Kohaku-XL-εpsilon | Rev2

CHECKPOINT
Original


Updated:

Kohaku XL epsilon rev2

Kohaku XL Epsilon的改進更新版。

加入Q****: 881970728

Rev2版本特性

  • 從Kohaku XL Epsilon Rev2加練而來

  • 154萬圖片,訓練5epoch。

  • 訓練集使用Pixiv高人氣畫師以及熱門IP為主。

  • PVC手辦也被加入訓練集,無須加練即可生成手辦圖片。

使用方法(務必閱讀本節)

  • Prompt格式: "<1girl/1boy/1other/...>, <character>, <series>, <artists>, <general tags>, <quality tags>, <year tags>, <meta tags>, <rating tags>"

  • 品質Tag: masterpiece, best quality, great quality, good quality, normal quality, low quality, worst quality

  • 分級Tag: safe, sensitive, nsfw, explicit

  • 年份Tag: newest, recent, mid, early, old

  • 解析度: 1024x1024,或接近的像素量

訓練參數與資訊

  • 使用硬體: 四張 RTX 3090

  • 訓練集圖片總數(含Repeat): 1,536,902

  • 總Epoch: 5

    • 總步數: 15015

    • 訓練耗時: 410小時(現實時間)

    • Batch Size: 4(每張卡)

    • Grad Accumulation Step: 32

    • 等效Batch Size: 512

  • Optimizer: Lion8bit

    • 學習率: 1e-5 for UNet / 2e-6 for TE

    • LR Scheduler: Cosine (with warmup)

    • Warmup Steps: 1000

    • Weight Decay: 0.1

    • Betas: 0.9, 0.95

  • Min SNR Gamma: 5

  • Noise Offset: 0.0357

  • ARB解析度: 1024x1024

  • Min Bucket Resolution: 256

  • Max Bucket Resolution: 4096

  • 混和精度: FP16

  • Caption Tag Dropout: 0.2

  • Caption Dropout: 0.05

The model already exists. If you are the creator, please contact us to claim it.
View duplicate model

Version Detail

SDXL 1.0
15015
5

Project Permissions

    Use Permissions

  • Use in TENSOR Online

  • As a online training base model on TENSOR

  • Use without crediting me

  • Share merges of this model

  • Use different permissions on merges

    Commercial Use

  • Sell generated contents

  • Use on generation services

  • Sell this model or merges

Comments

Related Posts

No posts yet