Usage:

To have this model work properly, add: framebinder, restrained
If the character isn't suspended in the air, add: suspended

If the character does have their legs spread, add: spread legs

WARNING: Due to the training method used, you NEED to specify character/background tags for them to appear. Otherwise it will generate exactly the character used for training.

The lora will work well with character and style changes, though the 3d style can be hard to remove sometimes.
Backgrounds work decently, though can often look flat or like photos, since all the backgrounds used to train the model were photos themselves.

Faces also are a bit low detail, but can be fixed with a detailed plugin, such as a-detailer for a1111.

Background:

A PonyXL model that makes a framebinder concept very similar to the one made for SD1.5 here: https://civarchive.com/models/135405/frame-hanger-concept

This was also made using a 3d model in blender with 60 images taken from different angles, clothes, and backgrounds in order to try and keep the lora flexible.

Due to the limitations of the training data used, future improvements for v2 may include:

• Replacing the photo backgrounds with 3d backgrounds

• Multiple characters in frame for better multi-person support

• A style filter pass over the rendered 3d images to introduce more style variation