Agnes Digital (umamusume)

LoRA of Agnes Digital (umamusume). Put LoRA file in your stable-diffusion-webui/models/lora folder and write LoRA notation in prompt to applying LoRA. <lora:your_loha_name:weight>

All of my works is free. If you'd like to support me, feel free to buy me a coffee. 👍

Here are two LoRA of agnes digital based on same data.

One is rank=16 LoRA, the other is rank=4 LoRA.

See difference of two LoRAs, do you see what has higher quality?

Recommend options

• rank=16. LoRA weight 0.7~0.9

• rank=4. LoRA weight 0.7~1.0

• Trigger words

  • base

    • agnes digital \(umamusume\)

  • Racing competition clothes :

    • single thighhighs, sleeveless, single glove, garter straps, white gloves, crop top, frills, sailor collar, multicolored skirt, multicolored clothes, layered skirt, red belt, white shirt

    • (optional) : glowing stick, drooling, saliva

  • Halloween costume :

    • twin braids, white thighhighs, double sleeves past wrists, hair bun, detached sleeves, official alternate costume, chinese clothes, jiangshi, ofuda, bloomers, hair rings

  • Casual clothes:

    • collarbone, casual, belt, jacket, shorts

Use animefull based model

• Settings

  • DPM++ Series(SDE Karras, 2M Karras, etc.)

  • about 20 steps, CFG scale 3.5~6.5

  • kl-f8-anime2.vae

  • CLIP skip = 1

  • If using highres

    • Latent series

    • Denoising strength 0.50~0.65

LoRA training info

• trained on sd-scripts by kohya_ss. Thank you a lot!

• based on Animefull-pruned

• 84 images of agnes digital

  • 4x low quality 28 images

  • 8x medium quality 27 images

  • 16x high quality 29 images

  • => 1 epoch = 792 images

• replaced character feature tags with agnes digital \(umamusume\)

  • horse girl, horse tail, pink hair, blue eyes, etc.

• using aspect ratio bucketing

• resolution : 768x768

• there are two different rank models. rank=16 and rank=4

  • lower rank, lower size.

• using dadaptation

• batch size = 4, 6 epochs trained

• clip skip = 1

This LoRA may be compatible with animefull-based models (ex. AbyssOrangeMix Series...)

All uploaded images are generated by AOM3A1