The Google Upscaler (also known as the image upscaling feature in Google's Vertex AI suite, with model IDs such as image-upscaler-001) is a proprietary image-enhancement model developed by Google. First introduced as part of the generative AI suite on Vertex AI, this model is designed to increase the resolution of existing, generated, or edited images by 2x or 4x without sacrificing visual quality or introducing aggressive artificial artifacts.
Unlike traditional upscalers that rely on simple pixel interpolation—which can produce soft or plasticky textures—the Google Upscaler uses a lightweight, fast super-resolution architecture. It operates via pure learned interpolation to predict and reconstruct fine details, such as subtle edges and textures, with minimal latency. It is highly valued as a fast, low-cost general-purpose upscaler that remains exceptionally faithful to the source image, maintaining original composition, colors, and lighting without hallucinating new elements.
To achieve optimal results, users are encouraged to start with well-illuminated, minimally compressed source images, as the AI works by enhancing existing data rather than generating entirely new scenes. The model supports input image files up to a maximum size of 10MB. For many workflows, beginning with a 2x upscaling factor is recommended to evaluate detail preservation and artifact control before opting for a larger 4x enlargement. A customizable parameter is also available to control the compression quality of the output, typically defaulting to 80 for JPEG exports.