Clarity AI offers an open-source AI image upscaler and enhancer, positioned as an alternative to commercial solutions like Magnific. It targets users seeking to improve image resolution and quality, with various integration options available.

How It Works

The project leverages AI models for image upscaling, with specific features like "Flux Upscaling" for enhanced detail in faces, text, and art. It supports custom safetensors checkpoints and LoRAs, allowing for tailored upscaling results. The architecture is designed for flexibility, offering integration via a Cog implementation, a ComfyUI node, and compatibility with the AUTOMATIC1111 Stable Diffusion WebUI.

Quick Start & Requirements

• Cog: cog predict -i image="link-to-image"
• ComfyUI: Install "Clarity AI" node via Manager, add API key (env var CAI_API_KEY, file cai_platform_key.txt, or node field). Alternatively, use the free-workflow.json from the ComfyUI-ClarityAI repository.
• A1111 WebUI: Requires specific prompt, negative prompt, sampler, and ControlNet settings as detailed in the README.
• Dependencies: Python, Cog, ComfyUI, Stable Diffusion WebUI. Specific model weights may need to be downloaded.

Highlighted Details

• Supports upscaling to 13kx13k resolutions.
• Offers output formats including JPG, PNG, and WebP.
• Features include pattern upscaling, multi-step upscaling, and pre-downscaling.
• Provides a ComfyUI node and API for integration.

Maintenance & Community

The project is actively developed by philz1337x, who also offers paid versions and services via ClarityAI.co. Updates and new features are frequently posted on Twitter.

Licensing & Compatibility

The repository is described as free and open-source. Specific licensing details are not explicitly stated in the README, but the project is presented as an alternative to paid services.

Limitations & Caveats

The "Flux Upscaling" feature is exclusively available via the paid ClarityAI.co app and API, not within this open-source repository. Some integration methods, like Cog, may require familiarity with the underlying tools.