A reconstruction is an estimate
A single photograph shows one view of an object. It does not record the exact back, underside, wall thickness or physical dimensions. Image-to-3D systems estimate those missing surfaces from visual patterns learned during training. A result can be useful for a creative scene or a prototype while still differing from the real object. It should not be treated as a measurement or an engineering scan.
Choose one object and show its shape
Use a sharp photo of one complete object on a simple background. A three-quarter view often reveals more shape than a perfectly flat front view. Leave some space around the object, avoid clipping its edges, and make sure important parts are not hidden behind a hand or another item. Reflective, transparent and very thin objects are especially challenging.
Understand this tool's workflow
MakePip uses a private processing worker for reconstruction. Your selected image is sent for processing only after you confirm the notice. The worker removes the background, reconstructs a mesh and exports a GLB file. Detailed shape uses TripoSG to estimate a finer mesh with a neutral material; it does not generate a texture. Quick draft uses TripoSR for a rougher mesh with estimated vertex colors. Neither mode guarantees a rig, animation or a production-ready asset. Existing transparent cutouts keep their alpha instead of being segmented again.
Inspect more than the front view
Rotate the result in the interactive viewer and look at the back, base and silhouette. Watch for floating pieces, fused parts, holes, surface noise and a shape that becomes implausible from another angle. A recognizable front view is not enough to establish that the entire mesh is correct. Try a cleaner input photo if the overall structure is wrong.
Take GLB into the next step
GLB packages glTF scene data into a binary file supported by many 3D editors and web viewers. Save the download within the result's 24-hour window. In a 3D editor, set an appropriate scale, simplify the geometry if needed and check orientation. For 3D printing, additionally inspect watertightness, wall thickness, overhangs and the printer's material requirements.
Could an object video help?
A video can supply real views of the back and sides that one photo cannot show. For multi-view reconstruction, keep the object stationary and walk slowly around it, maintaining overlap, focus and even lighting. Capture a complete lap, then a slightly higher angle. Avoid moving subjects, reflections, motion blur and changing zoom. Turning an object against an unchanged background needs careful foreground masking. MakePip’s Video to 3D tool samples 36 frames in your browser and uses multi-view photogrammetry to align cameras, recover depth and build a colored surface. It runs separately from the single-photo AI modes and uses the same private processing queue. The original video and its audio are not uploaded. Clips must be 5–90 seconds and no larger than 200 MB locally. A scan can take up to 30 minutes after processing begins, and capture quality strongly affects whether it succeeds.
Plan for worker availability
The reconstruction tools share a processor. They accept queued requests while it is busy or offline. You can join the queue while the worker is busy or offline; requests that cannot start within 12 hours expire. Queue position shows saved requests ahead of you, rather than a guaranteed start time. Optional ready emails become available once the sender mailbox is configured; you can always return using the same browser. Daily and queue limits keep this experimental feature manageable. Browser-only PDF and image tools continue working independently of the GPU worker.