Bondrewd LoRA

A LoRA for Bondrewd from Made in Abyss! Created with the dataset provided by Brenton#5035, big props to him. I will link the dataset at the bottom.

What to get

I am including most epochs, but I've personally had the best results with the 5th epoch, which I am renaming to *bestdad-BEST, while renaming the 4th to softer and 6th to harder. Not the greatest naming scheme possibly but I want it to be obvious which one should one download. Get bestdad-BEST.

Invoking

I've decided on two tokens this time, bdsks and bfsks. The first one, 'bdsks' (best dad) is the normal token, it's associated with Bondrewd's normal look. The second one, 'bfsks' (best furry), however, is associated with his transformed look, one where he exhibits animal characteristics.

If you're trying to get that transformed Bondrewd you should use 'bfsks', you can increase its weight. I also recommend including tags such as 'white fur', 'claws', 'tail'. It will be harder to get than the normal version though, because that look is not very consistent in both the artworks and anime.

Previews

All the previews have prompts included, so read that!

I've generated some on my mix holo-youcha, it's bad, don't use it, (https://huggingface.co/Aotsuyu/houjicha)[it's here.]

bestdad-000005 is just bestdad-BEST before renaming.

Prompt

(best quality), (bdsks), 1boy, solo, male focus, mask, black gloves, whistle, manly, school swimsuit, name tag
Negative prompt: (low quality, worst quality:1.4), (bad anatomy), out of frame, by bad-artist, monochrome, suit, jacket, breasts, artist name
Steps: 32, Sampler: DPM++ SDE Karras, CFG scale: 10, Seed: 3609972322, Size: 520x712, Model hash: 36784024d5, Model: holo-youcha, Clip skip: 2, ENSD: 31337, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: bestdadv2-000005(f6bc002a6491), AddNet Weight 1: 0.9, Discard penultimate sigma: True

Prompt

(best quality), bdsks, 1boy, solo, male focus, mask, black gloves, whistle, horror \(theme\), mksks style, dark, darkness, black theme, very detailed background
Negative prompt: (low quality, worst quality:1.4), (bad anatomy), out of frame, by bad-artist, monochrome, suit, jacket
Steps: 32, Sampler: DPM++ SDE Karras, CFG scale: 10, Seed: 657765566, Size: 920x512, Model hash: 36784024d5, Model: holo-youcha, Denoising strength: 0.3, Clip skip: 2, ENSD: 31337, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: bestdadv2-000005(f6bc002a6491), AddNet Weight 1: 0.9, AddNet Module 2: LoRA, AddNet Model 2: dpep2 768(5604eb3087e3), AddNet Weight 2: 0.3, Hires upscale: 2, Hires steps: 15, Hires upscaler: 4x-AnimeSharp, Discard penultimate sigma: True

Prompt

best quality, 1boy, mask, black jacket, white whistle, bdsks, pov across table, food, food focus, pizza, restaurant, (spoken blush:1.3), candles, speech bubble, romantic
Negative prompt: (low quality, worst quality), pelvic curtain, chinese dress, crop top, helm
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 826091359, Size: 568x768, Model hash: 6e430eb514, Model: anything-v4.5-pruned, Denoising strength: 0.35, Clip skip: 2, ENSD: 31337, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: bestdadv2-BEST(f6bc002a6491), AddNet Weight 1: 1, Hires upscale: 2, Hires steps: 15, Hires upscaler: 4x-AnimeSharp

Prompt

best quality, 1boy, mask, white whistle, bdsks, from above, spoken exclamation mark, covering breasts, (wariza), hand on own chest, looking up, undressing, bra, open jacket,
Negative prompt: (low quality, worst quality), pelvic curtain, chinese dress, crop top, helm
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 3011174594, Size: 568x768, Model hash: 6e430eb514, Model: anything-v4.5-pruned, Denoising strength: 0.35, Clip skip: 2, ENSD: 31337, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: bestdadv2-BEST(f6bc002a6491), AddNet Weight 1: 1, Hires upscale: 2, Hires steps: 15, Hires upscaler: 4x-AnimeSharp

Prompt

best quality, 1boy, (mask:1.1), white whistle, (bdsks:1.2), close up, sitting, crossed legs, ice cream
Negative prompt: (low quality, worst quality), pelvic curtain, chinese dress, crop top, helm
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 4235027472, Size: 568x768, Model hash: 55042b85bc, Model: pastelmix-better-vae-fp32, Denoising strength: 0.35, Clip skip: 2, ENSD: 31337, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: bestdadv2-BEST(f6bc002a6491), AddNet Weight 1: 1, Hires upscale: 2, Hires steps: 15, Hires upscaler: 4x-AnimeSharp

Prompt

prompt: mksks style, (highres:1.2), (ultra-detailed:1.2), (8k:1.2), (best shadow), high contrast, (((vibrant colors))), (Painterly:1.2), realistic lighting, beautiful detailed glow, bdsks, bondrewd, 1boy, solo, male focus, mask, black gloves, whistle, upper body, facing viewer, purple glow, 
Negative prompt: (bad-artist, bad_prompt), deformed, blurry, lowres, bad anatomy, text, error, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, artist name, ugly, disgusting, poorly drawn, mutilated, mangled
Steps: 16, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3070255080, Size: 512x768, Model hash: 943a810f75, Model: pastelmix-better-vae-fp32, Clip skip: 2

Model and epoch comparison

This is trained on base NAI so any models off of that should do fine.

Dataset

This is the dataset Brenton provided, they say anyone is free to use it for their own LoRA training and similar purposes.

https://mega.nz/folder/KnYy0KhT#5JYNVdAg0M7MuxWJdS7owA

Contact

If you have any questions, you can DM me on twitter.

My pixiv if you're up for lewds: Pixiv