Interesting observations in the generated files

[yuanxion]

ControlNet

We can get some video like edge, depth or pose, and then pass it to stable diffusion to control the video generation.

For a funny testing, I pass the depth video of a fox to Stable Diffusion,

and then ask it to generate a video with text prompt "oil painting of a deer, a high-quality, detailed, and professional photo", then I get a foxy deer, which looks like this:

[wangleflex]

Model: Text to video model with default inputs
Summary: When promts include color, tone and hue of video will be impacted, but when color is white, results are better than other colors.

some examples:

1. Promt: a brown bird is flying under blue sky
   Video snapshot:
   

2. Promt: a red bird is flying under blue sky
   Video snapshot:
   

3.Promt: a white bird is flying under blue sky
Video snapshot:

4.Promt: a white cat with blue eye is eating a small fish
Video snapshot:

5.Promt: a cute dog is playing with a blue ball
Video snapshot:

6.Promt: a cute dog is playing with a red ball
Video snapshot:

7.Promt: a cute white dog is running on street
Video snapshot: