Creation Checkpoint V1

中文/English

简介

预览版本。使用LoRA时，需要适当降低权重。

XL模型在训练当中。

✅ 提升线条利落性

• 相比前一版本，线条准确性得到提升，画面看起来更加干净利落。

✅ 提升手部拟合效果

• 生成图的人物手部“AI感”得到一定改善，姿势更加真实。

✅ 提升词义准确性

• 清理部分垃圾数据，降低生成失误率，提升生成准确性。

✅ 提升画面整体质量

• 改进了作画构图，降低“AI生成”感，更加贴合真人画师实际构图。

✅ 改善模型画风，提高画风兼容性

• 一定程度上消去了上一版本的画风“胶质感”“3D感”，但人物仍存在骨感。

下一步：

1. 进一步提升词义准确性，提高画面整体质量。

2. 改善人物裸体骨感，提高人物丰满度。

3. 改善头发线条末梢的拟合效果。

4. 解决光影问题。

5. 改善手部生成质量。

Introduction

Preview release. When using LoRA, you need to lower the weight appropriately.

✅ Improved line fluency

Compared to the previous version, the accuracy of lines has been improved, making the images look cleaner and more neat.

✅ Enhanced hand fitting effect

The "AI feel" of the hands in the generated images has been somewhat improved, making the poses more realistic.

✅ Improved semantic accuracy

Cleaning up some garbage data, reducing generation errors, and improving overall accuracy.

✅ Enhanced overall image quality

Improved drawing composition to reduce the "AI-generated" feel and make it more in line with the actual composition of human artists.

✅ Improved model style and compatibility

To some extent, the previous version's "gelatinous" and "3D" styles have been eliminated, but the characters still maintain a skeletal appearance.

Next steps:

1. Further improve semantic accuracy and enhance overall image quality.

2. Improve the naked skeletal appearance of characters and increase their fullness.

3. Enhance the fitting effect of hairline ends.

4. Address lighting and shadow issues.

5. Improve the quality of hand generation.

Style type: 2D

Model category: Checkpoint

Model format: Safetensors

Model size: 3.97GB

Embedded VAE model: Yes

MD5: 896566f31f7e774c18d9bd560fe76de1

SHA256: bc616239ac4609bc6af2ab8586948dfc2bf0b05119495985a7cd4869dcd55ffb