One image, 40 nested worlds: creator shares the full workflow behind his zero-cut infinite-zoom film
tetsuoai · x · 2026-10-21
Creator tetsuoai published the making-of for his infinite-zoom film: the whole piece is a single image containing 40 nested worlds, with the camera pushing toward one fixed point — no cuts, no fades, 16,423 frames and a claimed 8.9 quintillion× zoom.
Key workflow details:
- Map the song: Demucs splits audio into vocals/drums/bass, faster-whisper large-v3 timestamps every sung word, and Python scripts detect kicks, drops and quiet parts so visuals sync to music.
- Lock the look: iterate hero frames, refine with masked repaints, upscale best frames to 100 MP with Clarity Crystal for close inspection, then distill palette, line, light and finish into one style paragraph reused in every prompt.
- Build inside-out: the center world is made first, each new world painted around the previous one so every world sits in the middle third of its parent.
He also notes the film plus instructions can be pasted into Grok Build, Claude or AgenC to reproduce the pipeline.
Related event: AI creator's one-shot zoom video packs 40 nested worlds into a single image(2 posts)→
More from coding & agent
- Ant International's FlowState treats execution state as memory for long-horizon LLM agents — ant-intl · 2026-10-21
- Google's TEAL evolves reusable analytics tools for coding agents without ground-truth labels — google · 2026-10-21
- Tencent's ASAP evaluates agentic capabilities in pretrained base models before post-training — tencent · 2026-10-21
- What Agent Failure Slipped Past Your Tests Last? Building an Open-Source AI Verification Tool — Practical_Green5775 · 2026-10-21
- GEICO's $1,000-a-day AI war room: mapping legacy claims code with agent fleets — AI Engineer · 2026-10-21
- InternScience releases Sage-A1.5, a 35B MoE agentic model fine-tuned from Qwen3.6-35B-A3B — Skyline34rGt · 2026-10-21