Notice:
The following are strictly prohibited:
– Redistribution outside Civitai (e.g. SeaArt.AI, Tensor.art)
– Selling generated images
– Commercial or profit-based use
– Impersonating real people with outputs or using them to lure to paid services (e.g. Patreon)
– Sharing or selling this model or merged versions

This model is something I made purely as a personal hobby — just for my own fun! I decided to share it with everyone as a small gift, so please don’t be too picky or send lots of requests. I’m not trying to make any money from it, and I don’t want anyone else using it for profit either. Thanks for keeping things fun and respectful!

The Buzz I received has been turned into a way to thank the LoRA creators and to fund on-site training.

I’d love to see cosplay images you’ve generated using this checkpoint. Please share them here!

Realism may be slightly compromised when using LoRAs, but that’s by design — the goal is to preserve character fidelity while keeping the model responsive to detailed prompting. This is because setting the skin textures to be overly realistic from the start can sometimes make it harder to capture the finer details of characters baked into the base model or included in LoRAs — such as the lacework on 2B’s chest.

So, if you want to enhance realism, make sure to include quality tags like those shown in the Recommended prompt below.

Applying block weights to the LoRA can help maintain realism — for example: <lora:A:1:te_bw=-1,1,1,1,1,1,1,-1>

*Please see the article.

The base is configured with clean, youthful skin, but including prompts for freckles or moles can make the output look more realistic.

[ Illustrious model ] Illustrious_Lora is available. Famous characters don’t need Lora.

Sampling method : Euler a , Clip skip : 2

CFG Scale : 7 , Sampling steps :30

Use ADetailer. [Inpainting is not required for V40 or later.]

Recommended prompt :

((masterpiece, hyper quality, best quality, amazing quality, no anime, realistic face only, hyper realistic:2.9, hyper photorealistic:2.9, RAW photo, photo\(medium\):2.9, very awa, hyper realistic skin texture:2.9, absurdres, highres, dramatic lighting, intricate details, detailed face, detailed eyes)),

hyper photorealistic background:2.9, detailed background, high contrast, sharp focus, photorealistic detailed outfit:2.9,

Negative prompt:

unrealistic, 2D, 3D, anime, anime face, illustration, anime style, cartoon, painting, flat colors, manga, blurry, lowres, bad quality, low quality, worst quality, jpeg artifacts, text, error, fewer details, extra details, missing details, watermark, unfinished, displeasing composition, outdated style, early generation artifacts, chromatic aberration, signature, extra digits, artistic errors, username, scan artifacts, abstract elements, grainy, overexposed, underexposed, color bleed, overly smooth, bad hands, bad fingers, motion,

[ Pony model ] Pony_Lora is available. Famous characters don’t need Lora.

Sampling method : Euler a , Clip skip : 2

CFG Scale : 2 , Sampling steps : 20

Use ADetailer

Recommended prompt :

score_9, score_8_up, score_7_up, hyper photorealistic:2.5, realistic skin texture, detailed face, detailed eyes, detailed outfit, detailed background, photo\(medium\), sharp focus, BREAK,

Negative prompt:

2D, 3D, source_anime, source_cartoon, source_pony, source_furry, monochrome, bad fingers, bad hands, missing fingers, 5fingers, intersected fingers, disarticulated fingers, ugly fingers, extra fingers, extra arms, extra hands, extra legs, signature, username, motion,

[ SD1.5 model ] SD1.5_Lora is available.

CFG Scale : 2

Sampling steps : 8

Sampling method : Euler a

Clip skip : 2