MangaGen
It was a bit of a challenge, but I made a LoRA version of the manga-style model I was experimenting with in ckpt. It's divided into two LoRAs: the basic part and the concept LoRA.
MangaGen1
The base strength is 1. Specifies the strength of making the model manga-like. Setting it to 1 or higher has little effect on the model's appearance, but the image quality may decrease. In the case of color, the colors will become lighter.
The base strength is 2. This is an auxiliary LoRA, but think of it as something to adjust the number of frames, etc. The appropriate value varies depending on the model and the image.
This time, I tried to adjust it with color manuscripts in mind. In that case, negative prompts are recommended, which is rare for my model. The balance is slightly different from that in ckpt, and the image quality has improved, but there are too many close-ups of the face. The frame layout is also a bit monotonous. This is a future issue.
Many of the additional samples are output with
<lora:MangaGen1:1.25> <lora:mangaC:2.5>.
For comparison, the same prompts and seed bands are used in many cases, except for the LCM specification. Depending on the model, it may be suitable or unsuitable.