Popout · Create on iPhone
Five photo-to-3D cases, with the imperfect results kept in
We reviewed a recorded batch of five synthetic marketing photos: a family, friends, a couple, a portrait and a corgi. The actual App outputs show where depth framing and generated model details still need inspection.
What was captured, and how
The source batch was captured on 2026-11-08. Its five original images were made with ImageGen and depict fictional people or a pet. They are demonstration inputs, not private customer photos or customer outcomes.
- Depth scenes were made by the actual signed Debug App 1.0 (3) on iPhone 15 Pro Max, iOS 27.0.1. The batch used on-device depth estimation, foreground segmentation and background filling.
- Each source was exported by the App at medium and gentle amplitudes. The depth videos are 1080 × 1440, 30 fps, 6 seconds. Frames below are from the gentle exports at the two swing endpoints.
- Figurines were created through the real App-to-cloud workflow in the simulator after photo-use consent. The generated GLB models were downloaded and rendered by the App; turntables were exported at 1080 × 1350, 30 fps, 6 seconds.
- Web images are full-frame resized JPEG copies. Source frames and model views were not retouched. This review uses the existing batch; no new App generation was performed for this article.
Inspect both endpoints for every source
Mother and child
Gentle movement improved head framing. The mother’s sole still approached the bottom edge. In the figurine, the child’s dress changes from yellow in front to red at the back.
Three coastal friends
A man’s head and some shoes remained cropped in the gentle export. The figurine’s rear trouser color changes to blue for one subject.
Lakeside couple
Shoes remained close to or beyond the lower boundary. One trouser leg shows color drift on the generated model.
Forest portrait
Hair and shoes remained near or beyond the frame. A tree edge showed segmentation jaggies. Full rotation still requires a visual model check.
Garden corgi
Ears, body and paws remained in frame at both gentle endpoints. A slight segmentation halo remained around the subject.
A family model shows why the back matters
A flat preview cannot reveal every later model defect. The figurine guide separates the input, preview and model inspection steps.
What this small review can and cannot establish
The five inputs were selected for marketing variety. They are not random photos, and source resolution, subject count, pose and background differ. There is no blinded rating or controlled crop, blur or lighting experiment. We do not report a success percentage or claim that a particular input feature caused a defect.
Lower amplitude reduced cropping in the recorded batch but did not guarantee full-body framing. All five physical-device depth scenes contained foreground and background layers. Earlier simulator outputs with invalid depth were excluded from public examples.
AR camera capture, production authentication and StoreKit purchases were not evaluated by this material batch. A future photo-quality experiment must vary one property at a time, preserve every output and define failure criteria before running.