Stability AI has released a new Image Upscaling API, aimed at increasing image size without loss of sharpness or detail. This addition complements Stability AI's current range of image generation and editing APIs, including text-to-image, image-to-image, and inpainting services.
The company's latest offering seeks to address the challenge of low-resolution outputs from many popular text-to-image AI generation tools. Users can submit smaller images to the API and receive an upscaled version with enhanced detail. The new functionality has also been integrated into the most recent versions of their Stability for Photoshop and Stability for Blender add-ons.
Currently, two open-source models underpin the Image Upscaling API: Real-ESRGAN and Stable Diffusion 4x Upscaler. Real-ESRGAN can quickly double the resolution of a 512x512 image in approximately half a second. The Stable Diffusion 4x Upscaler requires more time, between 20 to 40 seconds, but provides increased detail and texture.
Importantly, you can upscale any images—even ones not generate by Stability AI, with the API. Stability AI plans to keep updating the Image Upscaling API with new models as they become available.