ESLPwn

41 Commits 1 Branch 2 Tags

3 Commits

Include renames

Author	SHA1	Message	Date
i12bp8	99bcc162b8	web-image-prep: chroma-contracted quantiser for newsprint illusion Deploy Image Prep to GitHub Pages / deploy (push) Failing after 31s Details Previous version pulled accent ink too aggressively (accentPull 0.85) so anything red-leaning collapsed to solid red. Real photographic feel from a 3-ink palette comes from W/B halftones simulating greys with the chromatic ink only sparsely overlaid - the newsprint illusion. Three knobs now make this work: - L_WEIGHT 3.5 (was 1.6): lightness completely dominates the metric - CHROMA_CAP: source a/b axes pre-scaled to the achievable accent chroma so the dither isn't asked for more colour than one ink can deliver. Excess chroma gets dropped via FS error diffusion through B+W neighbours. - accentPull 0.06 (was 0.85): only a tiebreaker bonus, used in the most strongly hue-aligned pixels.	2026-04-26 21:03:10 +02:00
i12bp8	ad9b6dd2fe	web-image-prep: hue-aligned tri-colour quantiser + serpentine dither The old quantiser pulled accent ink toward any chromatic pixel using a flat chroma bonus, which smeared red across shadows and produced flat red blobs without depth. Two fixes: - Distance now weights L (lightness) ~1.6x more than chroma, so dark pixels go to black and light pixels to white before colour competes. That preserves the photograph's tonal structure. - Accent attraction is the squared signed projection of the pixel's chroma onto the accent direction. Red ink only pulls red-leaning pixels (positive a*); blue/green never picks red. The squaring gives a smooth ramp, so faint reds get sparse red dither and saturated reds get dense red - matching how a real risograph layers a single chromatic ink under W/B halftones. Also flipped Floyd-Steinberg to a serpentine scan with mirrored kernel on reverse rows, killing the diagonal worm artefacts that gave smooth gradients an obviously-dithered look.	2026-04-26 20:59:15 +02:00
i12bp8	d608afb43b	Drop-folder image flow + web image preparer - Add web-image-prep/: self-contained HTML5 image preparer with perceptual Oklab 3-colour quantisation, Floyd-Steinberg/Atkinson/ Bayer dither, all supported tag profiles, BMPs that match the FAP reader exactly. Auto-deployed to GitHub Pages. - Add 'Set Image' under Targeted Payloads. Lists every BMP in apps_data/tagtinker/dropped/, scrolls long names, marks foreign- resolution files with a '~' prefix. - Stream-time nearest-neighbour rescaling so any BMP can target any tag at any page; image options screen is now Page-only. - Remove Phone Sync menu and the Android companion app: replaced by the browser preparer + drop folder. - Static, IR-friendly transmit screen ('Tinkering Tag', point hint, 3-dot loader, 'Flipped ;)').	2026-04-26 15:40:27 +02:00

Author

SHA1

Message

Date

i12bp8

99bcc162b8

web-image-prep: chroma-contracted quantiser for newsprint illusion

Deploy Image Prep to GitHub Pages / deploy (push) Failing after 31s

Details

Previous version pulled accent ink too aggressively (accentPull 0.85)
so anything red-leaning collapsed to solid red. Real photographic feel
from a 3-ink palette comes from W/B halftones simulating greys with
the chromatic ink only sparsely overlaid - the newsprint illusion.

Three knobs now make this work:
  - L_WEIGHT 3.5 (was 1.6): lightness completely dominates the metric
  - CHROMA_CAP: source a/b axes pre-scaled to the achievable accent
    chroma so the dither isn't asked for more colour than one ink can
    deliver. Excess chroma gets dropped via FS error diffusion through
    B+W neighbours.
  - accentPull 0.06 (was 0.85): only a tiebreaker bonus, used in the
    most strongly hue-aligned pixels.

2026-04-26 21:03:10 +02:00

i12bp8

ad9b6dd2fe

web-image-prep: hue-aligned tri-colour quantiser + serpentine dither

The old quantiser pulled accent ink toward any chromatic pixel using a
flat chroma bonus, which smeared red across shadows and produced flat
red blobs without depth. Two fixes:

- Distance now weights L (lightness) ~1.6x more than chroma, so dark
  pixels go to black and light pixels to white before colour competes.
  That preserves the photograph's tonal structure.
- Accent attraction is the squared *signed projection* of the pixel's
  chroma onto the accent direction. Red ink only pulls red-leaning
  pixels (positive a*); blue/green never picks red. The squaring gives
  a smooth ramp, so faint reds get sparse red dither and saturated
  reds get dense red - matching how a real risograph layers a single
  chromatic ink under W/B halftones.

Also flipped Floyd-Steinberg to a serpentine scan with mirrored kernel
on reverse rows, killing the diagonal worm artefacts that gave smooth
gradients an obviously-dithered look.

2026-04-26 20:59:15 +02:00

i12bp8

d608afb43b

Drop-folder image flow + web image preparer

- Add web-image-prep/: self-contained HTML5 image preparer with
  perceptual Oklab 3-colour quantisation, Floyd-Steinberg/Atkinson/
  Bayer dither, all supported tag profiles, BMPs that match the FAP
  reader exactly. Auto-deployed to GitHub Pages.
- Add 'Set Image' under Targeted Payloads. Lists every BMP in
  apps_data/tagtinker/dropped/, scrolls long names, marks foreign-
  resolution files with a '~' prefix.
- Stream-time nearest-neighbour rescaling so any BMP can target any
  tag at any page; image options screen is now Page-only.
- Remove Phone Sync menu and the Android companion app: replaced by
  the browser preparer + drop folder.
- Static, IR-friendly transmit screen ('Tinkering Tag', point hint,
  3-dot loader, 'Flipped ;)').

2026-04-26 15:40:27 +02:00