Neural Network Classifies Digits in Pixel Art
A silent, single-shot pixel animation turns digit classification into repeating blue signal passes and red return pulses.
Starter prompt / by Refs
The creator shared no short prompt. This is ours for making a film like it with Claude.
Animate a tiny neural network learning to recognize handwritten digits, in crisp pixel art on a black background. A 28×28 input grid sketches a digit, three hidden layers of round nodes and thin connection lines glow as activations pass through, and output labels 0 to 9 light up by confidence. Train a real small network in JavaScript so predictions visibly improve while a loss readout falls. Use p5.js at low resolution scaled up without smoothing, and render one minute at 1920×1080 with ffmpeg.
Field notes / why it works
A white input mark takes shape in the first second, while the answer waits until about 0:03.7. The locked diagram makes each new sample easy to compare, and 15 internal graphic changes keep the single 56-second shot active. Blue forward paths, cyan predictions, and red return pulses give an abstract classification process a legible visual rhythm.
Plate 01 / The format in one breath
A white pixel sketch appears in a box on the left, signals travel through three dense columns of nodes, and one of ten numbered outputs lights up on the right. The repeated blue forward pass, cyan answer, and red return pulse turn a technical process into a visual story without narration or captions. Calling the red pulse a learning or correction step is inferred from its direction and timing; no explicit loss readout appears.
Why it works
- The sparse marks at 0:00 become a recognizable stroke by 0:00.25–0:01.0, opening a visual question before the first output at 0:03.7.
- The fixed left-to-right layout lets viewers compare each new sketch with its prediction; examples change around 0:11, 0:27, 0:41, and 0:49.
- Bright blue paths and red return pulses supply movement within one uncut 56 s shot; 15 measured graphic changes replace edit cuts.
Plate 02 / Format card
| Platform · aspect · length | X · 16:9 · 1280×720 · 60 fps · 56.0 s |
| Pace | 1 shot · 0 cuts/min · 15 internal graphic changes, about 16.1 picture changes/min · 56 s average shot |
| Script | No spoken words or on-screen sentences; output labels are digits 0–9. |
| Voice | None. |
| On screen | 100% code-rendered pixel-style motion graphic (based on appearance and post description). |
| Captions | None; no explanatory text cards. |
| Sound | Quiet tonal music-like bed, −22.5 LUFS integrated, −5.3 dBFS true peak; 129 BPM estimate has low confidence. |
| Structure | Sketch → forward signal → answer → red return pulse; repeated for several examples, ending without a CTA. |
Pro / 13 plates + recipe
The first plates are open. The full dissection is Pro.
Get the timed shots, type and sound specs, and the runnable recipe.
Open the full blueprint- Plate 03 / The hook, frame by frame (first 3 seconds)Pro
- Plate 04 / Structure (beat sheet)Pro
- Plate 05 / Shot grammarPro
- Plate 06 / Visual style guidePro
- Plate 07 / Voice and deliveryPro
- Plate 08 / Sound designPro
- Plate 09 / PackagingPro
- Plate 10 / Script templatePro
- Plate 11 / Production planPro
- Plate 12 / PromptsPro
- Plate 13 / Edit recipePro
- Plate 14 / QA checklistPro
- Plate 15 / Make it yoursPro
- Recipe / runnable promptPro