Issue: Camera Zoom-Out Not Working Regardless of Prompt & Guidance Scale

[jeff999955]

First of all, thank you for your amazing work on this video diffusion model! I've been experimenting with generating videos using various prompts and guidance scales, but I've encountered a consistent issue:

No matter how explicitly I specify that the camera should zoom out (e.g., "zoom out," "pull away," "get away from the scene"), the model always generates videos where the camera zooms in or gets closer to the scene instead.

Steps to Reproduce

• Tried multiple variations of prompts emphasizing camera zooming out.
• Adjusted guidance scale, but it did not affect the behavior.
• The model still consistently generates a zoom-in effect.

Prompt:

This is a (motionless) video of a (static scene). The camera (zooms out) of the scene. (no moving objects, static scene, only camera movement). The video showcases a bedroom with a central focus on a bed adorned with various pillows and a dark-colored bedspread. The room is illuminated by natural light remain still from a window draped with dark curtains. A ceiling fan with a unique design hangs from the ceiling, and a mirror with a dark frame is remain still on the right side of the video. The walls are painted in a light beige color, and there’s a white door on the left. The flooring appears to be made of wooden planks. Overall, the room exudes a warm and cozy ambiance.

Expected Behavior The generated video should depict the camera zooming out from the scene, revealing a wider view.

Actual Behavior Instead of zooming out, the model consistently generates videos where the camera zooms in or gets closer to the scene.

Questions

• Is this a known limitation of the model?
• Did this issue occur during development?
• Are there any recommended ways to tweak the prompt or parameters to achieve a zoom-out effect?

Thanks in advance for any insights! I really appreciate your work on this project.

[Zueuk]

it tends to zoom in regardless of the prompt