ComfyUI-EulerDiscreteScheduler  by erosDiffusion

ComfyUI scheduler for high-fidelity image synthesis

Created 1 month ago
307 stars

Top 87.5% on SourcePulse

GitHubView on GitHub
Project Summary

This project provides the FlowMatch Euler Discrete Scheduler for ComfyUI, addressing the lack of direct exposure for this advanced scheduler used in official diffusers demos. It enables ComfyUI users to generate sharper, noise-free images with greater control over the sampling process, targeting users seeking enhanced image quality and experimental flexibility.

How It Works

The project integrates the FlowMatchEulerDiscreteScheduler, known for its use in diffusers' Z-Image demo, into ComfyUI's KSampler. It offers a custom node that exposes all scheduler parameters, allowing users to fine-tune sampling behavior and output SIGMAS for use with the SamplerCustom node, facilitating advanced configurations for potentially higher-resolution, noise-free outputs.

Quick Start & Requirements

Highlighted Details

  • Adds FlowMatchEulerDiscreteScheduler to ComfyUI's KSampler.
  • Provides an experimental custom node for detailed parameter control and SIGMAS output.
  • Aims for sharper, noise-free images, particularly with configurations like "9 steps, big res, zero noise."
  • Supports start and end steps for sampling control (added in v1.0.4).

Maintenance & Community

The repository is maintained by erosDiffusion. No specific community links (Discord, Slack) or details on major contributors/sponsorships are provided in the README.

Licensing & Compatibility

The project is released under the MIT License. This license generally permits commercial use and modification.

Limitations & Caveats

The custom scheduler node is experimental. Users may encounter incompatibility issues with the res4lyf package, requiring workarounds such as connecting SIGMAS to a basicScheduler node or disabling res4lyf. Specific peft package versions might be necessary to avoid installation failures.

Health Check
Last Commit

1 month ago

Responsiveness

Inactive

Pull Requests (30d)
0
Issues (30d)
5
Star History
146 stars in the last 30 days

Explore Similar Projects

Starred by Vincent Weisser Vincent Weisser(Cofounder of Prime Intellect), Patrick von Platen Patrick von Platen(Author of Hugging Face Diffusers; Research Engineer at Mistral), and
2 more.

IP-Adapter by tencent-ailab

0.2%
6k
Adapter for image prompt in text-to-image diffusion models
Created 2 years ago
Updated 1 year ago
Feedback? Help us improve.