Dataset of 330 sceencaps. It doesn't really know individual characters. Trained for one epoche. Later epoches wouldnt do background that good anymore and focus more on characters.

Screencaps used for training were uncropped wide screen format. But as you can see vertical seemss to work just fine.

Captions for the dataset are here:

https://huggingface.co/datasets/synta/monogatari_lora_captions

v0.1

Weight: 1.0-1.2

Trigger: monogatari style, SHAFT style

You can use only "monogatari style" as trigger, which makes color and sharpness punchier. If you add "SHAFT style" on top then it gets a subtle blur (in a good way) and is a little bit more desaturated. "SHAFT style" is generally the trigger specifically for sceneric/background screencaps without characters.

v0.1 STRONG

Disclaimer: Can do some characters a little bit.

Weight: 0.8-1.0

Trigger: monogatari style

kizu dim vibe (you can ignore this, just testing)

Oshino, a child with strawberry blonde hair and yellow eyes.

Hanekawa, a teen girl with brown hair and glasses.

Koyomi, a teen boy with short black hair and dark grey eyes.

Memen, a man with strawberry blonde hair and yellow eyes.

kiss-shot 1-3 ep10 / ep9

Captions: https://huggingface.co/datasets/synta/kiss-shot_krea2_lora_captions

Trigger (adding age is optional):

Vintrolek, an (8 years old) girl with short blonde hair and orange eyes.
Parasine, a (12 years old) girl with orange eyes and long blonde hair.
Pettalon, a (18 years old) female with a long blonde ponytail, a long hair ribbon and orange eyes.

You can probably alos just stack all 3 name triggers in a single sentence to trigger a basic style without specific characters like Vintrolek, Parasine, Pettalon