ComfyUI Node: H3 Trajectory Bank (checkpoint every step)

H3 Trajectory Bank (checkpoint every step):

H3TrajectoryBank is a specialized node designed to enhance the sampling process by wrapping a sampler and saving the trajectory latent, which is the noisy state the sampler carries, after each step. This node is particularly useful for AI artists working with video clips, as it allows for the efficient storage of the sigma schedule and trajectory data, consuming approximately 7 MB per step for a 5-second 1024 clip. This means a full 25-step run can be banked under 200 MB. The primary advantage of using H3TrajectoryBank is its ability to pair with H3TrajectoryLoad, enabling users to branch from any saved step without the need to recompute the initial steps. This flexibility allows for modifications such as swapping the model, LoRA, guider, or remaining schedule, and continuing the process seamlessly, thus saving time and computational resources.

H3 Trajectory Bank (checkpoint every step) Input Parameters:

sampler

The sampler parameter is a critical component of the H3TrajectoryBank node, as it specifies the sampling method to be wrapped and utilized during the trajectory saving process. This parameter determines how the trajectory latent is generated and stored at each step. By selecting an appropriate sampler, you can influence the quality and characteristics of the output, making it essential to choose one that aligns with your artistic goals.

dump_dir

The dump_dir parameter specifies the directory path where the trajectory data will be saved. By default, it is set to "/tmp/h3_trajectory". This parameter allows you to organize and manage the storage location of your trajectory data, ensuring that it is easily accessible for future use or analysis. It is important to ensure that the specified directory has sufficient storage capacity to accommodate the data generated during the sampling process.

every_n

The every_n parameter controls the frequency at which the trajectory data is saved. It is an integer value with a default of 1, a minimum of 1, and a maximum of 25. This parameter allows you to specify that the trajectory should be saved every nth step, with the last step always being saved. Adjusting this parameter can help balance between storage usage and the granularity of saved data, depending on your specific needs and the complexity of the sampling process.

H3 Trajectory Bank (checkpoint every step) Output Parameters:

SAMPLER

The output of the H3TrajectoryBank node is a wrapped SAMPLER. This output retains the original functionality of the sampler while incorporating the additional capability of saving the trajectory latent at specified intervals. This enhanced sampler can be used in subsequent processes, allowing for the continuation of sampling from any saved step, thus providing flexibility and efficiency in managing complex sampling tasks.

H3 Trajectory Bank (checkpoint every step) Usage Tips:

• To optimize storage usage, adjust the every_n parameter to save trajectory data at intervals that balance detail with storage capacity, especially for longer video clips.
• Ensure that the dump_dir has sufficient storage space and is organized in a way that makes it easy to locate and manage saved trajectory data for future use or analysis.

H3 Trajectory Bank (checkpoint every step) Common Errors and Solutions:

"Directory not found: <specified_dump_dir>"

• Explanation: This error occurs when the specified directory for saving trajectory data does not exist or is inaccessible.
• Solution: Verify that the dump_dir path is correct and that the directory exists. Ensure that you have the necessary permissions to write to the specified location.

"Insufficient storage space in <specified_dump_dir>"

• Explanation: This error indicates that there is not enough storage space available in the specified directory to save the trajectory data.
• Solution: Free up space in the specified directory or choose a different directory with sufficient storage capacity for saving the trajectory data.