Type | |
Stats | 352 |
Reviews | (38) |
Published | Feb 2, 2024 |
Base Model | |
Training | Steps: 876,640 Epochs: 50 |
Usage Tips | Clip Skip: 1 |
Hash | AutoV2 950F174E9F |
The v2 fix version has iterated the performance of characters, making it more inclined to 2.6D realistic characters, while ➡️ v1 is more minimalist, each with its own style!
You can try to compare, the following are some of my ramblings, you can skip~
lbc_Rebirth means the concept of rebirth. For this model, I have experienced a lot…
The most notable is the bottleneck of hardware computing power (yes, I am poor…), limited by a device equipped with only a GTX4060TI16GB graphics card, the calculation speed of the training process is obviously limited.
However, it is these difficulties that make me cherish every failure more, and also make me handle my dataset more carefully, because I know that the decisive factor of a model’s good or bad is the dataset in it. Of course, my model is also an inheritance of some excellent datasets in the lbc series (SD1.5 version), covering various aspects such as detail performance, artistic photography, minimalist two-dimensional style, and fantasy realism. Such a wide range of explorations makes lbc_Rebirth a diverse and comprehensive model.
“Rebirth” means rebirth, implying the bits and pieces and storms I have experienced in the process of model creation. By overcoming all difficulties, I have given this model a new life, a more powerful and comprehensive expressiveness.
Instructions for use:
Clipskip: 1-2, I usually use 1
Sampling method (Sampler) : Euler a
Samplingsteps: 30+
You can turn on HD repair, the recommended algorithm is: 4x-UltraSharp
Width: 640 or others are fine
Height: 960 or others are fine
CFGScale: 7
If you like my model, please support it a lot.
如果你有更好的建议请在评论区告诉我,或许你希望加入➡️我的TG聊天频道