Skip to main content
Artifact removal models clean up JPEG compression artifacts, blur, noise, and other image degradation. These models are essential for restoring quality to compressed comic scans and web images.

​
What are compression artifacts?

JPEG compression creates visible artifacts including:
  • Blocking: 8x8 pixel blocks visible in smooth areas
  • Ringing: Halos around sharp edges
  • Color banding: Reduced color depth in gradients
  • Blur: Loss of fine detail and sharpness
  • Noise: Random color variations
Artifact removal models are trained to reverse this degradation and restore image quality.

​
Available models

All artifact removal models operate at 1x scale, preserving the original resolution while improving quality.

​
OpenComic AI models

​
OpenComic AI Artifact Removal Compact

  • Model ID: opencomic-ai-artifact-removal-compact
  • Upscaler: upscayl
  • Scale: 1x
  • Latency: 0.5s (Very Fast)
  • Source: ollm/opencomic-ai-training
Ultra-fast compact model for general artifact removal. Excellent for batch processing and real-time workflows. Best for:
  • Quick artifact cleanup
  • Batch processing
  • Moderate compression artifacts

​
OpenComic AI Artifact Removal Lite

  • Model ID: opencomic-ai-artifact-removal-lite
  • Upscaler: upscayl
  • Scale: 1x
  • Latency: 2.97s (Fast)
  • Source: ollm/opencomic-ai-training
Larger model with improved quality and detail preservation. Best for:
  • High-quality restoration
  • Complex artifacts
  • Professional workflows

​
OpenComic AI Artifact Removal

Full-size model with maximum quality. Best results for heavily compressed images. Best for:
  • Archival quality restoration
  • Severely compressed images
  • Maximum quality requirements

​
JPEG restoration models

​
NMKD Jaywreck3 Lite

  • Model ID: 1x_NMKD-Jaywreck3-Lite_320k
  • Upscaler: upscayl
  • Scale: 1x
  • Latency: 2.98s (Fast)
  • Source: NMKD.de
Community model trained for JPEG artifact removal with 320k iterations. Best for:
  • JPEG-specific artifacts
  • Web images
  • Social media content

​
NMKD Jaywreck3 Soft Lite

  • Model ID: 1x_NMKD-Jaywreck3-Soft-Lite_320k
  • Upscaler: upscayl
  • Scale: 1x
  • Latency: 2.98s (Fast)
  • Source: NMKD.de
Softer variant that preserves more of the original image character while removing artifacts. Best for:
  • Natural-looking restoration
  • Subtle artifact removal
  • Preserving artistic style

​
SaiyaJin DeJpeg

  • Model ID: 1x-SaiyaJin-DeJpeg
  • Upscaler: upscayl
  • Scale: 1x
  • Latency: 8.2s (Slow)
  • Source: OpenModelDB
Specialized JPEG decompression model focusing on blocking and ringing artifacts. Best for:
  • Heavy JPEG compression
  • Blocking artifacts
  • Old web images

​
JPEG Destroyer V2

  • Model ID: 1x_JPEGDestroyerV2_96000G
  • Upscaler: upscayl
  • Scale: 1x
  • Latency: 8.22s (Slow)
  • Source: Hugging Face
Aggressive JPEG artifact removal with focus on severe compression. Best for:
  • Heavily compressed images
  • Multiple recompression artifacts
  • Extreme restoration cases

​
Using artifact removal models

​
Basic artifact removal

import OpenComicAI from 'opencomic-ai-bin';

await OpenComicAI.pipeline('./compressed.jpg', './restored.jpg', [
  { model: 'opencomic-ai-artifact-removal-compact' }
]);

​
Artifact removal with upscaling

Apply artifact removal before upscaling for best results: 
await OpenComicAI.pipeline('./input.jpg', './output.jpg', [
  { model: 'opencomic-ai-artifact-removal-compact' },
  { model: 'realesrgan-x4plus-anime', scale: 4 }
]);

​
Complete restoration pipeline

Combine descreening, artifact removal, and upscaling: 
await OpenComicAI.pipeline('./scan.jpg', './restored.jpg', [
  { model: 'opencomic-ai-descreen-hard-compact' },
  { model: 'opencomic-ai-artifact-removal-lite' },
  { model: 'realcugan', scale: 4, noise: 0 }
]);
For heavily compressed images, consider using multiple passes with different artifact removal models.

​
Performance comparison

ModelLatencySpeed CategoryStrength
opencomic-ai-artifact-removal-compact0.5sVery FastModerate
opencomic-ai-artifact-removal-lite2.97sFastStrong
1x_NMKD-Jaywreck3-Lite_320k2.98sFastStrong
1x_NMKD-Jaywreck3-Soft-Lite_320k2.98sFastSubtle
1x-SaiyaJin-DeJpeg8.2sSlowVery Strong
opencomic-ai-artifact-removal8.21sSlowVery Strong
1x_JPEGDestroyerV2_96000G8.22sSlowExtreme

​
Choosing the right model

For speed: Use opencomic-ai-artifact-removal-compact For quality: Use opencomic-ai-artifact-removal-lite or opencomic-ai-artifact-removal For JPEG-specific artifacts: Use 1x_NMKD-Jaywreck3-Lite_320k or 1x-SaiyaJin-DeJpeg For subtle restoration: Use 1x_NMKD-Jaywreck3-Soft-Lite_320k For severe compression: Use 1x_JPEGDestroyerV2_96000G or opencomic-ai-artifact-removal For batch processing: Use opencomic-ai-artifact-removal-compact

​
How artifact removal works

Artifact removal models use deep learning to:
  1. Detect artifacts: Identify compression patterns, blocking, and distortion
  2. Reconstruct details: Estimate original content before compression
  3. Remove artifacts: Eliminate blocking, ringing, and banding
  4. Restore sharpness: Recover lost edge definition and fine details
  5. Preserve structure: Maintain image composition and intentional blur
Artifact removal models operate at 1x scale. They improve quality without changing resolution. Combine with upscaling models to increase size.

​
Types of artifacts removed

​
JPEG compression artifacts

  • 8x8 blocking in smooth areas
  • Ringing halos around edges
  • Mosquito noise near text and lines
  • Color bleeding and banding

​
Other degradation

  • Gaussian blur from resizing
  • Color quantization artifacts
  • Noise and grain
  • Loss of fine detail

​
Pipeline order recommendations

  1. Descreen - Remove halftone patterns first
  2. Artifact removal - Clean up compression artifacts
  3. Upscale - Increase resolution with clean input
await OpenComicAI.pipeline('./input.jpg', './output.jpg', [
  { model: 'opencomic-ai-descreen-hard-compact' },      // 1. Remove patterns
  { model: 'opencomic-ai-artifact-removal-compact' },   // 2. Clean artifacts
  { model: 'realcugan', scale: 4, noise: 0 }           // 3. Upscale
]);

Build docs developers (and LLMs) love

Get started for freeTalk to us