TRStart creating
Menu

Wan VACE 14B

VACE is a video generation model that uses a source image, mask, and video to create prompted videos with controllable sources.

Input

Fields marked * are required. Review the price before generating.

The text prompt to guide video generation. For pose task, the prompt should describe the desired pose and action of the subject in the video.

Resolution of the generated video.

Aspect ratio of the generated video.

URL to the source video file. Required for pose task.

Uploading a file needs an account. Create accountMP4, WebM · up to 100 MB

URLs to source reference image. If provided, the model will use this image as reference.

Uploading a file needs an account. Create accountJPEG, PNG, WebP · up to 25 MB

URL to the first frame of the video. If provided, the model will use this frame as a reference.

Uploading a file needs an account. Create accountJPEG, PNG, WebP · up to 25 MB

URL to the last frame of the video. If provided, the model will use this frame as a reference.

Uploading a file needs an account. Create accountJPEG, PNG, WebP · up to 25 MB

Additional settings

Negative prompt for video generation.

Number of frames to generate. Must be between 81 to 241 (inclusive).

Frames per second of the generated video. Must be between 5 to 30. Ignored if match_input_frames_per_second is true.

Random seed for reproducibility. If None, a random seed is chosen.

Number of inference steps for sampling. Higher values give better quality but take longer.

Guidance scale for classifier-free guidance. Higher values encourage the model to generate images closely related to the text prompt.

Sampler to use for video generation.

Shift parameter for video generation.

Whether to enable prompt expansion.

Whether to preprocess the input video.

Tips and requirements

Acceleration to use for inference. Options are 'none' or 'regular'. Accelerated inference will very slightly affect output, but will be significantly faster.

The quality of the generated video.

The write mode of the generated video.

Number of frames to interpolate between the original frames. A value of 0 means no interpolation.

Tips and requirements

Temporal downsample factor for the video. This is an integer value that determines how many frames to skip in the video. A value of 0 means no downsampling. For each downsample factor, one upsample factor will automatically be applied.

Tips and requirements

If true, the model will automatically temporally downsample the video to an appropriate frame length for the model, then will interpolate it back to the original frame length.

Tips and requirements

The minimum frames per second to downsample the video to. This is used to help determine the auto downsample factor to try and find the lowest detail-preserving downsample factor. The default value is appropriate for most videos, if you are using a video with very fast motion, you may need to increase this value. If your video has a very low amount of motion, you could decrease this value to allow for higher downsampling and thus longer sequences.

The model to use for frame interpolation. Options are 'rife' or 'film'.

Tips and requirements

The transparency mode to apply to the first and last frames. This controls how the transparent areas of the first and last frames are filled.

If true, also return a ZIP file containing all generated frames.

Price$0.1268 / second

You sign up first. Your settings are kept.

API code for this input

Complete the required fields and fix input errors to see API code.

How to use Wan VACE 14B

Getting started

Explore a character or costume while using an existing movement sequence as guidance.

Review your result

Compare limb positions, foot contact and timing; check identity consistency between frames.

Inputs and settings

Supply video_url and prompt. Check preprocess for your guidance source; ref_image_urls can provide appearance references.

Choose resolution and frames_per_second for the shot. Keep seed fixed while refining guidance_scale; avoid adding interpolation until the motion is usable.

We use analytics cookies. Privacy policy