Update: Please if you see any of my models somewhere else report or let me know! Thnx
Please read version info (about this version)
Trained on base model(s) xl + sd1.5 + pony + illu + f1d + zit + zib
Update: Added perfection style model (Hands + Feet + all in one)
Update: Added Detailed style (Hand focus + all in one)
Update: Added SD 1.5 Model
Update: Pony version added! (Please read version info. before using it)
Update: Illustrious version added.
Update: added gun focus (weapon focus) F1D model.
Trained on base model xl 1.0 + FLUX.1-dev + Pony + Illustrious
Feel free to check feet model as well.
In addition to Archery style.
Feedback and ratings are much appreciated.
Other models to check:
Skin Tone (Cinematic Photography) Style XL (Human skin color)
Realistic Skin Texture style XL (Detailed Skin)
Description
This is a style version. Might be helpful for some users.
FAQ
Comments (30)
SDXL in the name plz!!!!
I would do that but It has another 1.5 version. Unless I separate them.
I generated a little and this is what I noticed: perhaps the trigger word in the plural is not very successful, since in the case when there is one hand in the image, the SD almost always tries to add a second one. At the same time, the quality of these mutations is excellent. It is also possible that there are not enough photographs of fingers from the side in the dataset, because... often unnaturally turns the nail forward.
Try different version. Not all versions work the same.
Doesnt work in many cases or make worse
Try Style version, "Perfect Hand Style"
I'm using afroditexl 2.0 model.
I can't tell if it is the same or worse with or without this Lora. I don't see any difference, still deformed hands. Maybe your Lora only works on close-up hands?
I don't know. I get a lot of reviews, some works well others don't. There is a style model you can give it a try, not sure if it might work or not.
You need to use it with Adetailer to get the best results, sometimes you have to reduce the inpainting weight to 0.2 or 0.3 depending.....some hands are beyond saving though :D
Still gives 6 fingers consistently.
The correct usage of this Lora is in ADetailer, and the effect is great
Hello, please ask the detailed usage, I can not use, thank you
@love530love Enable Adetailer, select hand_yolov8n.pt from the dropdown, set the inpainting weight between 0.2 to 0.4 (try 0.4 first and reduce if needed). High res fix helps too.
@Corbe Do you mean this Lora is not important ? and everything is in ADetailer?
@codegix In Adetailer, select hand_yolov8n.pt, and then enter this lora in the Adetailer text input box to enable it
@Corbe dont you need to change anything else? Which Checkpoint are you using? This methods works well with SD 1.5 and not any other versions of SDXL.
@Corbe also seems like it wont work for 1024x1024 ratio. Data is trained on 512x512
@NOOBDA RealVis or DreamDiffusion, I'm getting some good results
Do you plan to make Pony XL version?
Use 2 passes to get a better chance.
1st-pass: use the lora directly to get a roughly correct position and setup
2nd-pass: use detailer model and this lora to inpaint
I see people using this for pony xl, is that a mistake or is this actually making a difference?
It's a mistake, people are just idiots
LoRA trained on SDXL Base and other models is not compatible with Pony!
@prgfrg23 not true. ive used clothing loras in pony and it works just fine
@Vaa I should have written differently, LoRA trained on SDXL is compatible with Pony, but it works worse than LoRA trained on Pony
Conversely, it also works LoRA learning on Pony will not work very well on SDXL
The main problem with the Pony model is that there is almost no data left on the SDXL Base in this model
@prgfrg23 i believe some checkpoints are merges of ponyxl with sdxl, but I have to say personally using euler max sampler + this lora is still more miss than hit for me trying to inpaint anatomically correct hands.
I tried "hand style" one and it works pretty ok with pony checkpoint. At least hands get better with it.
It's hit and miss! Hopefully when SD3 is out thing might get better.
possibly one of the most valuable LORA's in the world right now.
Update: Added perfection style model (Hands + Feet + all in one) XL
Update: Added Detailed style (Hand focus + all in one)
I have'd, but still a lot negative results either