Question about SVD

[justinday123]

Thank you for your nice work. I have some questions about SVD. Original SVD can generate videos using text + image. But in your code, I can't find about using text. Did I lost it? Thanks.

[hadipash]

While the original SVD supports text-to-video generation, Stability AI released only an image-to-video model to the public. As a result, text-to-video functionality is absent in our code.

[justinday123]

Thanks for your response. You mean it's not able to generate video using image+text in SVD? Thanks.

[hadipash]

You can generate it in two steps: first with SDXL for text-to-image, and then with SVD for image-to-video.

[justinday123]

Thanks again. My aim was to use text for motion guidance such as
{Character image} + {Text: "a man swiming"}.
Is it unavailable?

[hadipash]

Not for SVD, you would need to use another model, such as HunyuanVideo-I2V or Wan 2.1.