Based on images with heads and emotions only. There are no full-length images in the dataset

support me on Boosty, all new versions are posted there (latest uploaded lora Danmakuman Retro Style (DRSv1.0) LORA / SDXL / PONY DIFFUSION)

P.P.S. new model Anime Screencap / LORA / PONY DIFFUSION and NEW ERA 3.0 (MODEL) PONY DIFFUSION / SDXL on Boosty

Lora should be uploaded along the path webui\models\Lora

Lora trained on this model 4th tail (?????? model), so it’s better to use it so that it’s “as intended” 

trigger words: characters sheet, multiple views,

additional trigger words: reference sheet, expressions

tips: use booru tags with different emotions in one prompt. For example:

one eye closed, looking at viewer, blush, smile, open mouth, close mouth, closed eyes, profile, :d, teeth, crying, angry, surprised, etc. This way the same character will be on the same art, but with different emotions (some may not work, experiment)

Of course you can use standard pony difusion prompts: score_9, source_anime, score_8_up, score_7_up (but the model works fine without them)

negative prompts: bad ???????, error, (worst quality, low quality:1.1), distorted, score_4

You can achieve more than just images with heads if you reduce the weight of the lora

weight: 0.7-1 (use any, depending on your preference)

LORA knows a little about retro, use 1990s (style) or 1980s (style) or retro artstyle (for amplification)

Thank you! :3