supersambat/4x-faceupldat-upscaler

4x face upscaler with denoising for lower quality inputs. DAT architecture trained on FaceUp dataset.

Public
0 runs
License
Weights
Paper
Playground Examples README Versions

4x-FaceUpLDAT Face Upscaler (Denoising Variant)

A 4x super-resolution model for face and portrait photos with built-in denoising. Designed for lower quality inputs — handles noise, compression artifacts, and blur better than the standard FaceUpDAT variant.

Model Details

  • Architecture: DAT (Dual Aggregation Transformer)
  • Scale: 4x upscaling
  • Training data: FaceUp (10,000 curated face images)
  • Training iterations: 80,000
  • Best for: Low quality, noisy, or compressed face photos

When to use this vs FaceUpDAT

  • FaceUpDAT — Use for clean, good quality face photos
  • FaceUpLDAT (this model) — Use for noisy, compressed, or degraded face photos

Usage

Upload a face/portrait image and the model will return a 4x upscaled and denoised version.

Credits & Acknowledgments

This model was originally created by Helaman and published on OpenModelDB. Full credit goes to the original author for training and releasing this model.

This Replicate deployment is a wrapper for convenient API access. The original model is licensed under CC-BY-4.0.

Model created