KXSR Atelier Ryza Character LorA for WAN 2.1 1.3B T2V

In collaboration with @machinedelusions [https://civarchive.com/user/machinedelusions]

KXSR Labs presents:
This LoRA enables the creation of videos featuring Ryza from the Atelier Ryza video game series. Specifically from Atelier Ryza 2 character model. The training data did NOT include NSFW. The model will consistently create the character at 1280x720 in horizontal aspect ratio generation using the 1.3B parameter WAN 2.1 Text2Video model. Can still create fun results at other aspect ratios/resolutions.

Use the trigger word "kxsr" to activate the model's specialized training.

Prompt Format

kxsr, in the style of [cgi/realism], [she/her] [POSE/ACTION: stands/walks/leans/poses] [describe location/background] [describe camera movement and angle] [describe additional environmental or lighting details]

Example Prompts

kxsr, in the style of cgi, she stands in a vibrant green garden with a castle behind. She adjusts her hat, then stretches her arms out to her sides. The camera pans left, showcasing the lush garden and castle. She maintains her pose, a gentle breeze rustling the trees. 

Recommended Settings

• CFG: 4

• Shift value: 4.0

• LoRA strength: 1.0

• ~65 frames

• 1280x720 horizontal aspect ratio

Technical Details

• Base model: WAN 1.3B Text2Video

• Training dataset: 35 clips + 4 HD images

• Resolution: 1280 x 720 (horizontal format)

• Frame count: 65 frames per clip

• For optimal results, maintain these specifications during inference

This LoRA works best when you provide detailed descriptions of both the subject and the surrounding environment while following the prescribed format.

Screenshot shows my typical inference testing setup for lora evals: