You supply a topic such as the origin of the butterfly effect, a period such as 1961, and the core scientific conclusion. The agent runs `scripts/generate_video_materials.py` with theme, era, core conclusion and output directory arguments, and the script calls a language model, which takes roughly one to two minutes. Four files come back: a narration script of about 900 to 1100 characters, a storyboard table of 30 to 35 shots at four to six seconds each, a list of Veo2 prompts, and a character design sheet.
Reference notes cover the retro style parameters, the script structure and time allocation, the Veo2 prompt format and a multi-agent architecture, and template files for each output sit under the assets folder. The conclusion you provide must be accurate, because the generated copy is built on it. The description presents the package as suited to the Jimeng platform. The instructions are written in Chinese.