ComfyUI-TCD-scheduler

This repository introduces custom sampler nodes for ComfyUI that implement Trajectory Consistency Distillation (TCD) as described by Zheng et al. It includes the TCDScheduler, SamplerTCD Euler A, and SamplerTCD nodes, enhancing the sampling capabilities within ComfyUI.

• Offers TCDScheduler for advanced sampling techniques.
• Provides SamplerTCD Euler A as a reliable sampling option while SamplerTCD is still in development.
• Features a unique gamma parameter to adjust the level of stochasticity during sampling.

Context

The ComfyUI-TCD-scheduler is a collection of custom nodes designed to integrate TCD sampling methods into the ComfyUI framework. Its primary purpose is to enhance the quality and flexibility of image generation by allowing users to utilize advanced sampling strategies.

Key Features & Benefits

The tool introduces several practical features, including the TCDScheduler which supports fine-tuning of sampling processes. The gamma parameter allows users to control the randomness in the sampling steps, providing a spectrum from deterministic to fully stochastic outputs, which can significantly impact the final image quality.

Advanced Functionalities

One of the notable advanced capabilities is the gamma parameter in the LCMScheduler. By adjusting this parameter, users can manage the balance between predictability and randomness in the sampling process, which is crucial for achieving desired artistic effects or maintaining consistency in generated images.

Practical Benefits

By incorporating the ComfyUI-TCD-scheduler, users can improve their workflows with greater control over the sampling process, leading to enhanced image quality and efficiency. The ability to manipulate stochasticity allows for more tailored results, making it easier to achieve specific artistic goals.

Credits/Acknowledgments

The development of this tool has been supported by contributions from @laksjdjf, who assisted in adapting the sampling methods. The project is based on the research presented in Zheng et al.’s work on TCD, and users are encouraged to refer to external resources for a deeper understanding of samplers in Stable Diffusion.