3 and comparison with other models
2024-01-16 01:47:37
CustomModel(その他)
animagine-xl-3.0
対象年齢：全年齢
デイリー入賞： 6 位
ウィークリー入賞： 120 位
参加お題：サイクルロードレース
It seemed like a good subject for comparison, so here are my personal impressions after touching Animagine XL V3, which has been the talk of the town
Conclusion
This is an ambitious and interesting model that aims to combine the stability and accuracy of the SDXL series with the expressive power of the SD1.5 animation series.
In the SDXL series, there is a feeling that this model can only produce the kind of expression that is currently possible.
Still some parts are too peaky, so we look forward to future version upgrades and the evolution of merged models based on this model.
＝＝＝＝＝＝＝＝Details that do not need to be read separately below: ＝＝＝＝＝＝＝＝
Features
A model specialized for animation, which is rare in the SDXL series.
The model itself has already learned animation characters and concepts.
(The model was additionally trained for 21 days with 2x A100 GPU and 80GB memory)
Strengths
Can produce anime and cartoon-like expressions that are difficult to achieve with other SDXL models.
Strongly suggestive of styles and effects.
It is easy to produce powerful and unexpected pictures among SDXL models.
⇒In this example, motion blur and wind effects can be created strongly.
Weaknesses
Stability, accuracy of structure, and lack of limb breakdowns are inferior to other SDXL series (but still superior to SD1.5).
Too high prompt sensitivity, so the picture tends to be broken when strong words are added.
It seems to sacrifice natural language processing more than other SDXL series, and the recommended prompt input method is unique (for the purpose of comparison, we ignored most of the recommended values).
⇒In this example, there are more broken limbs and frames than other SDXL models.
List of models to be compared with the same settings
1st : AnimagineXL_v3
2nd : AnimagineXL_v2 (the base model of AnimagineXL_v3)
Third: animaPencilXL_v100 (merged model using AnimagineXL_v3)
Fourth: bluePencilXL_v310 (merged source model of ↑)
5th: aikimixl_v1 (model of another SDXL system)
6th: brightprotonukeBPNNo_bpn13 (Model from a different SDXL lineage)
7th: flatpiececorexl_a1579 (SDXL model of a different lineage)
8th: himawarimix_xlV2 (SDXL model of a different lineage)
9th: fiamixReboot_v90 (SD1.5 series model)
*The above is my personal impression based on a little experimentation, so there may be some mistakes. For details, please refer to the official distribution page with related links, etc.
類似プロンプト作品検索
ログインするとプロンプトなどがチェックできます
※ 作品によっては掲載されていないことがあります
(masterpiece, best quality:1.3), illustration, (extream detailed), one girl, angel Descending
関連リンク
コメント
投稿
ダウンロード
ダウンロード
ダウンロード
ダウンロード
ダウンロード
ダウンロード
ダウンロード
ダウンロード
検証助かります
ダウンロード
新しい玩具に嬉々として飛びつきましたw
ダウンロード
私も何日か使ってみて、破綻も多いけれど良く動いてくれて面白いなと思いました。
ダウンロード
RefinerやHiresで真面目系のSDXLモデルと組ませてみるのも面白そうです
ダウンロード
ダウンロード
ダウンロード
これはナイス調査
ダウンロード
複数のSDXLモデル利用しているユーザをほぼ見かけないので、ちょっと比較してみようかなと思い
ダウンロード
ダウンロード
ダウンロード
ダウンロード
ダウンロード
ダウンロード
ダウンロード