22 seconds, 11 shots, 20+ characters — no rerolls
TanLuAI had Codex write the storyboard, video prompts and most character images, used Blender blockout video for camera moves, and rendered the 22-second scene in one pass with Seedance 2.5.
Twenty-two seconds, eleven shots, more than twenty characters, generated in one go with no rerolls. The clip is impressive, but the workflow is the real entry.
TanLuAI's inputs were small: a reference clip of a famous movie scene for the camera language, a new plot, and two images of the protagonist. From there Codex designed the storyboard, wrote the video prompts and generated most of the character images; even the scene images came from Codex's prompts. A Blender blockout video set the camera moves, and character images, blockout and prompts all went to Seedance 2.5 for the final render.
The author frames it as a process for people without a director's or cinematographer's instincts — and shares the full instructions and prompts in the replies. The skill being sold is not drawing or shooting; it is directing the tools.
Made with
探路AI
@TanLuAI
基于下面文章的教程，我做了一个升级版 流程，适合没有导演和运镜思维的朋友。 整个片段 22秒， 11个分镜 ，20+个人物 由GPT 6+Blender+Seedance 2.5 一次性生成，没有二次抽卡 从分镜脚本设计，到AI视频提示词 到大部分角色图的生成，全部由Codex完成 （完整指令和提示词 见评论🔽 🔽 🔽） https://t.co/RJfoDza8zr
976 likes
- likes
- 976
- replies
- 33
Captured when this was curated, not live. —
