Traditional algorithms rely on mathematics to decide which pixels to keep and which to discard. The new generation uses Neural Networks. An AI transcoder looks at a video game stream and "knows" that the sky in the background is less important than the gun in the player's hand. It applies perceptual optimization, keeping the subject sharp while heavily blurring the irrelevant background, achieving massive bitrate savings without perceptual quality loss.
| Parameter | Impact | |-----------|--------| | | Smaller GOP = better seek but lower compression; live usually 2–4 seconds. | | Latency mode | Low-latency (tune=zerolatency in x264) disables B-frames & lookahead. | | Rate control | CBR (for ABR ladders) vs VBR (quality, but buffer risk). | | Lookahead | Allows better bitrate distribution but adds frames of delay. | | Deinterlacing | Necessary if source is 1080i (broadcast) → 720p/1080p. | ip video transcoding live
| Resolution | Bitrate (H.264) | H.265/AV1 bitrate | |------------|----------------|-------------------| | 1080p | 4–6 Mbps | 2–3 Mbps | | 720p | 2–3 Mbps | 1–1.5 Mbps | | 480p | 1–1.5 Mbps | 0.5–0.8 Mbps | | 360p | 0.5–0.8 Mbps | 0.3–0.5 Mbps | | 160p (audio only) | 64 kbps | 64 kbps | Traditional algorithms rely on mathematics to decide which