Is there any guidance on prompt word input?

[mimimi999]

This model can generate higher quality speech than other models, but sometimes unintended sounds are generated, perhaps due to my poor prompt input.

I'm not sure what kind of prompts are being added to train this model, so is there a manual guide for inputting prompts for this generative model?

[SzczesnyS]

Thank you for your attention. Please ensure that your prompts describe the sounds and corresponding details clearly, without excessive description of the video content.

To address the issue of unintended sounds, we have updated the support for negative prompts, which will help reduce the occurrence of unwanted audio (e.g., negative prompt: “background music” can effectively suppress the generation of music).

[mimimi999]

Using negative prompts has reduced the amount of unnecessary music and other content that plays.

However, it seems like the generated quality drops if the video is extremely short. Does this mean that to maintain quality, it is necessary to generate videos of a certain length?

[SzczesnyS]

The optimal video duration is ​8-10 seconds. Videos that are significantly longer or shorter may result in reduced performance.