
What happened
The author piped a single drawing through ComfyUI into a 360° rotation video, logged 13 hair measurements in a JSON file, and let Claude Opus 5.5 generate Blender meshes, scoring each against the drawing with IoU and strand-width checks.
Why it matters
A model that passes every geometric check can still look wrong — a long-hair prototype passed with zero violations but read as a slitted curtain, forcing a redesign around closed strand meshes, layered depth, and lens-shaped cross-sections.
Summaries like this, every morning.
The project's most useful lesson is a measurement problem, not a modeling trick. The author built an inspection suite — face-rule violations, self-intersection inside a strand, contact with body and clothes — and treated all-zero as a pass. The long-hair prototype cleared that bar while looking like a slitted curtain, which pushed the pass condition itself to change from "the checks pass" to "it does not look like a curtain," tracked through thickness and layer overlap.
A second round of fixes came from looking at individual strands rather than whole sections. Splitting 29 strands into six types revealed that 21 of 29 still bent while thick at the tip, and that the median width 30 mm from the tip was 29.5 mm. The rebuilt version brought that to 0 of 29 and 12.1 mm, though the share of gaps showing the body through the tips rose from .091 to .153.
The tooling has a cost the author states plainly: texture-drawn lines are cheaper in time and token cost than editing hair structure and improve the look, but taste varies. Wave expression and sideways volume remain unfinished, with costume and animation work prioritized instead.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Visa launched its Agentic Commerce Benchmark, a framework to evaluate AI-powered shopping experiences for cons…

At the Next-Gen AI for Energy Summit in Houston on 3 November, Anthropic's Boaz Nur said AI will take over the…

Visa's Agentic Commerce Benchmark tested 22 AI models from 12 labs on shopping tasks like buying a TV or lapto…

Salesforce agreed to acquire Cresta, adding voice agents, real-time agent assistance, and conversation intelli…

In May 2026, SMFG said it will invest 100 billion yen in generative AI over three years, expanding a 50 billio…

The Japan enterprise agentic AI market was valued at $149.6 million in 2025 and is projected to grow at a 46.9…
