Negative prompt:(worst quality, low quality, illustration, 3d, 2d, painting, cartoons, sketch), tooth, open mouth,bad hand,bad fingers
The official v1 has been launched, and there is still a lot of room for improvement. The performance in terms of light and shadow is unstable, and we hope this can be resolved in the next version. Other parameters can be referred to in the figure legend.
v1正式版本已上线，目前依旧有很大的提升空间，光影方面表现不稳定，希望下个版本可以解决，其他参数可以参考图例。
-------------------------------------------------------------------------------------------
The official version is expected to be released within the next couple of days, and I hope it won't disappoint anyone.
正式版应该会在这两天与大家见面，希望不会让大家失望。
When generating images with dimensions smaller than 1024*1024, if you click on the "Hires" or "Adetailer" option for the image, you can use the SD1.5 version of the "Hires" or "Adetailer" feature. However, you need to select a magnification factor of 1.5x to prevent excessive distortion on the character's face.
当你生成图片的时候尺寸是小于1024*1024的时候，你点击图片的Hires or Adetailer，是可以使用SD1.5的Hires or Adetailer，但是放大倍数要选择1.5倍，这样可以稍微让角色的脸不那么崩。
Negative prompt(very important):(worst quality, low quality, illustration, 3d, 2d, painting, cartoons, sketch),
This model is not quite suitable for image quality prompts, as the results may backfire
Trigger Words please！！！！！！！
Test_v2
---Please describe in natural language.
---For example: xxmix_girl, a woman with long hair, wearing a white shirt, and a pair of earrings on her neck and shoulder, against a white background.
---Standard Start: xxmix_girl, a woman ____________________________
---At present, the model is still recommended for portraits or half-length images. The skin texture and delicacy have improved a lot compared to the Testv1 version. However, I haven't tested many application scenarios, so the effect of more flamboyant text may not be very good. This model is more suitable for portrait photos of Asian faces, as most of the training set is collected from magazines or flat photography.
---Some of the examples are images produced by webui, using Hires and the adetailer extension plugin. Some are images produced online by tensorart, so there may be some loss of effect in the online version. ---This version has a more mixed-race and mature woman's face shape. If you don't like this face, you can add my lora at https://tensor.art/models/620520927983161041 for face shape optimization. However, it is not recommended to set the value too high, as it may reduce the detail and texture of the skin.
---I hope the official version can meet with you soon. Thank you!
1.请用自然语言进行描述。
举个例子:xxmix_girl,a woman with long hair wearing a white shirt and a pair of earrings on her neck and shoulder, with a white background,
起手式：xxmix_girl,a woman ____________________________
2.目前模型还是推荐肖像或半身像，皮肤质感和细腻度要比testv1版本提高很多。不过我没有测试太多应用场景，可能比较花里胡哨的文字效果不一定很好，这个模型的亚洲脸比较适合肖像照片。因为大多数的训练集是采集的杂志或者平面摄影图。
3.图例有一些是webui出的图，有用到高清放大和adetailer扩展插件，有一些是tensorart在线出的图，可能在线版会有一些效果损耗。
4.这一版的脸型比较混血御姐一些，如果你不喜欢这个脸的话，可以加上https://tensor.art/models/620520927983161041 我的lora进行脸型优化，不过不建议数值开太高，开太高可能会减弱皮肤的细节质感。 4.希望正式版能够很快与大家见面，感谢！
-------------------------------------------------------------------------------------------
Test_v1
Please read the text description carefully first
Please pair with this Lora for better results
https://tensor.art/models/620520927983161041
At present, the characters have a strong sense of distortion, and the skin texture details are not clear enough. I hope to address these issues through continuous iterations.
XXMix_9realisticSDXL is a fine-tuned model based on the Stable Diffusion XL model, aiming to improve the poor performance of Stable Diffusion XL in terms of the attractiveness of Asian female characters. The current v1 version is still in the experimental stage, with many issues, but I will continue to iterate and optimize.
Trigger Words: xxmix girl woman (in some cases, it may be necessary to increase the weight of keywords)
The known issues include:

The character name or face style change may not work in some cases. In this situation, you can try to increase the weight of the trigger words.
Since the model is trained based on a 10241024 canvas size, other image ratios may result in a higher error rate. It is recommended to use a 10241024 canvas size as much as possible, and I will try to train with different canvas sizes in subsequent iterations.
Regarding reverse prompt words: The reverse prompt words we currently use can maintain good performance of the model in terms of Asian facial features. If you try to use different reverse prompt words, the results may not be as good as the ones I am currently using. However, if you find more suitable reverse prompt words for this large model, I hope you can share them with everyone.
Due to the aspect ratio of the images, this model is currently more suitable for generating half-body portraits or close-up portraits. If you need to generate full-body images, it is recommended to enable the Hires or use the faterdetail extension plugin.

目前角色失真感较强，皮肤纹理细节不够明显，我希望能在不断地迭代中改掉这些问题。
XXMix_9realisticSDXL是一个基于Stable Diffusion XL模型训练的微调模型，旨在提高Stable Diffusion XL在亚洲女性角色颜值方面糟糕的表现。目前的v1版本仍处于试验阶段，存在许多问题，但我会继续进行迭代优化。
触发词Trigger Words：xxmix girl woman（在某些情况下，可能需要增加关键词的权重）

目前已知的问题包括：

1.出现人物名字或者画风该脸型有几率不生效，这时，可以尝试增加触发词的权重。

2.由于模型是基于1024*1024的画幅进行训练的，因此其他比例的图片可能会导致较高的错误率。建议尽量使用1024*1024的画幅，我会在后续迭代中尝试使用不同画幅的训练集进行训练。

3.关于反向提示词：目前我们使用的反向提示词能够保持模型在亚洲脸部效果方面的良好表现。如果您尝试使用不同的反向提示词，效果可能不如我目前所使用的提示词。但是，如果您找到了更适合这个大型模型的反向提示词，我希望您能与大家分享。

4.由于画面比例的原因，目前这个模型更适合生成半身像或肖像。如果需要生成全身像，建议开启高清修复功能或使用faterdetail扩展插件。

XXMix_9realisticSDXLは、Stable Diffusion XLモデルをベースにした微調整モデルで、Stable Diffusion XLのアジア女性キャラクターの顔の魅力に関する悪いパフォーマンスを改善することを目的としています。現在のv1バージョンはまだ実験段階であり、多くの問題がありますが、私は引き続き反復最適化を行います。
トリガーワード：xxmix girl woman（場合によっては、キーワードの重みを増やす必要があります）

現在の既知の問題は以下の通りです。

1.キャラクター名や顔のスタイルの変更が効かない場合があります。この場合、トリガーワードの重みを増やしてみてください。

2.モデルは10241024のキャンバスサイズで訓練されているため、他の画像比率ではエラー率が高くなる可能性があります。できるだけ10241024のキャンバスサイズを使用することをお勧めします。私は後続の反復で異なるキャンバスサイズのトレーニングセットを使用して訓練を試みます。

3.逆プロンプトワードについて：現在使用している逆プロンプトワードは、モデルがアジア人の顔の特徴に関する良好なパフォーマンスを維持できるようにしています。異なる逆プロンプトワードを試してみると、私が現在使用しているプロンプトワードほど効果的でないかもしれません。ただし、この大型モデルに適した逆プロンプトワードが見つかった場合は、皆さんと共有していただけると嬉しいです。

4.画面の比率のため、現在のモデルは半身像やポートレートの生成に適しています。全身像を生成する必要がある場合は、高解像度修復機能を有効にするか、faterdetail拡張プラグインを使用することをお勧めします。

XXMix_9realisticSDXL은 Stable Diffusion XL 모델을 기반으로 한 미세 조정 모델로, Stable Diffusion XL의 아시아 여성 캐릭터 얼굴 가치 측면에서 나쁜 성능을 개선하려고 합니다. 현재 v1 버전은 여전히 실험 단계에 있으며 많은 문제가 있지만, 계속 반복 최적화를 진행할 것입니다.
트리거 워드: xxmix girl woman (일부 경우에는 키워드 가중치를 늘릴 필요가 있을 수 있습니다)
현재 알려진 문제는 다음과 같습니다.

1.캐릭터 이름이나 그림 스타일이 얼굴 형태에 적용되지 않을 확률이 있습니다. 이때 트리거 워드의 가중치를 늘려보십시오.

2.모델은 1024*1024 크기의 캔버스를 기반으로 훈련되었기 때문에 다른 비율의 이미지는 높은 오류율을 초래할 수 있습니다. 가능한 한 1024*1024 크기의 캔버스를 사용하는 것이 좋으며, 후속 반복에서 다른 크기의 캔버스 훈련 세트를 사용하여 훈련을 시도할 것입니다.

3.역방향 프롬프트 단어에 관해서: 현재 사용 중인 역방향 프롬프트 단어는 모델이 아시아 얼굴 효과 측면에서 좋은 성능을 유지할 수 있도록 합니다. 다른 역방향 프롬프트 단어를 사용하려고 하면, 현재 사용 중인 프롬프트 단어만큼효과적이지 않을 수 있습니다. 그러나 이 대형 모델에 더 적합한 역방향 프롬프트 단어를 찾으신다면, 다른 사람들과 공유해 주시면 감사하겠습니다.

4. 화면 비율 때문에 현재 이 모델은 반신상이나 초상화 생성에 더 적합합니다. 전신상을 생성해야 하는 경우, 고해상도 복원 기능을 활성화하거나 faterdetail 확장 플러그인을 사용하는 것이 좋습니다.

XXMix_9realisticSDXL - v1

XXMix_9realisticSDXL - v1

Sampler

Steps

CFG Scale

Seed

Clip Skip

Size