Restore old, damaged, and low quality photos with GFP-GAN

Here’s something fun to do with your family during upcoming holiday gatherings, whether virtual or in-person: Restore your old family photos using this photo restoration app, built with Baseten and powered by the GFP-GAN model.

GFP-GAN is a blind face restoration model that was published in June 2021 by Xintao Wang, Yu Li, Honglun Zhang, and Ying Shan. Read their paper here: Towards Real-World Blind Face Restoration with Generative Facial Prior. You can also check out their GitHub code.

This model caught our eye when we saw an impressive before and after example making the rounds. We were eager to try the model out ourselves and so decided to build our own photo restoration app with Baseten.

Here’s what building this photo restoration app with Baseten entailed:

  • We deployed the model on Baseten. Baseten provides a pretty simple yet flexible deployment flow, GFP-GAN’s GitHub repo contained the majority of code so it was a matter of installing the dependencies, abstracting the inference functions, and after a simple baseten.deploy(), we were ready to serve predictions. We’ll shortly add the model to our growing model zoo so all Baseten users can build apps with it!

  • We allocated a single GPU to run inference, which in combination with Baseten’s request buffering, should be able to handle a decent load.

  • We created a new Baseten application—all we need here is to serve predictions and build a simple frontend.

  • We created a worklet (code that defines the application’s business logic).

The worklet graph for the photo restoration app
  • We built a view (user-facing interface) that allows users to upload a photo, click a “Restore” button, and see the model output, a restored version of their photo.

  • We made the app public, so we could share it with users like you

And ta-da, our very own photo restoration app! Baseten really is the fastest way to build ML-powered apps.

Give the app a try and let us know what you think. Ready to build your own ML-powered app? Sign up for Baseten today!

Machine learning infrastructure that just works

Baseten provides all the infrastructure you need to deploy and serve ML models performantly, scalable, and cost-efficiently.

Machine Learning

NVIDIA A10 vs A100 GPUs for LLM and Stable Diffusion inference

This article compares two popular GPUs—the NVIDIA A10 and A100—for model inference and discusses the option of using multi-GPU instances for larger models.

Philip Kiely

September 15, 2023

Machine Learning

SDXL inference in under 2 seconds: the ultimate guide to Stable Diffusion optimization

Out of the box, Stable Diffusion XL 1.0 (SDXL) takes 8-10 seconds to create a 1024x1024px image from a prompt on an A100 GPU. Here’s everything I did to cut SDXL invocation to as fast as 1.92 seconds on an A100.

Varun Shenoy

August 30, 2023