mq_sdxl版本mq_sdxl_64_64_v2 (ID: 654462)

mq_sdxl_64_64_v2

By removing and adding some tags, the generated results are improved and more realistic (I think so). Training parameters are kept the same as mq_sdxl_64_64.

---

mq_sdxl_64_64

The improved version of LoRA training on SD XL (v1.0 VAE fix). The training tags are specially examined and modified. Training details are shown below.

• Network Rank (Dimension): 64

• Network Alpha: 64

• Training data: 11 pictures with 1024×1024 resolution

• max_train_steps: 1600

• Number of Repeats: 30

The goal is to generate an Asian woman with a round face.

Suggested checkpoint: SD XL

Suggested sampler: Euler a

---

mq_sdxl_64_32

The first version of LoRA training on SD XL (v1.0 VAE fix). The tags of training pictures are simpler and less than the newer versions. Training details are shown below.

• Network Rank (Dimension): 64

• Network Alpha: 32

• Training data: 11 pictures with 1024×1024 resolution

• max_train_steps: 1600

• Number of Repeats: 30

The goal is to generate an Asian woman with a round face.

Suggested checkpoint: SD XL

Suggested sampler: Euler a

描述:

An improved version, the generated face is more realistic.

训练词语: 1girl,mq