Cases: what it looks like in practice
No promises here, only work already done and numbers from our own hardware. Each row is the same person in different settings.
Numbers from our own hardware
71 sper frame with a character and face refinement
39 sper product frame
≈460finished frames per night shift
1376raw frames per shift before selection
0.47–0.69machine face similarity across frames, threshold 0.45 — all 12 pairs above
up to 1824 pxsource resolution
A face for a men's clothing line
- Task
- Frames of a man for product cards and the feed: office, street, gym. No shoot was ever arranged.
- How it was done
- The face was locked once from a studio portrait. After that we named scenes and got frames.
- Result
- One person in three settings. Machine recognizability check: 0.47–0.68, threshold 0.45, every pair above it.
A face for a women's line: store and fitness
- Task
- The same for the women's line: showroom, cafe, gym.
- How it was done
- Same order: a reference portrait, then scenes from a list.
- Result
- The row is one person: similarity between frames 0.47–0.60, every pair above the 0.45 threshold.
A brand character for social media
- Task
- The same person across the feed: office, street, coffee shop, studio.
- How it was done
- The character came from a selection of six face options made by the client.
- Result
- A finished character you can come back to in six months and shoot a sequel.
Product card: one item, four shots
- Task
- A store needs a set for every item, not a single photo: the main shot on white, the item in real life, a close-up and a frame with room for captions. A studio shoot means shipping the item.
- How
- The item was described once in words. After that only the scene changed: background, interior, close-up, empty space.
- Result
- A set of four shots of one item, 55 seconds per frame on our own card. The item travelled nowhere, no studio was booked.
