Trained on vanilla SD1.5 but works better with art-focused checkpoints like Elldreth's and others from this website. Training dataset contained game portraits, key art, and some cherrypicked fanart.
How to consistently get very stylized results:
-
Use Elldreth's lucid mix
-
Use "[{subject}:abstract background:0.5], expressionism, digital art, colorful background" prompt, replace {subject} with anything
-
Parameters: 20 steps, CFG=2, DPM++ 2M Karras sampler
-
Some usual negative prompts
See my comment below for examples of this method
描述:
训练词语:
名称: discoElysiumStyle_discoElysium.pt
大小 (KB): 85714
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: **Detected Pickle imports (3)**
```
torch._utils._rebuild_tensor_v2
collections.OrderedDict
torch.FloatStorage
```
病毒扫描结果: Success
