Type | |
Stats | 1,021 1,656 |
Reviews | (148) |
Published | Oct 1, 2023 |
Base Model | |
Training | Steps: 22,000 Epochs: 50 |
Trigger Words | street,taiwan roads,taiwan |
Hash | AutoV2 4386DB6D14 |
中文簡介在最下面
Here's the translation of your text into English:
"This is a LoRa (Long-Range Radio) trained using Taiwan streets.
Taiwan streets are characterized by right-hand driving, and you can easily spot air conditioners, advertising signs, utility poles, billboards, cars, pedestrians, old buildings, and corrugated metal houses on the roads.
During LoRa training, no specific central character is included, but low overfitting is applied during training. This makes it convenient to combine various characters and other elements during training, but it may result in less fitting during the final output.
Therefore, it is recommended to generate images twice (use Hires.fix during t2i or re-draw in t2t, with denoising strength set between 0.3 to 0.6).
lora proportion setting (0.3-0.6)
Keywords: street, Taiwan roads, Taiwan.
You can visit my Facebook page at
https://www.facebook.com/profile.php?id=100091921453420
for various commercial illustrations and designs!"
這是一個以台灣街道訓練的lora,
台灣街道特色為車輛為靠右駕駛,
道路上容易看到冷氣/廣告招牌/電線桿/廣告看板/汽車/行人/老舊建築/鐵皮屋,
lora訓練時沒有加入任何的主角人物,但在訓練時用低擬合進行訓練,
所以可以方便結合各種人物及其他東西,但在成型時比較容易不擬合,
所以圖片建議2次輸出
(在t2i時使用Hires. fix或是進t2t重繪,Denoising strength設定在0.3-0.6)
lora比重設定在(0.3-0.6),比重越低越不像台灣.
關鍵詞為street,taiwan roads,taiwan,
這是我的粉專 https://www.facebook.com/profile.php?id=100091921453420
如需各種商業製圖及設計都可以找我!