Frame-by-frame: 'high' reasoning tier yields smoother video motion than 'medium'
gabrielchua · x · 2026-11-05
Gabriel Chua compared video generations at low, medium and high complexity tiers frame by frame.
- All tiers land on frame 11, but the motion paths differ.
- At medium, the head is already above the knees by frame 9; high stays tucked longer and opens out on frame 10 before the feet touch down, producing smoother animation overall.
- Slowed down, the low version is nearly horizontal in the air on frame 12 and standing the next frame, while medium and high have already completed the rotation.
Takeaway: higher tiers allocate motion timing better rather than just shifting keyframes.
Related event: Frame-by-Frame Tests of GPT Image 2.6 Complexity Settings(3 posts)→
More from Multimodal
- Unverified leak: OpenAI's GPT images 2.6 reportedly generates animated images — aziz4ai · 2026-11-05
- Full prompt template for making looping GIF stickers with GPT-image-2.6 — op7418 · 2026-11-05
- GPT-image-2.6 launches with native GIF generation, demoed on viral fish stickers — op7418 · 2026-11-05
- AI video budgets should be measured by cost per finished second, not render price — NoaBlake · 2026-11-05
- Gemini 2.6 Instagram-grid vacation photos spark 'RL-optimized for male gaze' joke — flowersslop · 2026-11-05
- OpenAI ships ChatGPT Images 2.6 with GIFs and reference images, plus 50% faster API image models — gdb · 2026-11-05