Dev builds anime upscaler that redraws lines and gradients, not invent detail
Agitated-Pea3251 · reddit · 2026-10-05
A developer is building an upscaler for anime and illustration images, arguing that an upscaler should not "invent" content but only fix lines and gradients, which make up most anime and illustrations. The model doesn't need to know whether a pixel belongs to a face or the background: if it sees a black line, it redraws that line at 4x resolution; if it sees a gradient, it reproduces the same gradient at 4x. Theoretically all anime/cartoon-style images are just sets of lines, gradients and uniform colors, so no high-level understanding is needed. The author asks how it compares with the current best open-source models and invites feedback.
More from Multimodal
- Ego2Act benchmark: best video model completes only 67.8% of 110 real-world manipulation tasks — zmkzmkz · 2026-10-05
- Vivix W1 breaks AI video's speed-cost tradeoff to power endless playable AI livestreams — FellMentKE · 2026-10-05
- Prompt template: Silent Era Dream with film grain and faded monochrome hues — LudovicCreator · 2026-10-05
- Tutorial: Building a Floating City in Blender Using Claude Opus 5.5 and Blender MCP — sidahuj · 2026-10-05
- Voxel-style videos straight from Opus 5.5: custom skill output called commercially usable — AlchainHust · 2026-10-05
- Fable 5.5 single-handedly produces a Blender animation, from music to rigging — Angaisb_ · 2026-10-05