Pony version is now available. I will stick to Pony and derivates for further training.
A collection of sword stances.
It is a work in progress with a very bad dataset. If you generate some good quality images, please post them, they will be used for a further training.
Resolution of 1024x1024 is the most effective.
Further training could happen according to the interest.
Battoujutsu
A ready to draw/incoming attack stance.
Trigger words: battoujutsu sword stance, ready to draw
Stability/consistency:low, several generations needed. "sheath, sheathed,katana," will help with some checkpoints.
High guard
A sword parry stance (ochs).
Trigger words: high guard sword parry stance
Stability/consistency: medium, model is undertrained and may need emphasizing of the trigger sequence. It will be corrected within the next days.
Incoming attack
A self explanatory stance.
Trigger words: incoming horizontal sword attack OR incoming vertical sword attack
Stability/consistency: medium, further training needed.
Brave perspective
Can be used with or without weapon.
Trigger words: brave perspective +/- holding sword
Stability/consistency: high
Description
FAQ
Comments (2)
I really love this stance, but especially in combination with sword slashes all around. Do you you think that effect might be possible to train?
I found it to be very difficult to prompt and tried many things from controlnet to ipadapter with custom masks and whatnot to have that generate properly.
Something like this devil may cry wallpaper with vergil for example:
https://i2.wp.com/images.gamersyde.com/image_devil_may_cry_5_special_edition-42628-4599_0016.jpg
I think everything is possible with a good dataset and tagging. Asking in the bounties section here will surely find some interested creators.