Can a sub-cent AI model turn a 10-second film into a live 3D scene?
The film: a Japanese pagoda at night. Torii, stairway, sakura, lanterns, full moon, five camera shots. The brief: rebuild it in one HTML file. No 3D assets, just code. 18 tries. $0.42 in total. GPT-6 Luna: 8
English Version Generation mode: I2VA (Image-to-Video with Audio)｜Duration: 5.00s｜Aspect Ratio: 9:16
Keyframe Alignment: <Picture 1> aligns with 0.00 seconds of the target video and serves as the actual opening frame. Fully preserve the girl’s facial identity, pink wet hair,
It turns out that Big Fatty Fish is a Sichuan girl, hahaha😂
Over the weekend, I continued to improve DeepSeek Dock to make the sidebar more user-friendly.
I discovered the richness of the DeepSeek Harness open source community, and slowly DSH will become a small AI OS tailor-made for myself. Its functions only cover my daily work and entertainment.
You have reached the end of the archive
All of deepseek