AI summary by TwinMind
Why dark video is a terrible mess
Tom Scott
6m 15s
Color Limitations in Digital Video
- Modern digital video limited to ~16.7 million possible colors
- Screen technology converts compressed digital signals to instructions for red/green/blue lights
- Standard 8-bit color depth provides 256 shades per color channel (2^8)
- Results in 256 × 256 × 256 = 16.7 million total colors
- Professional digital cinema uses 10-bit (1 billion colors) but consumer content remains 8-bit
- Expensive HDR monitors can't fix color banding from 8-bit source material
Mathematical Color Distribution Problem
- 1920-pixel wide video with only 256 green shades creates color bands every 7-8 pixels
- Dark scenes severely limited: only ~6 different dark green shades available across entire picture
- Bright colors have more available shades, making banding less noticeable
Human Visual Perception Impact
- Color banding exists in bright gradients but appears invisible to human eye
- Relative difference perception: 201→202 feels tiny vs 1→2 (doubling)
- Absolute difference same (1 unit) but perceived magnitude differs dramatically
- Dark scenes make relative differences more apparent, exposing banding
Dithering as Mitigation Technique
- Gradual color transitions instead of hard lines between color bands
- Creates fuzzy boundaries that appear to blend from distance
- Natural scenes benefit from camera noise and sensor characteristics
- Real-world filming automatically creates dithering effect
Video Compression Challenges
- Raw HD video requires ~1 gigabit/second uncompressed
- All streaming services (YouTube, Netflix) use lossy compression
- Encoder software prioritizes viewer attention areas:
- Bright, sharp, well-lit, in-focus, moving elements
- Dark/out-of-focus areas deprioritized
- Limited bitrate allocation (50 kilobits per frame example)
- Compression removes dithering, creates solid color blocks in dark areas
- Results in visible banding in scenes viewers weren't expected to scrutinize
