Hello my friends. I made a complete rework of my LoRA for one of my favorite characters - Lulu from Final Fantasy X.

Version 3!

Features:

• 220+ dataset;

• It is not overfitted (previous version was overfitted due to small dataset and many 3D blender images);

• Smaller filesize.

• Tokens were reworked manually in Dataset tag editor for easy prompting. So read new tokens for better experience.

Main info:

Trained on animefull;

0.8-1 is the best radius.

Clip skip 2.

Tokens for prompt:

Main appearance:

Luluffx, mature female, 1girl,

black hair, single hair bun, hair over one eye, one eye covered, lipstick, makeup, purple necklace;

optional: braid;

Default dress:

off-shoulder dress, black dress, strapless, fur trim, multiple belts, garter straps, thighhighs

Recommended:

curvy, large breasts, cleavage,

perfect eyes, perfect face, detailed face, sweat

Extra info! I recommend you next prompt order for NAI-based mixes (you can see it in my preview arts):

1. highres, high quality (and masterpiece if it doesn't break character appearence from LoRA). Also other lighting / shadows / sharpness / fov / colorful if you need effects are here. Sometimes it doesn't work if style of anime or game was baked in LoRA.

2. Activation word (from lora civitai page, usually character name), what character is doing (with details), char expressions, camera position, background;

3. Hair details, face details, jewelry, crowns, hats, etc.

4. Clothes and shoes; Or nude, naked.

5. Body details (large breasts, cleavage). Genitals if you want generate it.

Also LoRA trained with some NSFW poses. So you can try it with LoRAs or without it.

Advices for better upscaling and fixing visual bugs:

·         use highres.fix (If you are sure that your prompt is easy for SD); I prefer x4_Fatal_anime_500000_G, animesharp, foolhardy_remacri, siax with 1.5x - 2x upscale.

· use extention after detailer for better faces. It is better than inpaint.

·         use SD Upscale / Ultimate SD Upscale script in img2img;

·         Use inpaint and sketch (or Photopea) for other details;

·         Use Photoshop / Photopea extension.

I will be glad to receive feedback from you and a demonstration of your art with my LoRA!

For previous versions:

This is NSFW model with mostly open dressed, lingerie and naked variants.

LoRA was trained on Perfect world model with 3D NSFW/SFW arts.

2nd ver was trained on animefill-pruned.ckpt.

Also second alt. version was trained on "vanilla" SD 1.5 for best compitability with realistic models like Deliberate v2.

V1.1 is also good! If you like it more - use it. But it was made without trigger words and more with deepbooru interrogates.

60 arts were used (4000 (2000 with batch size 2) steps with 10 epoches).

0.75-0.9 strength of Lora is the best radius.

Clip skip 2 for anime-based models and clip skip 1 for realistic SD1.5 models.

Some necessary Waifu diffusion prompt for better experience:

LuluFFX, mature female, eyebrows behind hair, hair over one eye, covered eye

black hair, hair stick, hair bun,

For dress:

black dress, strapless, bare shoulders, fur trim.

If you have troubles with distorted eyes or bangs:

• use highres.fix (I prefer not to do this);

• use SD Upscale script in img2img (For me 1.25-1.5x with AnimeSharp or UltraSharp is good on 0.2-0.3 denoising strengh);

• Use inpaint (but with SD Upscale and dddetailer I don't need this in 90% times);

• Use Photoshop.

I will be glad to receive feedback from you and a demonstration of your art with my LoRA!

Other works.

https://tensor.art/u/604375505701151021