Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/ollm/opencomic-ai-models/llms.txt

Use this file to discover all available pages before exploring further.

Descreening removes halftone dot patterns — the regular grids of ink dots that printing presses use to simulate continuous tones — from scanned comics, manga, and other printed media. When a printed page is scanned at typical resolutions, those dot matrices appear as moiré interference patterns or visible grain that no amount of ordinary sharpening or upscaling can fix cleanly. Descreening models are trained to recognise and eliminate these patterns while preserving the underlying artwork. Like artifact removal models, all descreen models operate at 1× scale (output resolution equals input resolution) and all use the upscayl ESRGAN-compatible backend. Weight files are located under <modelsPath>/descreen/models/.
Always run descreening before upscaling. Feeding a halftone-patterned image into an upscaler will amplify the dot structure, making it far harder to remove afterwards. The recommended pipeline is: scan → descreen → artifact removal (optional) → upscale.

​
All Descreen Models

Model IDDisplay NameNotes
opencomic-ai-descreen-hardOpenComic AI Descreen HardStandard tier; strongest halftone suppression
opencomic-ai-descreen-hard-liteOpenComic AI Descreen Hard LiteLite tier; faster with slightly reduced suppression
opencomic-ai-descreen-hard-compactOpenComic AI Descreen Hard CompactCompact tier; fastest and smallest footprint
1x_halftone_patch_060000_GHalftone Patch 060000 GGeneral halftone removal; NMKD.de
1x_wtp_descreenton_compactWTP DescreenTon CompactCompact community model; OpenModelDB

​
OpenComic AI Three-Tier System

The three OpenComic AI descreen models are all trained for “hard” descreening — aggressive suppression of coarse halftone dot patterns common in older printed comics — but compiled at different network sizes:
TierModel IDCharacteristic
Standardopencomic-ai-descreen-hardFull network capacity; most thorough pattern removal with the best edge preservation
Liteopencomic-ai-descreen-hard-liteReduced network; significantly faster while retaining strong descreening quality
Compactopencomic-ai-descreen-hard-compactSmallest and fastest; use for large batch jobs or real-time preview passes
For most single-image workflows, Lite is the recommended starting point. Switch to Standard if faint dot residue remains in flat areas after processing; use Compact when throughput is more important than maximum quality.

​
Community Models

  • Halftone Patch 060000 G is a broadly trained general halftone removal model. It handles a wide range of dot frequencies and angles, making it a useful fallback when the source material has unusual screen rulings.
  • WTP DescreenTon Compact is a lightweight community model focused on tone-dot patterns. Its smaller architecture makes it fast and low on memory, suitable as a quick first pass or for less severe halftoning.

​
Usage

import OpenComicAIModels from 'opencomic-ai-models';
import { execFile } from 'node:child_process';
import path from 'node:path';

const modelsPath = OpenComicAIModels.path;

execFile('upscayl-bin', [
  '-i', 'scanned-page.png',
  '-o', 'descreened-page.png',
  '-m', path.join(modelsPath, 'descreen', 'models'),
  '-n', 'opencomic-ai-descreen-hard-lite',
  '-s', '1',
]);
Shell
upscayl-bin \
  -i scanned-page.png \
  -o descreened-page.png \
  -m /path/to/models/descreen/models \
  -n opencomic-ai-descreen-hard-lite \
  -s 1

​
Full Pipeline Example

The following shell snippet chains descreening and upscaling for a complete scan-enhancement workflow: 
# Step 1 — remove halftone pattern
upscayl-bin \
  -i scan.png \
  -o descreened.png \
  -m /path/to/models/descreen/models \
  -n opencomic-ai-descreen-hard-lite \
  -s 1

# Step 2 — upscale the clean image
upscayl-bin \
  -i descreened.png \
  -o final.png \
  -m /path/to/models/upscale/models \
  -n realesrgan-x4plus-anime \
  -s 4

Build docs developers (and LLMs) love

Get started for freeTalk to us