Frame-accurate footage check
All 88 shots checked for format and length, then joined and compared with the master frame by frame — all 18,708 frames match.
Dosan's Gold Mountain: AI imagery for the years no camera recorded
A documentary of about 13 minutes on the gold-mining years around Jiufen and Jinguashi and one generation's story. Almost no footage survives from the early years, so AI-generated imagery fills the gaps; where records exist we use licensed archive photos. We handle footage checks, editing, narration recording, music editing, subtitles and grading — the film is in the edit now.
The challenge挑戰
What we set out to fix
Decades without footage
Early mining life left almost no footage, yet the story needs pictures.
Keeping real and generated apart
Real people must never get AI-generated faces, and every archive photo has to match its licence.
Not a frame out of place
The edit list, narration script and subtitles have to agree exactly.
What we built做法
How we built it
AI face rule
No AI faces for real people; shots that drifted from their prompt and show a clear face are flagged for the director.
Narration & subtitles
63 narration lines and 178 subtitles checked word for word across three documents; each subtitle stays within 16 characters, names and numbers unbroken.
Grading & delivery spec
Colour tags and pixel aspect unified, delivered to spec.
Results數字
Where it stands
88
shots (50 AI · 24 photos · 14 cards)
~13 min
running time
63
narration lines
178
subtitles
Built withStack
- FFmpeg
- Python
AI Film & ContentAI 影音製作
Want something
like this?
Start with a 30-minute call: we'll understand your situation, then propose the smallest useful version.