A picture is just numbers.
A small desktop app that gives a whole pile of photos a retro film look in one go: seven sliders, a before and after, and one button that runs the batch in the background. It came from an aha moment — an image is a grid of numbers, and numbers are something I can change with code.
01Why it exists
I made this for fun: I wanted to put filters on my own photos. Somewhere along the way it clicked that a picture is just data — numbers for every pixel, arranged so that our eyes read them as an image. Everything on a screen is code, and if it’s code, I can change it. That was my aha moment.
The first try was Pixelated Kisses, made for one look: a Joji album cover. It turns a photo into black, a warm grey and a hot red, adds grain, and can pixelate it. It’s on GitHub (opens in a new tab), with ready-made downloads for Mac and Windows.
Two weeks later I wanted something more general — a retro film look I could tune — and I wanted it for a whole folder at once, not one photo at a time. That became Filter Generator: seven sliders, eight steps, one click for the whole pile.
It stayed with me, too. This little app is one of the things that started RideLens, the car photo editor I’m building for iOS.
02Try it
This is the app’s filter, ported line by line from Python and running on your device: the same eight steps, in the same order, rounded the same way after every one. Point at the photo to follow one pixel through the whole pipeline.
Loading the photo…
Adrian’s
pic filter
Filter settings
Pipeline · one pixel, step by step
Pixel —
Hover or tap the photo to read a pixel; arrow keys move it one pixel at a time (Shift for ten).
A line-for-line port of the app’s apply_filter_logic() and of Pixelated Kisses’ filter, checked byte for byte against NumPy and Pillow, running in a background worker. It runs the same maths at preview size (900 px wide), so blur, grain and glow — all measured in pixels — look stronger than on a full-size photo. The grain is random as in the app, but drawn once per photo and rescaled as you drag, so it doesn’t flicker; “New grain” draws a fresh set. Your own photo never leaves this page.
03What it does
Drag photos anywhere onto the window, or pick them: JPG, PNG, BMP and WebP. Phone photos are turned upright from their EXIF data as they load.
Grain, warmth, brightness, shadow fade, lens softness, glow and vignette — each with its value printed above it, and sensible defaults.
Every photo runs through the filter on a background thread while a red progress bar fills, so the window stays responsive.
Flip between the original and the result, and step through the batch with Prev and Next — “Image 2 of 5”.
One photo opens a Save dialog. Several go into a folder you choose, each as name_filtered.jpg. Restart clears the lot.
Pixelated Kisses went out as zipped builds for macOS and Windows on GitHub Releases, made by GitHub Actions with PyInstaller.
04How it’s built
One Python file of 535 lines. Pillow opens, blurs, resizes and saves the photos; everything else is arithmetic on a NumPy array of pixels.
# 4. Vignette
if vignette > 0:
v_factor = vignette / 100.0
h, w, _ = img_arr.shape
y_coords = np.linspace(-1.0, 1.0, h)
x_coords = np.linspace(-1.0, 1.0, w)
Y, X = np.meshgrid(y_coords, x_coords, indexing='ij')
dist = np.sqrt(X**2 + Y**2) / np.sqrt(2.0)
mask = np.clip(1.0 - (v_factor * 0.45) * (dist ** 2), 0.0, 1.0)
img_arr = (img_arr.astype(np.float32) * mask[:, :, np.newaxis])
img_arr = np.clip(img_arr, 0, 255).astype(np.uint8)
# 5. Film Grain (Monochromatic Noise)
if grain > 0:
h, w, _ = img_arr.shape
noise_scale = (grain / 100.0) * 20.0
noise = np.random.normal(loc=0, scale=noise_scale, size=(h, w, 1))
img_arr = np.clip(img_arr.astype(np.float32) + noise, 0, 255).astype(np.uint8)05Build notes
Warmth doesn’t visit pixels one by one. The app works out 256 answers per channel once — a straight line blended with a retro curve — then applies them with NumPy indexing: lut_r[img_arr[:, :, 0]]. Even at warmth 0, floating point leaves 24 of those 256 values one lower than they went in.
Filter Generator draws one random number per pixel and adds it to red, green and blue alike, so the grain changes brightness, not colour. Pixelated Kisses drew separate noise for each channel, which speckles the image with colour.
Chromatic aberration is the three channels pulled apart: red scaled up by as much as 1.4% and cropped back, blue scaled down and centred on black, green left alone. It rides on the Lens Softness slider, so a softer lens also fringes more.
Only pixels with a luma above 180 feed the bloom. That layer is blurred by 5 to 25 pixels and added back at up to 75% strength — so a halo forms around bright paint and lights, not over the whole picture.