Skip to main content
Merge SeedVR2 Latents This node recombines sampled SeedVR2 latent temporal chunks into a single full-length latent. It uses a Hann window crossfade over the overlap region between chunks to create smooth transitions, or performs plain concatenation when no overlap is specified.

Inputs

Note: The temporal_overlap value must be greater than or equal to 0. The final chunk in the sequence may have fewer temporal frames than the other chunks. All chunks must have matching dimensions except for the temporal axis (T), and the first chunk must be 5-dimensional (B, C, T, H, W).

Outputs

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

Source fingerprint (SHA-256): 647653fc651462f62bcc9dca2919cfcaa4b476b4c679b0a1111698facab83f28