003 ASCII and Dither
ASCII 與像素抖色效果
為登入頁、官網或卡片中的圖片和影片加上 ASCII 與 Dither 效果,效果強弱可以隨時調整。
ASCII 和 Dither 依亮度把圖片重新畫成字元或像素紋理,保留明暗結構,去掉照片的細節;參數都可以調整,同一張圖既能當主視覺,也能退到背景裡當紋理。要點有以下五條:
- ASCII:適合終端機、程式碼一類的質感,例如開發者工具的登入頁。把畫面切成網格,每格依亮度換成一個字元,從「.」到「@」越來越密;大小是字級,間距是格子疏密,取樣強度是明暗對比。
- Dither:適合印刷網點或像素顆粒的質感。把畫面切成像素塊,以 4×4 的 Bayer 矩陣作為閾值,把每種顏色壓成幾階,遠看混成中間的灰階;大小是像素塊,間距是紋理的尺度,取樣強度越高,顏色階數越少,顆粒感越重。
- 遮罩:兩種效果都適用。在畫面上疊一層垂直漸層,讓畫面退後,效果更像紋理而不是雜訊;效果干擾到上層文字時,先調高遮罩。
- 取樣區域:只用於 ASCII,決定字元畫在暗部還是亮部,依畫面主體所在的明暗選擇。城市夜景的主體是燈光,取樣亮部;白底上的深色產品照片,取樣暗部。
- 色調:可以保留原色,也可以從五組配色中選一組,先把圖片的明暗對應到這組顏色上,再產生效果,讓不同圖片的風格保持一致。
參數
- ASCII
- 遮罩 90;取樣暗部;取樣強度 100;大小 37;間距 0
- Dither
- 遮罩 88;取樣強度 100;大小 15;間距 0
附帶的提示詞會先確認要處理哪張圖片或哪段影片,再為它產生效果,並在旁邊加上一個可以即時調整的參數面板。
Add an ASCII / ordered-dither effect to one specific image or video in
this project, with a live control panel next to it. Use the asset and the
place I name (an <img>, a <video>, or a file path or URL, and the page or
section it sits in); ask me if I have not named them. Apply the effect to
that asset only, in place: keep its size, position, rounded corners and
anything layered on top of it. If the project already has an image or
canvas effect pipeline, or slider, segmented-control and panel components,
build on them; otherwise create a reusable EffectLayer and EffectPanel.
Settings, one object shared by the layer and the panel:
{ filter: 'ascii' | 'dither', mask, palette, inkTone, strength, size,
spacing }
Keep separate values per filter, so switching filters brings back that
filter's own settings.
What each control does in each filter, its range, and the recommended
value (ASCII / Dither):
- Mask effect, 0-100 (90 / 88): a vertical veil over the picture so the
effect reads as texture rather than noise. Same in both filters.
- Color, Original or one of five palettes (Original): gradient-maps the
asset's luminance into one color family before filtering.
- Sample tones, Shadows | Highlights (Shadows; ASCII only, hidden for
Dither): which end of the tonal range becomes dense characters.
- Sampling strength, 0-100 (100 / 100). ASCII: contrast of the tone curve
and how bright and opaque the characters are. Dither: how few levels
each color channel keeps (8 at 0, 2 at 100) and how fully the quantized
color replaces the original.
- Size, 0-100 (37 / 15). ASCII: font size, 7-24px. Dither: pixel block,
2-9px.
- Spacing, 0-100 (0 / 0). ASCII: grid step, 8-24px across and 9-26px
down, never tighter than the glyphs. Dither: scale of the Bayer
pattern, 1-4 blocks per matrix cell.
Helpers used throughout:
- range(v, a, b) = a + (b - a) * v / 100
- luminance L = (0.299 R + 0.587 G + 0.114 B) / 255
- s = strength / 100
- "The box" is the asset's rendered size on the page.
Sampling:
- Cover-fit the asset into the box and draw it into a small sampling
canvas (about 480px on the long side), then read it with getImageData.
A sample is the pixel under a point in the box, composited over white.
- With a palette, recolor that canvas first: map each pixel's L through a
256-entry lookup built from the palette's four stops.
- Per image or video frame, take about 12,000 evenly spaced samples:
low = 8th percentile of L, high = 92nd percentile.
- Shadows means dark ink, Highlights light ink.
ASCII:
- Characters, from empty to full: " .,:;-=+xX08S#&@" (starts with two
spaces).
- Font: round(range(size, 7, 24))px monospace (22 instead of 24 when the
box is under 720px wide), centered in each cell.
- Grid: x step = max(range(spacing, 8, 24), widest glyph * 1.08); line
height = max(range(spacing, 9, 26), height of "@" * 1.12, font * 1.04).
The first cell sits half a step in from the edge.
- Tone of a cell: raw = dark ink ? 1 - L : L; normalized = dark ink ?
(high - L) / (high - low) : (L - low) / (high - low), clamped 0-1 (use
raw alone when high - low < 0.025); mix = 0.72 + 0.14 s;
t = raw (1 - mix) + normalized mix. Then a tone curve: black = 0.22 s,
white = 1 - 0.08 s, lev = clamp((t - black) / (white - black), 0, 1),
tone = lev (1 - s) + smoothstep(lev) s.
- Skip cells with tone <= 0.02. Character = chars[round(tone * 16)]; skip
spaces.
- Ink: the sampled color lifted in HSV: value += (1 - value) *
(0.24 + 0.18 s + 0.24 tone); saturation *= 1 + 0.16 + 0.2 s (clamped).
Alpha = clamp((0.12 + 0.7 tone) (0.72 + 0.28 s), 0.06, 0.94).
Dither (ordered, Bayer 4x4):
- pixel size = max(2, round(range(size, 2, 9))) (7 instead of 9 under
720px); pattern scale = max(1, round(range(spacing, 1, 4))); levels =
max(2, round(range(strength, 8, 2))) per channel; blend =
range(strength, 0.32, 1).
- Matrix [[0,8,2,10],[12,4,14,6],[3,11,1,9],[15,7,13,5]]; threshold =
(m + 0.5) / 16, indexed by floor(x / pattern scale), floor(y / pattern
scale).
- For each block sample its center, push saturation away from luma by
1 + 0.06 + 0.18 s, then per channel: scaled = c / 255 * (levels - 1);
q = floor(scaled) + (fraction > threshold ? 1 : 0);
out = c + (q / (levels - 1) * 255 - c) * blend.
- Write the blocks into a canvas of box / pixel size and draw it scaled up
with image smoothing off.
Mask and layering:
- The mask is a vertical gradient with stops at the top, 42% and the
bottom. Base opacities: on a light page white 54/20/38%, on a dark page
black 44/20/50%. For a mask value v: stop = v <= 50 ? base *
(v / 50)^1.08 : base + (100 - base) * ((v - 50) / 50)^1.08.
- ASCII: draw the (recolored) picture, the mask, then the characters.
Dither: draw the dithered picture, then the mask on top.
- Changing filter or palette cross-fades from the last frame over 150ms.
Palettes (four stops at 0, 0.4, 0.74, 1):
- meadow: #2B361F, #677D45, #AEBA78, #E5E1BD
- blue: #192A4D, #326AC5, #A2C2F1, #EDF3FC
- amber: #3C200F, #A74C15, #EF9A46, #F7E7CA
- lavender: #292144, #6552AB, #B3A5E4, #F0ECFB
- mono: #151515, #575757, #A6A6A6, #ECECEC
Rendering:
- Draw into a canvas laid over the asset with the same size and corner
radius, at devicePixelRatio capped at 1.6x and about 2048 x 2048.
- Image: redraw when a setting changes (debounced 72ms).
- Video: redraw on every new frame (requestVideoFrameCallback, falling
back to requestAnimationFrame) while it plays and is on screen; keep
the video element underneath as the source.
Control panel:
- Place it next to the asset, below it when there is room, otherwise
beside it, never covering it. About 420px wide, centered on the asset.
- Rows, in order: Mask effect, Filter (two tabs: ASCII | Dither), Color
(round swatches), Sample tones (ASCII only), Sampling strength, Size,
Spacing (shown with %).
- Each row 46px tall: a 136px label column, then the control. Slider: a
22px track with 8px radius and a filled part, a 9x30 rounded-rectangle
thumb (3px radius), a number box with the value centered, and a snap
mark at 50. Swatches: 22px circles, each filled with its palette's two
middle stops at 135deg (Original: a tiny crop of the asset); the
selected one gets a ring with a small gap.
- Every change applies live while dragging. Add a "Reset to recommended"
action that restores the values above for the current filter.
- Follow the page's light or dark theme and the project's tokens.




















