
V4:
Strenght: 0.99 For most cases - Solo or Mixing everytime perfect. If the result is bad the other lora might be trained badly, so lower it!
Clip Skip: 2
Steps: I always use 35
Sampler: Everyone comes with bad and good sites, for realistic go with DPMPP_2m and with many LoRA´s go with Euler_A.
Im happy to close my first personal "Perfect" LoRA. Personally i think no Face LoRA have any quality like this now on CivitAi. Thats what i wanted to achieve. Why do I think that? I wanted a LoRA that creates incredibly good faces and is flexible with other LoRAs. That's why I say "perfect" because I rarely found perfect loras from other creators, especially face loras. My special thanks goes to @__2_ his article gave a insane boost to the final result ty! Of course you can continue to test it 100 times, but at some point it is very acceptable. I have now trained exactly 43 LoRAs for Kim schiele. What adds up to about 2150 minutes training time the past weeks. I think im insane.
V3:
LoRA strength: 0.8 - 1.05
Clip Skip: 2
Steps: 20. Use Heun or Euler for best results.
V2:
LoRA strength: 0.8 - 1.05. When combining with other LoRA´s go 0.8 and when solo use go higher :)
Clip Skip: 2
Steps: Everything above 30. Best: 50
V1:
My second face model i can upload with a clear conscience. I feel like i'm getting better and better and i'm learning. There's still a lot i'm not sure about in the training process but we'll see. Please give me some feedback would be great!
Feedback please!
Please give me some feedback would be great! :))
I also do requests! Also Private requests.
描述:
Basically V2 but even more finetuned!
训练词语: k1m5ch1ele
名称: KimSchieleLoraV3.safetensors
大小 (KB): 73847
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success