KEEP: Kalman-Inspired Face Super-Resolution

Discover KEEP for advanced video face super-resolution. Ready to experience the power of AI? Start your journey here!

Platform: Replicate

Video Face EnhancementFace Super-ResolutionVideo Restoration

68 runs

L40S

Commercial

🚀Function Overview

A deep learning model that enhances video quality by super-resolving faces using Kalman-inspired feature propagation, with options to restore background and upscale facial features for coherent frame-to-frame results.

Key Features

Uses KEEP algorithm for temporal-consistent face super-resolution
Supports facial landmark detection and alignment
Optional RealESRGAN integration for face/background enhancement
Configurable batch processing for long videos
Face cropping and alignment controls for pre-processed videos

Use Cases

•Restoring old or low-resolution personal videos
•Enhancing video conference/streaming quality
•Improving facial clarity in archival footage
•Pre-processing for AI-based video analysis
•Upscaling social media content

⚙️Input Parameters

input_video

string

Input video file

draw_box

boolean

Draw box around detected faces

bg_enhancement

boolean

Enable background enhancement using RealESRGAN

upscale

integer

Upscaling factor for enhancement

max_length

integer

Maximum number of frames to process in one batch for KEEP model

has_aligned

boolean

Set to true if faces in the video are already aligned and cropped (512x512)

only_center_face

boolean

Process only the center face if multiple faces are detected

detection_model

string

Face detection model to use

face_upsample

boolean

Upsample restored faces using RealESRGAN

bg_tile

integer

Tile size for background upsampler

save_video_fps

number

FPS for the output video. If None, uses original FPS

💡Usage Examples

Example 1

Input Parameters

{
  "draw_box": true,
  "input_video": "https://api.replicate.com/v1/files/YmFjMzJiY2YtY2ViMC00MjJjLTg0MTItYWFmYmU4NGYyOGU1",
  "bg_enhancement": false
}

Output Results

https://replicate.delivery/xezq/zoY2cDWpp7ofKi93fpGmoSNksUrloQzl28xAefP0kX0rcP6SB/tmp65mci3dvdownload_restored.mp4

Quick Actions

Use NowView Documentation

Technical Specifications

Hardware Type: L40S
Run Count: 68
Commercial Use: Supported
Platform: Replicate

Related Keywords

Video Face Super-ResolutionTemporal-Consistent Face EnhancementRealESRGAN IntegrationBatch ProcessingFacial Landmark DetectionVideo RestorationEnhancing Video Conference QualityUpscaling Social Media Content

Related Models

Luma Reframe Video

Change the aspect ratio of any video up to 30 seconds long, outputs will be 720p

Frames to Video Merger

Convert a set of image frames (JPG or PNG) into a high-quality MP4 video. Automatically handles sorting and frame order for smooth playback.

twha Video Generation Model

A model for generating videos from text prompts and starting images.