SingularUnity - Wan2.1 MotionCraft

April 15, 2025 Update:

After Additional Testing this model works best with clothed subjects, normal size breasts, and medium shots perspective. I will be making another version of this model that is more explicit because the generation seem to be fairly chaotic with some tearing between dance frames.

ARTFUL POISE Details.

You know what, I'm not entirely sure what I was going for on this. It's got some walking, some breast physics and dynamics, some dancing. I suppose it was meant to be a more cinematic kinda artsy, kinda sexy model.

This is not like V2.0 of Motion Craft, it's a different rendition using content from Motion Craft v2.0 but different captioning and a wider scope. Use at your own peril and enjoyment.

Triggers are more of a guide, refer to the sample prompts based on caption data to derive a prompt.

It's 5 am and I'm tired. This is WIP!!!!

Sample Prompts:

A cinematic zoom in captures a woman with long, flowing black hair standing against a serene backdrop of a misty lake. The camera pans around her, highlighting her smooth, fair skin and the subtle texture of her hair as it moves in the breeze. The natural light enhances her features, creating a soft, dreamy atmosphere.

A woman dressed in a nuns habit, with her breasts prominently displayed and adorned with band-aids, exudes a sense of confidence and playfulness. Her body movements are fluid and natural, capturing the essence of soft, realistic dynamics. The camera angle provides a close-up view, highlighting the detailed texture of her skin and the intricate details of her costume. The overall scene is vibrant and engaging, with the subjects expressive smile and dynamic poses adding a touch of humor and charm.

A woman in a nun costume stands confidently, her breasts jiggling and bouncing as she dances. The camera captures her from a medium close-up angle, emphasizing the realistic soft-body dynamics and the natural movements of her breasts. The detailed lighting highlights the texture of her skin and the white nun attire.

A mesmerizing sequence captures a woman in a poised, full-bodied pose against a vibrant yellow background. Her long, wavy black hair cascades gracefully around her, and her large, full breasts are accentuated with realistic soft-body dynamics, showcasing natural movements and jiggles as she moves. The camera angle is slightly tilted, enhancing the visual appeal of her form and the intricate details of her body, while the soft, smooth transitions and lighting highlight the natural curves and contours, making every movement appear lifelike and engaging.

In this captivating video, a woman with voluminous, curvy breasts and curly hair is captured in a series of dynamic poses, her body moving in a way that highlights the realistic soft-body dynamics and natural movements. The camera angle is low, emphasizing the size and texture of her breasts as she walks towards the camera, creating an intimate and engaging visual experience.

March 30, 2025 Update.

Due to increased demand, I've included the META within some of the videos. it wont be a regular thing but with this tool, hopefully it helps folks.

Additional training and inclusion of different camera motions.

Additional testing was completed, and there appears to be better prompt adherence as well.

Prompting:

These can be used directly or mixed as needed to trigger the LoRa effectively. It's important that you are detailed in your prompting. Try mixing them together and trying different seeds.

1. A smooth cinematic camera orbit around the subject.

2. Fluid motion maintaining steady distance and perspective shifts.

3. Uninterrupted rotation showcasing the subject from all angles.

4. Consistent-speed camera orbit enhancing spatial depth.

5. Graceful, immersive camera movement with no abrupt changes.

6. A cinematic zoom in. 

7. A cinematic zoom out. 

8. A cinematic pan. 

March 27, 2025 Update.

🚨This is an experimental LoRa 🚨

This LoRa is meant to be used with the 14B I2V Model of Wan2.1 and specializes in producing fluid panning shots and dynamic rotations of subjects, enhancing the storytelling capabilities of your visuals.

Previous LoRa's of this type tend to interfere with the overall context of the input image or require very specific images. The purpose of the experimental training was to be able to train the model to execute specific frame dynamics without interfering with the overall context of the image or prompt.

It is trained on just 5 carefully selected video clips over 3,000 training steps. Despite its lightweight training setup, the results are surprisingly effective, proving how much potential even small-scale training experiments can unlock.

Stay tuned and stay blessed.

I2V Image Selection Guide:

So far I have only tested this model with images of centrally placed figures, which can be either full-bodied portraits, landscapes, or close-ups. It doesn't seem to matter who or what it is; realistic or hyper-realistic photos tend to get good results. I have not tested this LoRa in conjunction with other LoRa's. If you do please share your results.

Prompting:

These can be used directly or mixed as needed to trigger the LoRa effectively.

1. A smooth cinematic camera orbit around the subject.

2. Fluid motion maintaining steady distance and perspective shifts.

3. Uninterrupted rotation showcasing the subject from all angles.

4. Consistent-speed camera orbit enhancing spatial depth.

5. Graceful, immersive camera movement with no abrupt changes.