APISR | Anime Image/Video Upscaler

1. APISR for Anime Image Resolution | ComfyUI Upscale Workflow

This ComfyUI upscale workflow integrates the APISR (Anime Production-oriented Image Super-Resolution) model for upscaling low-quality, low-resolution anime images and videos. Additionally, it incorporates the 4xAnimateSharp Model for comparison purposes.

By default, the workflow is configured for image upscaling. To upscale videos, simply replace “load image” node with “load video” node and replace “save image” node with “combine video” node for a seamless transition.

2. Introduction to APISR

APISR (Anime Production-oriented Image Super-Resolution) is designed to restore and enhance low-quality, low-resolution anime images and video sources that have undergone various degradations in real-world scenarios. It caters to the increasing interest in anime super-resolution (SR) by deviating from the traditional application of photorealistic SR techniques, which may not adequately accommodate the distinctive characteristics of anime.

Two primary challenges are identified in anime SR: the degradation of hand-drawn lines due to compression or aging, and the presence of unwanted color artifacts. To tackle these, APISR introduces an image degradation model specifically designed to simulate compression effects and assist in the restoration of hand-drawn lines. Additionally, it proposes a technique for enhancing faint, aging hand-drawn lines by merging them from overly sharpened images.

For the issue of color artifacts, often arising from GAN-based SR networks, APISR introduces balanced twin perceptual loss. This method combines perceptual features from both photorealistic and anime domains, aiming to reduce color inconsistencies and improve the visual quality of super-resolved anime images.

APISR introduces a dataset curation pipeline specifically designed for anime, an image degradation model that targets the unique challenges of anime restoration, and an innovative approach for addressing color artifacts in GAN-based SR network training. These methods collectively focus on enhancing the super-resolution process for anime by selecting the most informative frames for datasets, restoring degraded hand-drawn lines, and improving visual quality by addressing color inconsistencies. The effectiveness of APISR's approach is validated through comprehensive evaluations on real-world anime SR datasets, showing significant advancements over existing techniques.

3. How to Use ComfyUI APISR for Anime Image Resolution

The APISR model is very user-friendly; simply select the model corresponding to the upscale size you desire.

For more information, check it on the github and paper