Two-person team spent 12 days and $400+ credits making a 6-min AI short film with Codex + Seedance 2.5
APPSO · wechat · 2026-09-06
APPSO's two-person team spent 12 days and over ¥3,000 in credits building a 6-minute AI short film with Codex and Seedance 2.5, and published a full post-mortem. The workflow: refine the screenplay with Codex (making AI describe how characters act, not just what happens); build an asset library — the most time-consuming step, where direct film-screenshot references triggered copyright review and forced a two-day rework around abstract visual traits; generate shots with three-layer prompts (overview, second-by-second timeline, constraints) using Seedance 2.5 in Xiaoyunque, plus free-canvas for asset management and local-edit for precise fixes; finally assemble in editing since AI rarely produces usable takes in one shot. They compare their process to the 95-min AI feature HellGrind (15 people, 14 days of generation, sub-$500k budget), concluding the tools change but the filmmaking logic doesn't. Prompts, Skills and assets are shared.
More from coding & agent
- Dev proposes open-source self-healing agent loop: trajectory eval turns failures into skills — EnchantedHawk · 2026-09-06
- Polygres turns your Postgres into a hybrid search context layer for AI agents — Scobleizer · 2026-09-06
- If you can't automate your dev workflow with AI, that's a bad sign — paul_cal · 2026-09-06
- Running Qwen 3.8 27B Locally to Drive Blender via MCP and Build a 3D Llama — jacek2023 · 2026-09-06
- tokenbender's trick: ask the model for two prose narratives to detect long-task drift — tokenbender · 2026-09-06
- KV cache pressure tool shows vLLM's advertised 2M-token cache can retain 3M after fixes — t4a8945 · 2026-09-06