LoRA created from the public domain of 20 images of the art of Tsukioka Yoshitoshi (1839-1892)

• Recommended weight: (0.5-1) I prefer the value of 0.8 but if you are getting extra limbs you can lower the weight.

• Prompt sentence: <lora:4si4n-step00000800:1> 4si4n, (SUBJECT / SCENE), poster from the series Fifty-Three Stations of the Tokaido by Utagawa Kunisada Toyokuni III/Kunisada III.

• Example: <lora:4si4n-step00000800:1> 4si4n, It depicts a scene of a fight of princess peach fighting bowser, poster from the series Fifty-Three Stations of the Tokaido by Utagawa Kunisada Toyokuni III/Kunisada III. princess peach is holding machine gun, bowser, epic, masterpiece composition, movement,

(I know, the artist mentioned on the prompt is a different one, but it's how the florence captioner identified the dataset so I leave it that ways and it works.)

I'm not an expert in creating loras, I still have a lot to learn. So if you have any comments or suggestions on this model I would greatly appreciate it. I will probably upload better versions of this model because I'm in my lora experimentation era xD.

Shout out to civitai for the streams they made and to @jboogx_creative and @kftiger for the tutorials on it. They helped me get out of my comfort zone and send me to create loras and finally create some decent ones for myself. 🤘🏽

Training info:

Trained in kohya_ss, using an NVIDIA RTX 3060 12gb, 20 images on the dataset. 22 epochs. 13 repeats. 2 batch size. max_train_steps: 1144