Type | |
Stats | 418 13,842 |
Reviews | (64) |
Published | May 18, 2024 |
Base Model | |
Training | Steps: 2,535 Epochs: 13 |
Usage Tips | Clip Skip: 1 Strength: 0.8 |
Trigger Words | tank top shirt low-cut chest low-cut cleavage high-cut crewneck extreme low-cut dropped side holes |
Hash | AutoV2 5AB7990E60 |
This is a SDXL version of my SD 1.5. SDXL isn’t as clueless about “armholes” as SD 1.5 but it is still hard to do it so…
Are you tired of SD giving you this completely prude tight fit high-cut armholes tank tops? Do you want your generated dork himbos and bimbos showing off their muscles at the gym?
Let’s fix it!
Lora weight works from 0.3 up to 1.0.
Start with 0.7
There is no trigger word. But you should reference a tank top, sleeveless shirt, gym stringer or singlet. If your males are getting too much belly crop top shirts or shirtless it helps to use “Shirt:1.2” and just “tank” without the word “top”.
Supporting prompts:
shirt, tank top, tank, sleeveless shirt, (high crewneck:1.3), low-cut chest (males), low-cut cleavage (female), extreme low-cut dropped side holes, side boobs (female), side pecs (males), (male/female) nipple slip, side view, profile, front view, full body medium shot, long shirt, loose shirt, muscle body, thin body
Negative supporting prompts:
skin holes, shirtless, jewelry, necklace, muscular, head out of frame
I did a lot of testing without the lora and SDXL is better at this task comparing to SD15 but has side effects. Here using “low-cut dropped side holes” have way worse side effect because it has more resolution, it will generate more holes to the shirt, and we don’t want that. So the other best tags are “low-cut large dropped armholes” and “low-cut dropped sleeve opening”.
So, if you really don’t want to use my lora I guess you could prompt like this:
(extreme low-cut large dropped armholes, extreme low-cut dropped sleeve opening, low-cut cleavage, side pecs:1.3)
*pecs or boobs
I still rather use my Lora. Especially because it can do “extremes” much better.
It was trained with “low-cut dropped side holes”. With the lora the holes won’t happen as much. But it might be a good idea to use “skin holes” on negative.
SDXL is unfortunately not as good with necklines without the lora. It does know what a “high-cut neckline” is so you CAN use this tag if you want. Works especially good with the lora. The training data was trained with the SD1.5 captions “high-cut crewneck”, so I still suggest to use that tag instead of “neckline”.
But for “low-cut neckline” it won’t work without the lora. I still think the best one is “low-cut chest” or “low-cut cleavage”.
Warning: If you use cleavage for men, they will get boobs!
_________________________________________________________________________________________________
Try out SD15 and TI here (will update later)::
Stringer Gym Tank Top / Dropped Side Holes / Large Armholes – SD 1.5
Stringer Gym Tank Top / Dropped Side Holes / Large Armholes – SD 1.5 Textual Inversion (I'll probably never update this, my TI tests were bad, it is too hard for SDXL)
_________________________________________________________________________________________________
The base checkpoint is the “sdXL_v10VAEFix” 6.7GB. So, it should be very flexible with any checkpoint. As of right now, I don’t recommend any other finetuned model as they don’t give expressive better results compared to the original one, but you should use the one you like.
CFG: 5.5 - 9
DPM++ 3M Exponential (50 steps or more)
DPM++ 2M Karras (25 Steps or more)
DPM++ SDE Karras
DPM++ 2S a Karras
V2 is with block weight analysis. V1 is all blocks.
Because the base and the contours of the shirt and body are created in the beginning of the image denoise, Loractl is a GREAT option if you want the lora to not interfere too much with facial structures and details that get added later for a specific character or celebrity Lora. I would recommend starting high weighted (0.8) at 0 steps. At 40% of the steps drop the weight to 0.35 or even 0.
For example <LORANAME:0.8@0,[email protected],[email protected]>
Want to have some “fun”? Install wildcards dynamic prompts extension: https://github.com/adieyal/sd-dynamic-prompts . And wildcard manager. Here is a good, crafted script I made for testing. Paste on prompt:
{side view|from below|[0.2::(front:0.6)]} (medium shot:1.1) of a YetAnotherWildcardCollection-main/States/Nationality {man|woman} {18|28|45|80} years old on a YetAnotherWildcardCollection-main/Background/*, wearing a {dark|ligh| } ({white|black|red|pink|purple|green}:1.1) (tank:1.1) {(high crewneck:1.2)|(extreme low-cut neckline:1.2)}, {(extreme low-cut dropped side holes:1.1)|low-cut dropped side holes| }, stringer (shirt:1.2), { |(nipple slip:1.1)}, natural lighting, 4k uhd, dslr, soft lighting, high quality, Fujifilm XT3
Problems with the current Lora:
Using “extreme low-cut dropped side holes” will give holes to other things. Use “skin holes” on negative
This epoch (13) is kind of a little bit overtrained to white or plain colored backgrounds while lower epochs were more flexible. But I did a lot of testing and I do think epoch 13 was the best one to get the shirt effect, so I uploaded this one.
Bias to muscular body type. Just prompt it away
Nipples showing weirdly below the shirt. I always see this with SD, don’t think it’s exclusively a problem with this lora
Some settings: Trained 1024 res. More than 205 pictures. 2 steps each image. 3 different folders/concepts (2_stringer shirt / 2_low-cut chest stringer shirt / 2_low-cut cleavage stringer shirt). "prodigy, 13 epoch, BATCH 3, rank 8/1,Scale weight norms 1, snr gamma 5, Noise offset 0.0357, no regularization images”
Hopefully you can leave some results and some comments. Any idea is appreciated. Thank you.