---
language: 
  - en
thumbnail: "https://huggingface.co/mirav/ashojo/resolve/main/00345-3313293082.png"
---
Model trained on shojo manga artists.<br>
Currently not on civitai because this is very much a beta. If you use this model, please provide me with feedback! I'm Mira!#1697 on Discord. <br>
See the images for examples.<br>
I think it performs best with the [aid28 and/or badv5](https://civitai.com/models/16807/useful-quality-embeddings-animeillustdiffusion) negative embeddings, but this is a bit subjective. Aid28 is more authentically retro, but badv5 is probably more like what most people want.<br>
I use [ClearVAE](https://huggingface.co/JosefJilek/loliDiffusion/blob/main/loli_clearvae_main_FP16.vae.pt), but any VAE should be fine. <br>
I recommend R-ESRGAN 4x+ Anime 6B for upscaling, but other upscalers [are fine](https://huggingface.co/mirav/ashojo/blob/main/upscalers.png).<br>
The sketchier gens use Latent (Nearest).<br>
See [artists.txt](https://huggingface.co/mirav/ashojo/blob/main/artists.txt) for names to add to your prompts. How well it replicates each style varies, but each produces a different result.<br>

Example on recommended settings:
<img src="https://huggingface.co/mirav/ashojo/resolve/main/xyz_grid-0012-888.jpg">