KXSR Cinematic Chasm Concept 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 cinematic sequences featuring characters and imaginative scenery centered around a large chasm in the center of the frame. The typical view will be from a heightened viewpoint looking down with subtle camera panning in horizontal aspect ratio generation. 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, [describe the environment/setting in detail] [describe key landmarks or characters] [describe any notable atmospheric elements or lighting] [describe any events or transitions]

Example Prompt

kxsr, A vibrant, stylized jungle scene unfolds, showcasing a partially ruined, whimsical city perched atop a cliff overlooking a deep chasm filled with rushing water. A lone female character with pink hair and a backpack sits on the cliffs edge, gazing down at the scene below.  The city displays a mix of dilapidated structures and lush vegetation.  Small boats navigate the chasm.

Recommended Settings

• CFG: 4.1

• Shift value: 4.0

• LoRA strength: 1.0

• ~73 frames

• 1280x720 horizontal aspect ratio

Technical Details

• Base model: WAN 2.1 - 1.3B Text2Video

• Training dataset: 98 clips

• Resolution: 1280 x 720 (horizontal format)

• Frame count: 73 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: