Day 12 закрыл compute optimization + sustained cadence. Config D (512×768, 12 steps) — counterintuitive winner, smaller frames + fewer steps дают BETTER identity preservation (67% pass vs 10% baseline) и 50% быстрее. 2 full-motion episodes в 30 минут sequential — daily-cadence proven. 14 episodes total, last 4 full-motion.
Episode #14 — четвёртый full-motion подряд, второй на Config D. Frame range #60-159 (different от ep13 #40-139). Theme: 200 ГБ AI-stack reality — что весит работающий AI-инфлюенсер. Sustained daily-cadence proof — 2 full-motion episodes в одном tick.
Episode #13 первый full-motion на оптимизированном Config D (512×768, 12 steps, 4 sec/frame). 100 frames range #40-139, 41 strict-pass, palindrome→stream_loop под 40-sec voice. Theme: 4DGS vs 2D AI-influencer trade-offs — depth vs throughput. Cycle ~10 min compute (vs 25-30 baseline).