This is my first attempt to create a human LoRa.
I'm not sure why, but sometimes the images are generated in an anime style, and other times in the style of the original design.
I suspect that the more details you include in the prompt, the more it leans towards an anime style.
If anyone knows more about this, I'd really appreciate any insights! 😊
It is possible to create pictures of her with and without her helmet, but with it the results doesn't look as good as i wished:
For the helmet try:
"Racing Gear, Helmet"
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
V2.0
trained on "3D"-tag
trained more specific to clothes, so it might be easier to dress her with different clothes.
______________________________
Character: laDash,
Style: 3D / source_anime
Appearence: ponytail, dark-skinned female, brown hair, multicolored hair, yellow/orange hair extensions, brown eyes,
Original cloth: blue boots, yellow/orange striped boots, blue suit, textured suit, yellow/orange coat, blue striped coat, multicolored hair, yellow/orange hair extensions, silver earrings
______________________________
Example for original Character appearence:
P-Prompt:
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, 3D, Expressiveh, very aesthetic, pretty eyes, intricate iris, perfect eyes, expressive eyes, masterpiece, best quality, BREAK, laDash, ponytail, dark-skinned female, (dark skin:1.2)*, brown eyes, brown hair, multicolored hair, yellow/orange hair extensions, small breasts, BREAK, blue suit, textured suit, yellow/orange coat, blue striped coat, silver earrings, BREAK, 1girl, solo, looking at viewer, smile, outdoors, teeth, looking back, from behind, grin, blurry, lips, blurry background, tight, photo background,
Example for alternative appearence:
P-Prompt:
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, source_anime, Expressiveh, very aesthetic, pretty eyes, intricate iris, perfect eyes, expressive eyes, masterpiece, best quality, BREAK, laDash, ponytail, dark-skinned female, (dark skin:1.2)*, brown eyes, long hair, brown hair, multicolored hair, yellow/orange hair extensions, small breasts, BREAK, white pants, black sweater, pants, tight, tight pants, earrings, long sleeves, jewelry, sweater, BREAK, 1girl, solo, looking at viewer, smile, ass, outdoors, teeth, looking back, from behind, grin, blurry, lips, blurry background, photo background,
* helps if the skin is generated too bright
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
Illustrious V1.0
trained on the same tags as for the Pony version v2.0
______________________________
Character: laDash,
Style: 3D (without "3D" you might get more results in Anime Style)
Appearence: ponytail, dark-skinned female, brown hair, multicolored hair, yellow/orange hair extensions, brown eyes,
Original cloth: blue boots, yellow/orange striped boots, blue suit, textured suit, yellow/orange coat, blue striped coat, multicolored hair, yellow/orange hair extensions, silver earrings
______________________________
Example for original Character appearence:
P-Prompt:
very aesthetic, pretty eyes, intricate iris, perfect eyes, expressive eyes, masterpiece, best quality, BREAK, laDash, ponytail, dark-skinned female, dark skin*, brown eyes, brown hair, multicolored hair, yellow/orange hair extensions, small breasts, BREAK, blue suit, textured suit, yellow/orange coat, blue striped coat, silver earrings, BREAK, 1girl, solo, looking at viewer, smile, outdoors, teeth, looking back, from behind, grin, blurry, lips, blurry background, tight, photo background,
Example for alternative appearence:
P-Prompt:
very aesthetic, pretty eyes, intricate iris, perfect eyes, expressive eyes, masterpiece, best quality, BREAK, laDash, ponytail, dark-skinned female, dark skin*, brown eyes, long hair, brown hair, multicolored hair, yellow/orange hair extensions, small breasts, BREAK, white pants, black sweater, pants, tight, tight pants, earrings, long sleeves, jewelry, sweater, BREAK, 1girl, solo, looking at viewer, smile, ass, outdoors, teeth, looking back, from behind, grin, blurry, lips, blurry background, photo background,
*the Illustrious Version generates her skincolor just fine. But you can always adjust the tone as you like
Description
FAQ
Comments (3)
Did you tag the applicable images in training "3D"? From what I've seen not doing that is what causes everything to be a weird half-3D (though not normally inconsistently applying 3D). Since it's a pony model, you might try adding "source_anime" to your generation prompt (but not your training data)
I also see your example images don't list any clothes. Did you not tag the outfit parts? From what I can find, I think it would be "turtleneck, blue suit, orange coat" though I can't find any pics from below the knees to give footwear tag advice.
Oh yeah, i might try another Training with the 3d Tag.
For the clothes i used some picturres just from the parts like "blue boots" "yellow/Orange coat", "blue suit", "textured suit".
I wasn't Sure to use these Tags on every image in the dataset. I thought of i do that, it might be a Problem to generate different clothes.
After work i'll try another training.
@LittleApe You want to be consist with tags since that's how PDXL learns. If you don't tag clothes, the model learns the clothes more as part of a character's body. The effects are way worse in 1.5, but it's still not great.