Compare
See what changes.
Review alternative crops of the same photo. A ratio and a position are separate choices, and both affect what survives.
Back to work
Crop Arena / Smart crop evaluation
What the crop leaves out
We needed the best crop. But people don’t take photos to fit our layouts.
A face near the edge. A group spread across the frame. A moment that matters because of what surrounds it. The same crop can preserve one photo and lose the point of another.
I vibe coded Crop Arena to make that problem visible, and make the case for investing in smart crop for the team.
- My contribution
- Built the evaluation platform through vibe coding.
- Purpose
- Make the case for team investment in smart crop.
- Approach
- Compare crop ratios and positions through manual, AI and heuristic evaluation.
- Deliverable
- A working prototype, shown in the full demo below.
The demo could not load. Open the video directly, or read the visual walkthrough below.
01 / The problem
A layout needs consistency.
A photo needs context.
Design needs to decide where a crop should sit. It is tempting to answer with a single ratio and a fixed position, then apply that answer everywhere.
But everyone frames a photograph differently. What belongs in the crop changes with the image: the people, their positions and the surrounding scene. At the scale of millions of photos, checking each one by hand is not an answer either.
My argument was that this needed an intelligent system, not one more universal crop rule. To get the team investment that work needed, I wanted to put the alternatives in front of people, not leave the discussion at the level of preference.
02 / What I built
Make the trade-offs
visible.
Crop Arena turns the question into an evaluation: compare different ratios and positions on actual photos, then examine what gets selected. The platform brings manual, AI and heuristic evaluation into the same workspace.
The point was not to assume that an AI judge would always be right, or that human judgment would settle everything. It was to test the same underlying question across those approaches: does one ratio or position really work for every photo?
Choose
Make a judgment.
The manual flow presents small groups, narrows the candidates and asks for a final top-three ranking.
Evaluate
Look beyond one winner.
Manual, AI and heuristic paths make it possible to explore the crop decision through different kinds of judgment.
03 / A companion to the recording
Inside the arena.
One photo, several answers.
The recording opens in the Arena workspace with an image list, evaluation tabs and progress tracking. The manual round shows several crops of the same selfie, with different ratios and positions. Selecting a crop marks it as best before continuing.
Narrow, then rank.
The manual comparison progresses through smaller groups to a final round. The top-three ranking makes the choice explicit, rather than treating every plausible crop as equally good.
Inspect the result.
The demo moves through heuristic results for different photos. Ranked crops sit beside a selection summary and rubric scores. The workspace also exposes an AI evaluation path; the recording is a prototype walkthrough, not a controlled benchmark of the three methods.
Change the photo.
Selfies, group portraits and event photos bring different framing problems into the same interface. Later manual comparisons return to the core question: what does this particular crop keep, and what does it leave out?
04 / The investment story
Build the case
before building at scale.
Crop Arena was the platform I built to argue for the work behind smart crop. Instead of asking the team to accept that fixed cropping was insufficient, I gave us a way to evaluate the alternatives and see the trade-offs on real photos.
The deliverable here is that working evaluation platform. The goal was investment in smart crop; this story does not claim that funding was secured, that smart crop shipped, or that one evaluator outperformed another.
Keep exploring
