philz1337x

clarity-upscaler: Enhance and Upscale Images with AI

Clarity-Upscaler is a Python image-to-image project for increasing image resolution and enhancing details with Stable Diffusion workflows. It suits users comfortable with Cog or image-generation tools who want a configurable alternative to hosted upscaling services.

PythonAGPL-3.0AIGenerative AIComputer Vision
View on GitHub Compare
Stars
5.1k
Forks
535
Stars, last 7 days
+2

Our take

Health score

21/100

How it is scored
Activity
20
Community
22
Issues
21
Pull requests
21

+2 stars in the last 7 days

Overview

Clarity-Upscaler is an AI image upscaling and enhancement project built around image-to-image generation. It aims to increase resolution while refining image details, with options for workflows such as multi-step and tiled upscaling.

The repository is an implementation for Cog rather than a turnkey desktop application. It is most relevant to developers and experienced image-generation users who want to run or adapt the code. The README also describes ComfyUI and AUTOMATIC1111 approaches, while noting that the hosted Flux upscaler is not open source.

Key Features

  • AI-based image upscaling and enhancement.
  • Multi-step upscaling and pre-downscaling options described in the project updates.
  • Support for custom safetensors checkpoints and LoRAs.
  • Anime and pattern upscaling support described in the README.
  • Sharpening and output in JPG, PNG, or WebP formats.
  • Cog prediction workflow, with separate ComfyUI and AUTOMATIC1111 usage paths documented.

Use Cases

  • Image-generation users who want to enlarge an image while adding or refining detail.
  • ComfyUI or AUTOMATIC1111 users who want to incorporate the project into an existing Stable Diffusion workflow.
  • Developers evaluating a Cog-based image enhancement implementation for integration or customization.
  • Creators working with anime or patterned images who want to try the documented upscaling options.

Getting Started

For Cog, download the weights and run a prediction:

python download_weights.py
cog predict -i image="link-to-image"

See the README for setup details and the ComfyUI and AUTOMATIC1111 options.

Considerations

  • The repository identifies itself as a Cog implementation, so using it directly assumes familiarity with Cog; the README recommends other options for newcomers.
  • AI image-to-image upscaling can alter image details rather than simply enlarge pixels. Review results for fidelity to the source.
  • Flux upscaling is offered through the linked service and API, but the README explicitly says that Flux feature is not open source.
  • The project is licensed under AGPL-3.0. Review the license obligations before redistributing or integrating the code.

Found this useful?

Share it with someone who would like clarity-upscaler.