Turning finished anime art back into line drawings

What it does Anime2Sketch is an inference-only tool that extracts clean line art from finished anime, illustrations, and manga, and can simplify rough hand-drawn sketches into polished drawings. Built on a WACV 2022 adversarial open-domain adaptation method, it ships with pretrained PyTorch weights and two model variants: a standard set and an artifact-free version for dark or low-contrast images.

The interesting bit Most sketch extractors are tuned for natural photographs; this one targets the flat colors and stylized edges of anime and illustration. The repository is essentially a curated inference kit—training code lives in a separate project—so the value is in the ready-made weights and narrow domain focus.

Key highlights

• Handles finished anime art, illustration, and rough hand-drawn sketches
• Two pretrained weight sets: standard and artifact-free for dark or low-contrast inputs
• Processes images at 512×512 by default to manage memory
• Supports CPU and GPU inference, with Docker packaging available
• MIT licensed

Caveats

• No training code included; retraining requires the separate AODA repository
• Pretrained weights are hosted on Google Drive, not in the repository itself
• Default resizing to 512×512 may lose fine detail on high-resolution artwork

Verdict Grab it if you need to recover or clean up line art from existing anime assets. Skip it if you are after a general-purpose photo sketch filter or want to train your own model from scratch.