ComfyUI Node: MVEx Latent Mask To Mask

MVEx Latent Mask To Mask:

The MVEx_LatentMaskToMask node is designed to transform a latent-resolution mask back into its original pixel resolution, making it suitable for visualization or testing purposes. This node is particularly useful when you need to see the detailed effects of a mask that has been compressed into a latent space. By expanding each latent frame to paint every pixel frame in its group, the node ensures that the detail collapsed during the latent reduction process remains collapsed, preserving the integrity of the original mask's influence. This functionality is crucial for artists and developers who need to verify the accuracy and impact of masks in their projects, especially when working with video or image data that has undergone compression.

MVEx Latent Mask To Mask Input Parameters:

mask

This parameter represents the latent-resolution masks, with one mask provided per latent frame. It is essential for defining the areas of interest or influence within the latent space that need to be expanded back to pixel resolution. The mask parameter is crucial for ensuring that the correct regions are visualized or tested after the transformation process.

vae

The VAE (Variational Autoencoder) input is optional and is used when the compression method is set to auto. It specifies the VAE whose latents the mask corresponds to, ensuring that the transformation process aligns with the specific characteristics of the VAE model used. This alignment is important for maintaining the accuracy of the mask expansion, especially when dealing with different VAE configurations.

spatial

This integer parameter defines the number of pixels per latent pixel on each axis, with a default value of 8 and a minimum of 1. It determines the level of detail retained during the expansion process, affecting how finely the latent mask is translated back into pixel space. Adjusting this parameter allows you to control the granularity of the expanded mask.

token_spatial

This integer parameter specifies the number of latent pixels per model token on each axis, with a default value of 1 and a minimum of 1. It influences how the mask is unified over each token block, ensuring that the strongest value in each block paints the entire block. This parameter is particularly relevant for models like MiniMax H3, where the mask is read per token block.

head_frames

This integer parameter indicates the number of pixel frames in the model's leading frame group, with a default value of 1 and a minimum of 0. It is used to define the structure of the frame groups, particularly for models like Wan, Hunyuan, and LTX, which have a leading group. Setting this parameter to 0 indicates that the model has no leading group.

head_latents

This integer parameter specifies the number of latent frames produced by the leading group, with a default value of 1 and a minimum of 0. It is crucial for determining the output structure of the latent frames, especially for models with a defined leading group like Wan, Hunyuan, and LTX.

chunk_frames

This integer parameter defines the number of pixel frames in each repeating group after the leading group, with a default value of 4 and a minimum of 1. It controls the temporal structure of the frame groups, with different values for models like Wan, Hunyuan, and LTX. Setting this parameter to 1 disables temporal reduction, making it suitable for image models.

chunk_latents

This integer parameter indicates the number of latent frames produced by each repeating group, with a default value of 1 and a minimum of 1. It is essential for defining the temporal output structure of the latent frames, particularly for models like Wan, Hunyuan, and LTX.

MVEx Latent Mask To Mask Output Parameters:

output_mask

The output of this node is the expanded mask in pixel resolution. This output is crucial for visualizing or testing the effects of the latent mask in its original resolution, allowing you to see the detailed impact of the mask on the pixel data. The expanded mask retains the collapsed detail from the latent space, ensuring that the original mask's influence is accurately represented.

MVEx Latent Mask To Mask Usage Tips:

• Ensure that the mask input is correctly aligned with the latent frames to achieve accurate expansion results.
• Adjust the spatial parameter to control the level of detail in the expanded mask, balancing between granularity and performance.
• Use the vae input when the compression method is set to auto to ensure proper alignment with the VAE model used.

MVEx Latent Mask To Mask Common Errors and Solutions:

Mask reduction on device failed

• Explanation: This error occurs when the mask reduction process fails on the specified device, possibly due to memory limitations or missing operations.
• Solution: The node automatically retries the operation on the CPU, which is slower but more reliable. Ensure that your system has sufficient resources, or consider reducing the complexity of the input data.

Incorrect mask alignment

• Explanation: This issue arises when the input mask does not match the expected latent frames, leading to inaccurate expansion results.
• Solution: Verify that the mask input is correctly aligned with the latent frames and matches the VAE model's configuration if applicable. Adjust the input parameters as needed to ensure proper alignment.