Skip to main content

​
Overview

Generate audio that matches given video content + prompt using ThinkSound.
  • Output Type: video
  • Estimated Cost: 0.5 * duration credits
  • Handler: replicate

​
Parameters

​
Required Parameters

​
video
video
required
URL of the base video

​
Optional Parameters

​
caption
string
Brief description of the video content
A short caption describing what’s happening in the video to help the model understand the context.
  • Label: Caption
​
cot
string
Detailed description of the sound generation process
A detailed description that begins with the sound process and includes texture, atmosphere, and timing details. This helps the model understand exactly what audio to generate and how it should match the video.
  • Label: Chain of thought
​
cfg_scale
float
default:5
Classifier-free guidance scale
Controls how closely the model follows the prompt. Higher values mean stricter adherence to the prompt.
  • Label: CFG Scale
  • Minimum: 1
  • Maximum: 20
​
num_inference_steps
integer
default:24
Number of inference steps
More steps generally produce higher quality but take longer to generate.
  • Label: Inference steps
  • Minimum: 10
  • Maximum: 50