Skip to main content
This node sets context windows for LTXV-like models during sampling. It divides the video generation process into overlapping windows to manage memory usage and improve temporal coherence.

Inputs

ParameterDescriptionData TypeRequiredRange
modelThe model to apply context windows to during sampling.MODELYes-
context_lengthThe length of the context window in real frames. Must be 8*n + 1. (default: 145)INTYesMinimum: 1
Maximum: nodes.MAX_RESOLUTION
Step: 8
context_overlapThe overlap of the context window in real frames. (default: 40)INTYesMinimum: 0
Step: 8
context_scheduleStep-dependent scheduling algorithm for context windows. (default: UNIFORM_STANDARD)COMBOYesSTATIC_STANDARD
UNIFORM_STANDARD
UNIFORM_LOOPED
BATCHED
context_strideThe stride of the context window; only applicable to uniform schedules. (default: 1)INTNoMinimum: 1
closed_loopWhether to close the context window loop; only applicable to looped schedules. (default: False)BOOLEANNoTrue
False
fuse_methodThe method to use to fuse the context windows. (default: PYRAMID)COMBOYesOptions from comfy.context_windows.ContextFuseMethods.LIST_STATIC
freenoiseWhether to apply FreeNoise noise shuffling, improves window blending. (default: True)BOOLEANNoTrue
False
retain_first_frameRetain the first latent frame in every context window (may help retain initial reference). (default: False)BOOLEANNoTrue
False
split_conds_to_windowsWhether to split multiple conditionings (created by ConditionCombine) to each window based on region index. (default: False)BOOLEANNoTrue
False
Note: The context_length parameter must follow the formula 8*n + 1, where n is a positive integer. The node automatically adjusts the value to meet this requirement by converting real frames to latent frames. The context_overlap is also converted from real frames to latent frames (divided by 8).

Outputs

Output NameDescriptionData Type
MODELThe model with context windows applied for sampling.MODEL
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): ad0b8c54acaab1853f6fe98dbad675478f033caf867df954b40b7db5208f5ae4