Skip to content
EN
English 简体中文 soon 日本語 soon
AI design tools Site unreachable

SAM 3D

SAM 3D is a 2D image-to-3D model tool designed around generating 3D objects or 3D mannequins from a single 2D image.

Visit official site

What SAM 3D is

SAM 3D is a research model published by Meta. It rebuilds three-dimensional objects, and human figures too, from one flat image. What this entry recorded was a hosted service built on that model, and the domain it lived at no longer works.

The underlying capability is real and the wrapper is not, and separating the two is the whole point of this note. Searching for the name now returns a cluster of lookalike sites offering the same research model as a service.

What the model does

The idea behind single-image reconstruction is that a photograph contains enough information to infer the shape behind it, and the model produces a mesh from one picture rather than from a sequence or a scan.

Several independent sites have appeared offering this capability as a hosted tool, and at least one states outright that it is not affiliated with Meta. They advertise the same promise: upload an image and receive a model. Operators are not always identifiable, and the naming is close enough that a search result can easily be mistaken for an official one. Because the model itself is public, the same result can be obtained from a service that will still be running next year.

Who it is for

The capability fits environment and game artists, people visualising products, developers working in augmented reality, and hobbyists or researchers who need a three-dimensional asset from a snapshot rather than a scanner.

Anyone who came looking for this specific site should look elsewhere. The ability is what matters, and it exists in maintained services and in the original research release, neither of which depends on the address that is down.

Points to weigh

Do not enter credentials or payment details at the address recorded here. It returns a hosting fault instead of a product, which matches an abandoned deployment, and an address left like that is precisely what somebody else registers next. An old password or a stored card number left on that name would be handed to whoever controls it next.

The method has inherent limits worth understanding before any work depends on it. The back of an object is never photographed, so the model invents it, which makes everything hidden a reasonable guess rather than an observation. For a prop or a background shape that is usually harmless; for anything measured, printed or fitted it is not.

Last, establish who is actually selling the service and on what terms. The model comes with its own licence controlling commercial use, and a third-party site wrapping it does not pass that licence on by default. A buyer should establish which model is actually running, who operates the service, and what the terms permit before building anything commercial on the output.

Pros & cons

✓ What we like

  • Reconstructs an object or a body from a single photograph
  • The underlying research model is publicly documented
  • Suitable for props, background assets and rapid concept work
  • Achievable without any scanning equipment

! What to watch out for

  • The specific hosted site recorded here is no longer running
  • Lookalike sites use the name with unclear operators
  • Unseen geometry is inferred, so output is not measurement-accurate

FAQ

Is this tool still available?

Not at the address recorded here, which now returns a hosting error rather than a product.

Can I use it for anything precise?

No. The back of an object is never photographed and is guessed at, so the mesh is an inference rather than a record.

Can I use the output commercially?

The research model's own licence governs that, and a third-party wrapper does not transfer it automatically.

Last reviewed: 2026-09-17

More AI design tools tools

View all →

How we review