Dev finds Claude-researched beauty algorithm beats CapCut's approach
vista8 · x · 2026-11-06
Indie developer vista8, adding beauty/whitening filters to the Cap screen-recording app, had his agent study CapCut's approach and then compared real frames with AI analysis — concluding CapCut's beauty filter performs worse than what Claude produced directly through research. The attached screenshot shows Claude's full quantitative report: candidate SkinWhitener vs. the current slider on facial brightness ΔL, saturation ΔC, background contamination (68% of wall pixels brightened vs. 0%), plus three concrete flaws and a recommendation not to ship it, with notes on how to split the slider if needed.
More from coding & agent
- Halai Play lets anyone build and ship games via MCP with Claude or ChatGPT — aziz4ai · 2026-11-07
- Open-source tool turns Garmin health data into daily LLM agent check-ins — unixterminal · 2026-11-07
- Microsoft Ignite 2026 packs five agent-focused sessions in San Francisco — lee_stott · 2026-11-07
- Cua Driver Bench: 58% pass rate on background macOS desktop tasks vs Codex 33% — TianbaoX · 2026-11-07
- YC-backed Trepo debuts kitchen hardware that lets AI agents shop for groceries in the background — ycombinator · 2026-11-07
- Cognition ships Devin Review 2.0: 5× cheaper, 1.5× faster PR review — smtabatabaie · 2026-11-07