Sigma Curves:
SigmaCurves is a sophisticated node designed to provide a step-locked sigma editing capability within the ComfyUI framework. Its primary function is to allow you to manipulate the sigma values used in the sampling process of AI models, which are crucial for controlling the noise levels during image generation. The node plots one point per sampling step, with the y-axis representing the normalized sigma at that step, allowing you to visually adjust and fine-tune the sigma curve. By selecting a base scheduler, SigmaCurves seeds the plot with the scheduler's natural shape, which you can then modify by dragging individual y-values or applying curve types to specific ranges. This flexibility enables you to customize the noise schedule to achieve desired artistic effects, making it a powerful tool for AI artists looking to have more control over the image generation process.
Sigma Curves Input Parameters:
model
The model parameter specifies the AI model that will be used in conjunction with the SigmaCurves node. It is essential for determining the sigma values as different models may have different characteristics and requirements for sigma adjustments. This parameter does not have a specific range of values as it depends on the available models in your environment.
scheduler
The scheduler parameter determines the base scheduling algorithm used to seed the sigma curve. It influences the initial shape of the sigma plot and can be adjusted to match the characteristics of different models. The choice of scheduler can significantly impact the final output, as it sets the foundation for further sigma adjustments.
steps
The steps parameter defines the number of sampling steps to be used in the sigma curve. It directly affects the granularity of the sigma adjustments, with more steps allowing for finer control over the noise schedule. The minimum value is typically 1, and there is no strict maximum, but it should be set according to the desired level of detail and computational resources available.
denoise
The denoise parameter controls the level of noise reduction applied during the sampling process. It ranges from 0.0 to 1.0, where 0.0 means no denoising and 1.0 means full denoising. Adjusting this parameter allows you to balance between preserving details and reducing noise in the generated images.
curve_data
The curve_data parameter is a JSON object containing the sigma values and other related data. It allows you to input custom sigma curves, providing a way to override the default scheduler output. If the curve_data is empty or malformed, the node defaults to using the real scheduler output, ensuring that the node behaves like a regular scheduler until you start editing.
Sigma Curves Output Parameters:
sigmas
The sigmas output parameter provides the final sigma values after all adjustments have been made. These values are crucial for the sampling process, as they determine the noise levels at each step. The output is typically a list or tensor of sigma values that can be used in subsequent nodes or processes to generate images with the desired characteristics.
Sigma Curves Usage Tips:
- Experiment with different schedulers to see how they affect the initial sigma curve and choose one that aligns with your artistic goals.
- Use the denoise parameter to fine-tune the balance between detail preservation and noise reduction, especially when working with high-resolution images.
- If you have a specific noise schedule in mind, use the curve_data parameter to input your custom sigma values and achieve precise control over the image generation process.
Sigma Curves Common Errors and Solutions:
Sigma Curves: invalid curve_data; falling back to scheduler.
- Explanation: This error occurs when the curve_data parameter is empty or contains malformed JSON, preventing the node from using the custom sigma values.
- Solution: Ensure that the curve_data parameter contains valid JSON with the correct structure and values. Double-check for syntax errors or missing fields in the JSON object.
Sigma Curves: calculate_sigmas failed for '<scheduler>': <error_message>
- Explanation: This error indicates that the calculation of sigma values failed for the specified scheduler, possibly due to an unsupported scheduler or an issue with the model.
- Solution: Verify that the chosen scheduler is compatible with the model and that all required inputs are correctly specified. If the problem persists, consider using a different scheduler or model.