Comic Book Image Processor

Local desktop app for batch-processing slab or auction comic scans into transparent WebP or layered PSD files, then reviewing the results inside the app.

Features

• Input, output, review, and debug folder pickers
• Batch export as PSD (mask on separate layer for Photoshop, default) or WebP (single layer with alpha)
• Straighten — rotate so top edge is horizontal (affine only, no perspective skew)
• Crop padding — max 24px around comic edges
• Boundary inset — shrink mask inward to avoid slab/plastic edges
• Tone controls: luminance autocontrast, shadow/highlight clip, gamma, white suppression
• Start, cancel, and single-file test run
• Animated processing state and completion summary
• Batch manifest export to batch_manifest.jsonl
• Built-in Review tab with:
  • original image preview
  • debug overlay preview
  • final output preview
  • buttons for Approve, Reject, Needs Recrop, Mask Too Aggressive, Wrong Object, Skip
  • keyboard shortcuts: A, R, N, M, W, S, ←, →
  • "Show unreviewed only" filter
  • "Approve All Unflagged" batch action
  • notes field
  • saved review decisions in review_results.json
• Preset save/load

Install

pip install -r requirements.txt

Run

python main.py

Recommended first use

1. Choose input, output, review, and debug folders.
2. Click Test First File.
3. Inspect the output PSD/WebP and debug image.
4. Adjust settings if needed.
5. Run the full batch.
6. Open the Review tab and label flagged or questionable results.

Review data files

After a batch, the app writes:

• batch_manifest.jsonl — one record per processed file
• batch_summary.json — summary counts
• review_results.json — your review decisions and notes

These files are designed to support later tuning or model-training work.

Learn from Results

After reviewing a batch, use Learn from Results in the Process tab to:

• Suggested threshold — Analyzes approved vs rejected items and suggests a confidence threshold so similar problematic items get flagged for review
• Method stats — Shows approval rate for edges vs nonwhite detection methods

Click Apply Suggested Threshold to update the Review threshold setting.