AI vs pro artist at 3D cartoon characters: 94/100 shape score, faces still wrong
ycombinator · x · 2026-11-08
TensorTest × Atrum had three AI models build the same cartoon character in 3D and compared each with a professional artist's version. The best build (GPT-6 Astra) scored 94/100 on shape similarity (F-score at 1% of character height) yet still got the face wrong: the artist's character looks serious with heavy lids and a slight frown, while the AI's looks wide-eyed. The face is only 3% of the character's surface area, so the score barely moves — overall shape is near-usable, but small high-semantics regions remain a clear weak spot.
More from Multimodal
- Naked mole rat driving a convertible: AI video goes viral for eerie realism — Megalania34 · 2026-11-08
- Custom YuE2 LoRA recreates 1980s British two-tone ska, released on Hugging Face — Robonotes1760 · 2026-11-08
- Gothic AI Portrait With Striking Detail Draws Community Praise — LudovicCreator · 2026-11-08
- Viggle-Remake hands-on: wiring its MCP to ComfyUI workers with Claude for full video remake — thefi3nd · 2026-11-08
- Prompt Card: a local tool that extracts prompts and settings from AI-generated files — lordstryfe · 2026-11-08
- Rumor: GPT Images 3.0 in early alpha, may jump to GPT-6-class base model — mark_k · 2026-11-08