This is my test experiment at a hand-detailer improvement model. If it actually works, I might make more or at least a more extensive one.
Note: While it does actually work, I made improved version of the model 'anime hands-improved' preference and update will be given to that one.
Testing out the experiment by doing hands only as the image input on transparent background. As such, you will need base model, character loras, style loras and other negative/enhancers as well.
Trained on 30 images, 10 of each pose, tried doing 3D-360 view of each pose. Most shots covered from finger tip to elbow, all included the wrist and upper forearm. VRoid is not incredibly high poly or detailed on model I used with texture, but the outlines and ambient occlusion of posing app should give it the needed discerning details.
Trained on SDXL with Haruka v2, recommended stats are unknown as of yet due to experimental nature. Tags and triggers associate to the 'pose' of the hand. Model is trained on 3 poses.
- V-sign, aka 'peace sign' or 'victory hands'
- standard fist
- the rocker sign aka 'devil horns' hand sign.
