feat: add post-processing script for analog print effects #33

New issue

Closed

opened 2026-05-26 14:05:01 -04:00 by jasper · 0 comments

jasper commented

2026-05-26 14:05:01 -04:00

Contributor

Summary

Add a post-processing script (scripts/postprocess.py) and corresponding SKILL.md documentation for analog print effects applied to raw generated images. Inspired by the approach in plntrprotocol/nous-branding.

Motivation

Raw AI-generated images are too clean for the Nous/Hermes aesthetic. The skill needs a two-stage pipeline: generation + mandatory post-processing for analog print degradation. This is the single highest-leverage improvement we can make — it works regardless of which image generation provider is used.

Proposed Solution

Add nous-branding/scripts/postprocess.py (Pillow + numpy) implementing 14 analog print effects:

Warm color grade (shadow tones to amber/gold)
CRT scanlines (faint horizontal retro-display lines)
Film grain (fine Gaussian noise across entire image)
Bayer dither (4x4 ordered dither matrix)
Warm vignette (edges darkened + warm-shifted)
Chromatic aberration (subtle RGB channel offset at edges)
Screen print texture (simulated halftone dot pattern)
Paper fiber (subtle canvas/paper substrate texture)
Ink bleed (slight edge smearing at dark boundaries)
Palette compression (pulls clean RGB renders toward constrained print inks)
Xerox threshold (degraded photocopy contrast and threshold breakup)
Registration offset (subtle cyan/orange plate misalignment)
Plate wobble (subtle row-wise distortion)
Print scuffs (sparse scratches and imperfect ink pickup)

Then update nous-branding/SKILL.md with a Post-Processing section documenting usage and intensity calibration.

Filed by Jasper (AI agent on behalf of Magnus Hedemark)

## Summary Add a post-processing script (`scripts/postprocess.py`) and corresponding SKILL.md documentation for analog print effects applied to raw generated images. Inspired by the approach in [plntrprotocol/nous-branding](https://github.com/plntrprotocol/nous-branding). ## Motivation Raw AI-generated images are too clean for the Nous/Hermes aesthetic. The skill needs a two-stage pipeline: generation + mandatory post-processing for analog print degradation. This is the single highest-leverage improvement we can make — it works regardless of which image generation provider is used. ## Proposed Solution Add `nous-branding/scripts/postprocess.py` (Pillow + numpy) implementing 14 analog print effects: 1. Warm color grade (shadow tones to amber/gold) 2. CRT scanlines (faint horizontal retro-display lines) 3. Film grain (fine Gaussian noise across entire image) 4. Bayer dither (4x4 ordered dither matrix) 5. Warm vignette (edges darkened + warm-shifted) 6. Chromatic aberration (subtle RGB channel offset at edges) 7. Screen print texture (simulated halftone dot pattern) 8. Paper fiber (subtle canvas/paper substrate texture) 9. Ink bleed (slight edge smearing at dark boundaries) 10. Palette compression (pulls clean RGB renders toward constrained print inks) 11. Xerox threshold (degraded photocopy contrast and threshold breakup) 12. Registration offset (subtle cyan/orange plate misalignment) 13. Plate wobble (subtle row-wise distortion) 14. Print scuffs (sparse scratches and imperfect ink pickup) Then update `nous-branding/SKILL.md` with a Post-Processing section documenting usage and intensity calibration. --- Filed by Jasper (AI agent on behalf of Magnus Hedemark)