Spider-Verse used a scikit-learn regressor trained on ~600 frames per character to ink its art
lmoroney · x · 2026-10-19
Laurence Moroney confirms machine learning was part of Into the Spider-Verse's production, revisiting a circulating clip.
- Sony Pictures Imageworks artists hand-drew ink lines; lines following predictable geometry (noses, ears, hands) fit modeling well
- The team trained scikit-learn's GradientBoostingRegressor inside Houdini on 600 artist-drawn frames per character, using camera angle and expression changes as features
- Artists started each shot from predictions and nudged them; those corrections became a second round of training data
Moroney frames it as a classic small model doing a focused job alongside artists, and suggests a replicable pattern: collect a few hundred examples of a repetitive manual task, train a simple regressor, and measure time saved on corrections.
Related event: Spider-Verse Used Machine Learning to Draw Its Signature Ink Lines(2 posts)→
More from Multimodal
- SenseTime co-founder: encoder-free native VLMs key step toward unified multimodal AI — liuziwei7 · 2026-10-19
- Reddit user shares "Brutus Part 1", a dark fantasy short film — akasan · 2026-10-19
- Krea2 vs MOSS vs Qwen2.1 vs Flux3 hands-on: Flux3 flops hard — CaptainBackwoods447 · 2026-10-19
- Animator uses image models as fast style tests, concludes hand-drawing wins — lmoroney · 2026-10-19
- AVC-Reward: an event-centric framework for evaluating audio-video generation consistency — ZhejiangUniversity · 2026-10-19
- Tsinghua and Z.AI study shows image-generation training expands unified multimodal models' knowledge — zai-org · 2026-10-19