Reactor: Face Swap and Restoration Model Assets

This repository provides 8 pre-trained model assets and weights for the ReActor face swap extension, maintained by Gourieff and updated in March 2026. The collection includes specialized files for face detection, swapping, and restoration, such as InsightFace and GFPGAN weights. It serves as the central asset hub for ComfyUI and Stable Diffusion web UI integrations.

Use Cases

• Executing face replacement tasks using the inswapper_128.onnx model
• Improving facial clarity in low-resolution images using the GFPGANv1.4.pth restoration model
• Implementing facial feature correction within ComfyUI workflows using the RestoreFormer_PP.onnx asset

Strengths

• Includes 8 distinct model files covering the full face-swap pipeline from detection to restoration
• Provides multiple versions of restoration models including GFPGAN v1.3 and v1.4
• Offers cross-platform compatibility through both ONNX and PTH file formats

Limitations

• Lack of raw training data prevents model fine-tuning or verification of training bias
• Mixed licensing across different model files (e.g., DeepInsight vs. TencentARC) creates complex compliance requirements

Provenance

Source

Aggregated from multiple research entities including DeepInsight, TencentARC, and sczhou

Collection Method

Curated collection of pre-trained weights from various facial recognition and restoration research projects

Freshness

Last updated March 2026, reflecting the most recent versions of the supported model architectures.