Inspiration
On social media, there's a huge trend of then-and-now photos. I think as humans we really love seeing our progress over the years. A lot of families want to recreate old photos but never actually do. Jīnxī makes it easier to do that.
What it does
Jīnxī is a Uliya Skill for UGREEN AI NAS. It has two modes.
Before the occasion. You tell it the occasion and who is coming. It finds past photos of those people. It offers 3 to 5 ideas, each marked Recreate or Reinterpret. For each idea you pick, it writes a Recreation Recipe: who stands where, the angle, the lens and the light.
After the occasion. It finds the new photos and matches them to the Recipes. You confirm each pair. It tags the series and builds a Then & Now album.
How we built it
The Skill is Markdown only, built to the Uliya Skill Development Standard. It has no scripts and no outside services.
It delegates to three UGOS experts. The photo expert handles face groups, search, EXIF, the vision model and albums. The file expert handles tags and notes. The knowledge expert saves the Recipes.
We built it with uliya-skill-kit 0.7.3, then checked it against kit 0.7.7. It passes both checkers with 0 errors and 0 warnings. Claude helped us draft and review the files.
Challenges we ran into
We had no NAS hardware in Round 1. So we traced every step by hand against the capability catalog.
Accomplishments that we're proud of
- Proud of the idea generation without the real hardware
What we learned
- Limitations and also how powerful skills can be
What's next for 今昔 (Jīnxī)
- Run it on real NAS hardware and fix what breaks.
Built With
- ugreen
