This repository provides the official ComfyUI interface for HelloMeme, a diffusion model integration that embeds high-level and fidelity-rich conditions for image and video generation. It is designed for users of ComfyUI looking to leverage advanced conditional generation techniques for creative media production.

How It Works

HelloMeme integrates "Spatial Knitting Attentions" into diffusion models, allowing for the embedding of complex, high-level conditions. This approach enhances the fidelity and control over generated content, particularly in video generation tasks where maintaining consistency and detail is crucial. The system is built as a ComfyUI node, enabling seamless integration into existing workflows.

Quick Start & Requirements

• Install via ComfyUI custom nodes.
• Requires ComfyUI, Python 3.10+, and specific dependencies managed by ComfyUI.
• Recommended checkpoints include realisticVisionV60B1_v51VAE and disneyPixarCartoon_v10.
• Example workflows are provided in the workflows directory.

Highlighted Details

• Supports both image and video generation.
• HelloMemeV2 offers improved expression consistency, better checkpoint compatibility, and reduced VRAM usage.
• Includes HMControlNet2 for facial expression extraction and a super-resolution feature.
• Optimized for lower VRAM usage, with specific workflows tested on 2080Ti.

Maintenance & Community

• Developed by HelloVision/HelloGroup Inc.
• Regular updates with new features and optimizations (e.g., V3, Modelscope support).
• YouTube demos are linked for new features.

Licensing & Compatibility

• License details are not explicitly stated in the README.
• Compatibility with third-party checkpoints is noted as improved in V2.

Limitations & Caveats

The README mentions that face proportion in the reference image significantly affects generation quality, and provides a CropReferenceImage node to address this. Specific VRAM requirements may vary depending on the chosen workflow and model.