I Built an AI Upscaler with 6 Specialist Models — Here's What I Learned
Why one model isn't enough When I started building an AI upscaling service, I thought I'd just wrap Real-ESRGAN and call it done. Wrong. A model trained on photographs produces blurry edges on digital art. A model trained on anime smears skin textures in portraits. And video? Completely different challenge. After testing dozens of models, I picked 6 that each excel at a specific image type: Model…
Building an AI upscaling service revealed the importance of model selection over scale factor. A skilled model can produce superior results even when scaling by 2x, while a less suitable model could struggle with a 4x upscale. The free tier of the service saw high conversion rates, as users who experienced positive outcomes tended to upgrade. Watermarks detracted from conversion rates, discouraging users from fully embracing the service.
Video upscaling posed unique challenges compared to image upscaling. Achieving temporal consistency across frames proved to be the most difficult engineering hurdle. After testing numerous models, the creator ultimately chose six that excelled at specific image types: Flare for natural textures, Prism for digital art, Lumen for portraits, Mirage for illustrations, Motion for video, and Motion X for enhanced video quality.
The architecture of UpRes was streamlined into a single API endpoint, allowing users to submit image URLs, select a model and scale factor, and receive the output URL upon completion. Most jobs completed within 15-25 seconds.
A key takeaway was the critical role of model selection in upscaling success, more so than adjusting the scale factor. Many users were enticed to upgrade after trying the free tier and witnessing impressive results, but watermarks significantly impeded conversion. The video upscaling challenge took three months to master, and the concept of Multi-Clip Processing (MCP) emerged as a game-changer, enabling developers to upscale images simply by saying "Claude, upscale this image."
The service is available at https://upres.ai (5 free images, no card), with CLI support via npm, API documentation at https://upres.ai/api-playground, and the source code hosted on GitHub at https://github.com/auroracapital/upres-cli. The creator expressed enthusiasm for learning about users' upscaling needs and preferred models.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
