WARNINGS:
I generated some nude pictures for the batch of images to be sure it can make the body and clothes. So just be sure to add clothing prompts.
Due to the nature of the theme of what this Lora generates, it is in the territory of short stack, and chibi. Which isnt that far from child like bodies. You can add child, chibi, loli, in the negatives, also you can use Tall, in positives prompts.
I recommend strength 1-2.5 for Power Puff Limb and/or Power Puff Addon. Personally got best results on 1.5
Trigger words, and what they are supposed to do:
pplimbs: activator and general use.
Summons the monstermonster defeated.ppbody: activator and general use.
pplegs: Focuses on making the legs, (I recommend just using this prompt. it mostly focuses on legs, decreasing weird arms and T posing)
pparms: Focuses on the arms,
pphands: Some images had hands, so i prompted it. So it can make some hands, or put it in negative to get the nubs.
From behind, back view: to look from behind (Side view and other viewing angles seems to work now too.)
Power Puff Addon:
This model is an addon (and can work by itself), and a test for something in the future.
I tried editing the pictures used for training to have transparent backgrounds and focus on certain things. Fixing the big head issue. and removing weird hands.
Added extra poses, to try to make it more flexible.
Added some pictures to give me some lead way for a new project. Featureless feet/soles.
Trigger words for Addon:
ppaddon: this is a general activator tag, all or most pictures have this tag
ppalegs: pictures focused on legs. anything pelvis and below.
ppafeet: this was tagged on any images below the knee. Sometimes it will create a foot with 0-3 toes. still buggy. future project. (Featureless feet)
ppabody: This does the same as ppaddon, and legs, but it was tagged on images that had a torso in it too.
soles: This is also a tagged for future featureless feet project. this is so i can generate some images with soles.
Negatives:
Notes:
starts to not work when adding things that collide with the legs and arms. or things that needs a foot.
Why is everyone so Angry? Why weird eyes? cause of the bug eyed powerpuff girls... hopefully over time it will be fixed. Kinda fixed but for pony only. Use the addon. Try using just
Sometimes I use this prompt string i found that makes things look nice.
score_9, score_8_up, score_7_up, score_6_up, depth of field, perfect soft lighting, best shadows, perfect reflections, volumetric, detailed image, high resolution, 5k resolution, photo,Description
Finally an update I can post! I must first say DO NOT USE "PPLIMBS" i know you might see it in the meta data. Apparently for me at least, that spawns some weird monster. (ill post a pic at the bottom of the description).
I got the Lora to work at a strength of 2.5. Higher, and the character started to glow.
NSFW WARNING: I generated some nude pictures for the batch of images to be sure it can make the body and clothes. So just be sure to add clothing prompts.
Trigger words, and what they are supposed to do:
pplimbs:
activator and general use.Summons the monsterpplegs: Focuses on making the legs, (Try using "thick legs" or "fat legs" in addition for more thicker and/or rounded legs.)
pparms: Focuses on the arms, (Try using "thick arms" or "fat arms" in addition for thicker and/or rounded legs.)
pphands: Some images had hands, so i prompted it. So i can make some hands, or put it in negative to get the nubs.
From behind, back view: It kind of works, but doesn't always have the right leg shape. (I will try to train this more in the future, but for now, try to generate a front facing position, then IMG2IMG with added prompt "from behind, back view"
Negatives: (Not sure how well they work)
fingers, hands, toes, soles, shiny skin, feet
Notes:
starts to not work when adding things that collide with the legs and arms. or things that needs a foot.
So I tested this on 3 main checkpoint models I use on automatic1111. AutismmixSDXL, UnrealSpice, and Pony Realism.
The text2img works on Autism, but more cartoon like. It kinda works on UnrealSpice, didnt really work.
If you want to try to get some realistic type pictures, try making a cartoon version and then Img2Img it using a realistic lora.. Usually works for me.
When I Img2Img, i add this prompt string. it helps.
score_9, score_8_up, score_7_up, score_6_up, depth of field, perfect soft lighting, best shadows, perfect reflections, volumetric, detailed image, high resolution, 5k resolution, photo,and play around with the denoising levels.
Plans for future update:
Im gonna change the prompts. If anyone know how i can make it work on just strength 1, let me know.