Inspiration

Photographers get their "must-have" shots from three places: the contract, the client's email and a shot list. After a long shoot there are thousands of photos and no time to check every request by hand. So the missing shot (say, the posed photo with Grandma) usually turns up weeks later, when the client opens the gallery and it's too late to take it.

We wanted the NAS the photos already live on to answer one question at a break in the shoot: did we get every shot they asked for?

What it does

Got the Shot is a Skill for the UGREEN AI NAS. Every shot they asked for, and proof you got it.

  • It reads the contract, brief and shot list (or a list you type) and merges them into one checklist, marking which lines are must-haves.
  • It searches the shoot for each line and checks the best matches with the vision model.
  • It shows a coverage board (covered, thin or missing) and a reshoot list, must-haves first, while you can still take the shots.
  • After you say yes, it tags the best photos, makes a selects album and saves a coverage report for the client. It never moves, edits or deletes your originals.
  • Archive mode checks past shoots for repeat clients: "You already have 5 of 8 shots."

How we built it

A Skill is just instructions in Markdown. The NAS assistant (Uliya) reads them and hands each step to the NAS's built-in experts:

  • the document expert reads the briefs
  • the photo expert searches photos, checks them with the vision model and makes the album
  • the file expert finds folders and adds tags
  • the knowledge expert saves the report

Each shot line is scored from the vision check's YES answers. If a line asks for $q$ photos (for example $q = 3$ angles):

$$ \text{status} = \begin{cases} \text{Covered} & \text{if } #\text{YES} \ge q \ \text{Thin} & \text{if } 0 < #\text{YES} < q \text{, or there are only UNSURE answers} \ \text{Missing} & \text{if nothing was found, or every answer was NO} \end{cases} $$

We checked every step against UGREEN's official capability list and ran the official checker until it passed with zero errors and zero warnings. Then we walked all three of our test cases through the Skill step by step and reviewed each other's changes in a shared repo. AI coding assistants helped us draft and check the Markdown (details are in the package's disclosures file).

Challenges we ran into

  • Search works on what the NAS already recognized. The search that stays inside one shoot finds people, objects, text and times, not free descriptions. So we split every request into those clues, and use the event's timeline for small details like the rings.
  • The official capability list changed our plan. One tool we planned to search with only looks up names, and the album tool needs photo IDs, not file paths. We rewrote those steps to match.
  • Being honest. One photo can't prove "3 different angles", and the vision model shouldn't be asked who someone is. So the person search confirms who's in the photo, the vision check only confirms the scene, and anything not found says "check manually" instead of "you missed it".
  • One shot at testing. Finalists get a single 24-hour session on a real NAS, so we wrote probes, backup plans and pre-tested fixes ahead of time.

What we learned

The best AI-NAS tools lean on what the NAS already knows: faces, objects, text and capture times. They don't need a model to look at every photo. And for a working photographer, "not found, check manually" is more useful than a confident guess.

What's next

If we're finalists, we'll run all three test cases on a UGREEN iDX6011 Pro and record the demo. Reading requests from a client's voice memo or mood board is pending NAS testing, and we'll only add it if it works.

Built With

Share this project:

Updates

Submission history