ComfyUI Node: Diversity Boost

Diversity Boost:

DiversityBoostCore is designed to enhance the composition diversity of distilled models, which are often used in AI art generation. This node focuses on restoring the richness and variety in compositions by employing a method known as HF attenuation combined with a DCT composition push at the initial step. The primary goal of this node is to counteract the loss of diversity that can occur during the distillation process, ensuring that the generated outputs maintain a high level of detail and complexity. By integrating these techniques, DiversityBoostCore helps in achieving more dynamic and varied artistic outputs, making it an essential tool for artists looking to push the boundaries of AI-generated art.

Diversity Boost Input Parameters:

model

This parameter represents the AI model that will be used for generating art. It serves as the foundation upon which the diversity enhancements will be applied.

strength

The strength parameter controls the amplitude of the composition push. It ranges from 0.0 to 2.0, with a default value of 0.50. A value of 0 means only cleanup is performed, while 0.5 provides a moderate push, and 1.0 results in a strong push. Adjusting this parameter affects how aggressively the node enhances diversity in the composition.

clamp

The clamp parameter is a safety measure for DCT field values, with a range from 0.1 to 3.0 and a default of 1.0. It ensures that the values do not exceed a certain threshold, preventing potential artifacts or distortions in the output.

noise_type

This parameter allows you to select the frequency spectrum of random DCT coefficients, with options including "pink," "white," and "blue." The default is "pink," which is recommended for a stronger composition push. The choice of noise type influences the texture and complexity of the generated art.

n_periods

The n_periods parameter, ranging from 1 to 10 with a default of 2, determines the Butterworth cutoff. A value of 2 is typically used to preserve the DCT signal, affecting the smoothness and continuity of the composition.

dc_preserve

This parameter, ranging from 0.0 to 1.0 with a default of 0.0, controls the preservation of DC amplitude. A value of 0 maximizes diversity, while higher values retain more of the original amplitude, impacting the overall balance between diversity and fidelity.

energy_compensate

The energy_compensate parameter is a boolean option that, when enabled, rescales the output energy to match the original. This helps maintain consistency in the overall energy level of the composition, ensuring that the enhancements do not lead to unintended changes in brightness or contrast.

Diversity Boost Output Parameters:

model

The output model is the enhanced version of the input model, now with improved composition diversity. This output retains the original model's characteristics while incorporating the diversity enhancements, resulting in more varied and dynamic artistic outputs.

Diversity Boost Usage Tips:

• Experiment with the strength parameter to find the right balance between diversity and fidelity for your specific artistic goals.
• Use the noise_type parameter to explore different textures and complexities in your compositions, with "pink" being a good starting point for most cases.
• Adjust the clamp parameter if you notice any artifacts or distortions in the output, ensuring that the values remain within a safe range.

Diversity Boost Common Errors and Solutions:

ValueError: Invalid strength value

• Explanation: This error occurs when the strength parameter is set outside its allowed range of 0.0 to 2.0.
• Solution: Ensure that the strength value is within the specified range and adjust it accordingly.

TypeError: Unsupported noise type

• Explanation: This error arises when an invalid option is selected for the noise_type parameter.
• Solution: Choose a valid option from the available choices: "pink," "white," or "blue."

RuntimeError: Model output mismatch

• Explanation: This error indicates a discrepancy between the expected and actual output model.
• Solution: Verify that the input model is compatible with the node and that all parameters are correctly configured.