ComfyUI Node: Anima Booster Loader (BSS)

Anima Booster Loader (BSS):

The AnimaBoosterLoader is a specialized node designed to optimize the loading of Anima DiT models within the ComfyUI framework. Its primary purpose is to enhance the performance of these models by leveraging advanced techniques such as SageAttention and torch.compile. This node is particularly beneficial for users who work with the Anima DiT 2B model, as it ensures maximum performance optimization. The AnimaBoosterLoader achieves this by automatically detecting and applying the appropriate data type, specifically bfloat16, which is crucial for maintaining the integrity of the model's output. By doing so, it prevents common issues such as black images that occur due to value range mismatches when forcing fp16. Additionally, the node integrates seamlessly with ComfyUI's standard pipeline, mirroring the functionality of the built-in UNETLoader, while offering enhanced speed and efficiency through its unique optimizations.

Anima Booster Loader (BSS) Input Parameters:

model_name

The model_name parameter specifies the name of the Anima DiT model you wish to load. This parameter is crucial as it determines which model file will be accessed and loaded into the system. The model is expected to be stored in the bfloat16 format, and the loader will automatically handle this data type to ensure compatibility and optimal performance. There are no explicit minimum or maximum values for this parameter, but it must correspond to a valid model file name within the designated directory.

sage_attention

The sage_attention parameter controls the application of the SageAttention optimization to the model. This parameter can significantly impact the model's execution speed and efficiency. Options for this parameter include different modes of SageAttention or disabling it entirely. The default setting is typically to apply SageAttention unless specified otherwise. This parameter allows users to fine-tune the model's performance based on their specific needs and the capabilities of their hardware.

Anima Booster Loader (BSS) Output Parameters:

model

The model output parameter represents the loaded and optimized Anima DiT model. This output is crucial as it is the primary object that will be used for further processing and inference tasks. The model is returned in a state that is ready for immediate use, with all necessary optimizations applied, ensuring that it operates at peak efficiency. Users can expect this output to be fully compatible with the ComfyUI framework and any subsequent nodes or processes that require a model input.

Anima Booster Loader (BSS) Usage Tips:

• Ensure that the model files are stored in the correct directory and named appropriately to avoid loading errors.
• Utilize the sage_attention parameter to enhance model performance, especially if you are working with large datasets or require faster processing times.
• Avoid forcing the model to use fp16 as this can lead to incorrect outputs; rely on the default bfloat16 setting for best results.

Anima Booster Loader (BSS) Common Errors and Solutions:

Failed to load nodes: <error_message>

• Explanation: This error indicates that there was an issue loading one or more nodes, possibly due to missing files or incorrect configurations.
• Solution: Verify that all necessary files are present in the correct directories and that the configurations are set up correctly. Check the logs for specific error messages that can guide you to the root cause.

SageAttention is not installed or failed to load

• Explanation: This message suggests that the SageAttention module is either not installed or encountered an error during loading.
• Solution: Ensure that the SageAttention module is correctly installed and accessible. If the problem persists, consider disabling SageAttention to use the native PyTorch Scaled Dot Product Attention as a fallback.

torch.compile failed: <error_message>

• Explanation: This error occurs when the torch.compile function fails to optimize the model, possibly due to compatibility issues.
• Solution: Check the compatibility of your PyTorch version with the torch.compile feature. If necessary, update PyTorch or adjust the model settings to bypass the compilation step.