Test LoRa. Trained on the last 100 anime images I downloaded (mainly early 2000s artbook and megami magazine scans).

It was captioned as if it was real life photography. Thus this LoRa should work well with photo prompts. My tests showed that it works with tags or anime based llm prompts also.

Sampler: Euler A

Weight: 0.9-1.2

ep6 = bigger eyes, yelloish darker hue

ep8 = brighter image, a bit smaller eyes

ep6 might actually perform a stronger style although being trained less than ep8.