Four upscaling models: general, portrait, anime, night

"AI upscale" sounds like one button. Skin, cel lines, and night noise are different priors. The AI Video Enhancer (and the same family on AI Ultra HD) ships General Repair, Portrait Repair, Anime Repair, Night Scene Enhance, plus Special Scene Repair for fog/rain/colorize/film grain on the restore old video path. This page is which prior to load — not a model-family name drop.

• A natural-photo prior invents grain inside flat anime fills and softens the line
• A general prior is most likely to plasticize a face — the one region viewers police
• Night noise looks like "detail" to a daylight prior, so a generic pass can get dirtier

• General — default for messy real-world plates: old exports, screen recordings, social re-encodes. Start here when unsure
• Portrait — reconstruct the person, not a beauty filter. Interviews, family, wedding faces
• Anime — protect hard edges and clean flats. The usual way a general pass makes animation worse
• Night Scene — separate grain from structure before rebuild. Concerts, surveillance-like ISO, night streets
• Special Scene — weather, colorize, film grain. Full archival order lives on restore old video

1. Person is the subject → Portrait
2. Animation / illustration → Anime
3. Dark and noisy → Night Scene
4. Everything else → General, then 1s/3s/5s after download if two candidates are close

• Priors rebuild plausible detail, not lost photons — see 480p to 4K
• Mixed live-action + heavy illustration: pick the subject. There is no perfect hybrid
• Unread text still cannot be guessed. Tiers are speed stops — why tiers are speed tiers