FinishingR
When I created the merges for RMBtest1 to 4, I merged the basic parts and then added LoRAs to finish them. Although the settings were slightly different for each, the LoRAs ware generally similar, so I summarized the LoRA parts. The basic strength when used is 1.0.
The basic difference from the actual ones is that RMBtest focused on Asians, so Asian-oriented face correction LoRA was used (except for RMBtest1, where I mistakenly put in European-oriented), but FinishingR uses European-oriented face correction LoRA (I use this for most of the realistic models I create).
The 30 or so types of LoRAs are distributed as follows:
Costume 12
Background 7
Character adjustment 5
Art style 5
Asset 2
No matter what base I use, by adjusting these parts, I feel that the final character will be close to the one I usually create (although if I create it properly, I will add more adjustment LoRAs). In addition, I adjust it for photography, control the overall image quality, and add fantasy elements that are generally lacking in photography. In summary, using this allows me to create characters with faces that I like, and enhances photographic expression.
The costumes are merged at a low intensity of about 2% each, but the direct impact is small. The costumes become a little more diverse, and the tendency of photo models to be underdressed can be alleviated.
FinishingA
Similarly, FinishingA is used to improve the face of a character (to my taste) and the anime-style painting of an anime model. The base strength when using this is also 1.0.
Since the character elements are strengthened, the handling of the character during processing also changes, which ultimately affects the composition. Its effect is easier to understand than FinishingR. In terms of content, it is like character (or style) LoRA + x, so it affects the character of the original model. It is not strong enough to replace the character like Character LoRA, but it has an effect in a certain vector, so you need to be careful about where you use it.
Both FinishingR and FinishingA strengthen things other than the character, so they can be used to improve the output of a model that is a little dull (depending on your character preferences). These are LoRAs that perform the usual corrections together and also correct the character, but since I do this purely for my taste, if you want to make a character to your taste, you should make additional adjustments. Painting and other things are up to your taste, so it may be quicker to make something similar yourself. By the way, in the case of my models, the anime painting has been done in the Urushihara style since KawaiiAnimetic 1.0.
Who creates the AI faces that are often seen in both photo and anime?
描述:
训练词语:
名称: FinishingR.safetensors
大小 (KB): 253506
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
