Memory Storyboard

Say a memory, get a storyboard built only from your real NAS photos and videos - every scene cites its source, then save it and make a matching album after you confirm.

Comment

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Memory Storyboard

Inspiration

Families keep years of photos and videos on a NAS, but reliving one event — "our Jeju trip with Mina last summer" — still means scrolling thousands of files. AI recap tools help, yet they often add scenes, places, or feelings that never happened. I wanted a recap you can trust: a story where every line points back to a real file.

What it does

Memory Storyboard turns one spoken memory request into a short, ordered storyboard (6–12 scenes) built only from photos and videos that really exist in the NAS photo library.

  1. Confirms who, when (required), and where.
  2. Finds matching media and picks scenes spread across the event, describing only what is visible.
  3. Drafts the storyboard; each scene cites its source item.
  4. After the user approves, saves the storyboard as a text file.
  5. After the user approves, creates an album with exactly the scenes used.

If nothing matches, it names the condition that removed everything (for example "no photos of Mina between June and August") and stops instead of substituting other media.

How I built it

A parent orchestration Skill written with the official uliya-skill-kit 0.7.3. It delegates all NAS work to registered experts — ug-photo → photo-agent, ug-document → document-agent, ug-knowledge → knowledge-agent — and passes only the real upstream result to each next step. Saving and album creation are state changes that wait for user approval. All five delegation states (completed, failed, cancelled, waiting_user, waiting_approval) are handled explicitly.

Challenges

  • Keeping the story honest: a source check rejects any scene whose item id is not in the selected list, with one rewrite attempt before the draft is marked unverified.
  • Fitting the official authoring contract: the kit's static checker now reports static-pass (0 errors, 0 warnings).

What's next

Testing on a real UGREEN AI NAS: person-name matching, video description quality, and album ordering.

Disclosure

No third-party components. AI tools: Claude (Anthropic) via Claude Code with the official uliya-skill-kit, for drafting and checking the Skill.

Built With

Share this project:

Updates

Submission history