Replicate is joining Cloudflare
ctrl+k

pbarker/gfpgan-video

GFPGAN for human face video upscaling

Public
33.5K runs
License
GitHub
Paper
Playground API Examples README Versions

Input

*file
Input
string
GFPGAN version. v1.3: better quality. v1.4: more details and better identity.

Default: "v1.4"

number
Rescaling factor

Default: 2

Sign in to run this model

Each run costs approximately $0.27. Alternatively, try out these models for free.

Sign in with GitHub

By signing in, you agree to our
terms of service and privacy policy

Output

Generated in
Tweak itIterate in playgroundReportView full prediction

Examples

View more

Run time and cost

This model costs approximately $0.27 to run on Replicate, or 3 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia L40S GPU hardware. Predictions typically complete within 5 minutes. The predict time for this model varies significantly based on the inputs.

Readme

This model doesn't have a readme.

real-esrgan

nightmareai/real-esrgan

Upscale images with adjustable scale (up to 10x) and optional face restoration. Takes a single image as input and returns an upscaled image, with GFPGAN-based face enhancement to improve facial details when enabled. Recommended maximum input resolution is 1440p.

76.7m runs
Official
gfpgan

tencentarc/gfpgan

Restore faces in old, low-quality, or AI-generated photos from a single input image, returning an enhanced image with improved identity fidelity. Optionally upscale the output by a scale factor. Choose between versions v1.2, v1.3 (better quality), v1.4 (more detail and identity), or RestoreFormer to balance detail and identity preservation.

98.9m runs
codeformer

sczhou/codeformer

Restore and enhance faces in old or AI-generated photos. Takes an image and returns an improved image, recovering facial details with CodeFormer while optionally enhancing the background via Real-ESRGAN and applying configurable upscaling. Provides dedicated face upsampling for high-resolution outputs and a fidelity control (0–1) to balance identity preservation and visual quality.

45.9m runs
gfpgan

xinntao/gfpgan

Restore and enhance faces in photos. Accepts a single image and outputs a restored image with sharper facial details, reduced artifacts, and preserved identity, useful for old photographs and AI-generated faces, with optional upscaling. Supports versions v1.2, v1.3, v1.4, and RestoreFormer; v1.3 emphasizes quality while v1.4 adds more detail and better identity consistency.

43.4m runs
codeformer-video

pollinations/codeformer-video

Restore and enhance faces in videos. Takes a video as input and outputs an enhanced video using CodeFormer-based face restoration with adjustable upscaling, optional face upsampling for high-resolution or AI-generated faces, and optional Real-ESRGAN background enhancement. Improve old, low-quality, or artifacted footage while preserving temporal consistency.

526 runs
video-upscale

topazlabs/video-upscale

Upscale videos to 720p, 1080p, or 4K and convert frame rate between 15–120 fps. Accepts a video input and outputs an upscaled video, enabling resolution enhancement and frame rate conversion for delivery to modern displays or restoration workflows.

759.6k runs
Official
real-esrgan-video

lucataco/real-esrgan-video

Upscale videos to FHD, 2K, or 4K. Accepts a video input and returns an upscaled video using Real-ESRGAN. Choose between RealESRGAN_x4plus for general footage, or anime-focused RealESRGAN_x4plus_anime_6B and realesr-animevideov3.

284.3k runs
keep

zsxkib/keep

Restore and upscale faces in videos. Accepts a video input and outputs an enhanced video with temporally consistent face super-resolution using KEEP (Kalman-Inspired Feature Propagation). Supports flexible face detection (retinaface_resnet50, retinaface_mobile0.25, yolov5n, yolov5l), optional background enhancement and face upsampling via RealESRGAN, and options to process only the center face or draw bounding boxes. Works with pre-aligned 512x512 face crops or unaligned footage, and can preserve or override output FPS.

108 runs