General Models

Models optimized for general photographic images and natural scenes

General models are designed for natural images, photographs, and real-world content. These models excel at upscaling portraits, landscapes, and everyday photography.

Available Models

RealESRGAN_x4plus

Primary model with best quality for 4x upscaling

RealESRGAN_x2plus

Optimized for 2x upscaling

realesr-general-x4v3

Lightweight model for fast processing

RealESRNet_x4plus

MSE-based model for smooth results

Model Specifications

RealESRGAN_x4plus

Recommended: This is the flagship model for general images, providing the best quality for 4x upscaling.

Property	Value
Scale	4x
Architecture	RRDBNet
RRDB Blocks	23
Features	64
Growth Channels	32
Download	RealESRGAN_x4plus.pth

# Download and use
wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth -P weights
python inference_realesrgan.py -n RealESRGAN_x4plus -i inputs -o outputs

RealESRGAN_x2plus

Property	Value
Scale	2x
Architecture	RRDBNet
RRDB Blocks	23
Features	64
Growth Channels	32
Download	RealESRGAN_x2plus.pth

# Download and use
wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.1/RealESRGAN_x2plus.pth -P weights
python inference_realesrgan.py -n RealESRGAN_x2plus -i inputs -o outputs

realesr-general-x4v3

Lightweight: This model consumes much less GPU memory and processing time, making it ideal for resource-constrained environments.

Property	Value
Scale	1x, 2x, 3x, or 4x (variable)
Architecture	SRVGGNetCompact
Convolution Layers	32
Features	64
Size	Small (S)
Download	realesr-general-x4v3.pth

# Download and use with custom scale
wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth -P weights
python inference_realesrgan.py -n realesr-general-x4v3 -i inputs -s 2  # 2x upscaling

Denoise Strength Control

The realesr-general-x4v3 model supports adjustable denoise strength:

# Weak denoise (keep more original texture)
python inference_realesrgan.py -n realesr-general-x4v3 -i inputs -dn 0.2

# Strong denoise (smoother result)
python inference_realesrgan.py -n realesr-general-x4v3 -i inputs -dn 0.8

The -dn parameter ranges from 0 (weak denoise) to 1 (strong denoise). Default is 0.5.

RealESRNet_x4plus

This model uses MSE loss instead of GAN training, resulting in over-smooth effects. Use this if you prefer less sharp, more natural-looking results.

Property	Value
Scale	4x
Architecture	RRDBNet
RRDB Blocks	23
Training Loss	MSE (no GAN)
Download	RealESRNet_x4plus.pth

wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.1/RealESRNet_x4plus.pth -P weights
python inference_realesrgan.py -n RealESRNet_x4plus -i inputs -o outputs

Legacy Model

ESRGAN (Official)

The original ESRGAN model trained on DF2K and OST datasets:

Property	Value
Scale	4x
Download	ESRGAN_SRx4_DF2KOST_official-ff704c30.pth

Discriminator Models

Discriminator models are used for fine-tuning and training. You don’t need these for inference.

Discriminator	Corresponding Generator Model
RealESRGAN_x4plus_netD.pth	RealESRGAN_x4plus
RealESRGAN_x2plus_netD.pth	RealESRGAN_x2plus

Model Comparison

Quality vs Speed

Quality:  RealESRGAN_x4plus > RealESRGAN_x2plus > realesr-general-x4v3 > RealESRNet_x4plus
Speed:    realesr-general-x4v3 > RealESRNet_x4plus > RealESRGAN_x2plus > RealESRGAN_x4plus
Memory:   realesr-general-x4v3 < RealESRNet_x4plus < RealESRGAN_x2plus < RealESRGAN_x4plus

Use Case Recommendations

Best Quality

RealESRGAN_x4plus

Use for final production work where quality is paramount.

Fast Processing

realesr-general-x4v3

Use for batch processing or when speed matters.

2x Upscaling

RealESRGAN_x2plus

Use when you only need 2x resolution increase.

Smooth Results

RealESRNet_x4plus

Use for artistic or stylized results without sharpening.

Usage Examples

Basic Inference

# Using RealESRGAN_x4plus (recommended)
python inference_realesrgan.py -n RealESRGAN_x4plus -i inputs

# Using lightweight model
python inference_realesrgan.py -n realesr-general-x4v3 -i inputs

# Custom output scale
python inference_realesrgan.py -n RealESRGAN_x4plus -i inputs -s 2  # 2x instead of 4x

Advanced Options

# With tiling for large images
python inference_realesrgan.py -n RealESRGAN_x4plus -i inputs -t 400

# With face enhancement
python inference_realesrgan.py -n RealESRGAN_x4plus -i inputs --face_enhance

# Custom output folder and format
python inference_realesrgan.py -n RealESRGAN_x4plus -i inputs -o results --ext jpg

Next Steps

Anime Models

Learn about anime-specific models

Video Models

Explore video upscaling models