Free No account needed

Hi3DGen

Free research demo that turns one photo into a detailed, untextured mesh you can export as STL

Hi3DGen is a free browser tool for free single-color figurines and sculpts from one photo. Free. It runs on Hugging Face ZeroGPU, which gives a daily GPU allowance of 2 minutes without an account, 5 minutes with a free account and 40 minutes with a paid PRO account. Exports STL, OBJ, PLY and GLB. Last checked 26 September 2026.

Open Hi3DGen See alternatives

Opens huggingface.co

At a glance

Price: Free. It runs on Hugging Face ZeroGPU, which gives a daily GPU allowance of 2 minutes without an account, 5 minutes with a free account and 40 minutes with a paid PRO account.
Runs on: Browser
Exports: STL, OBJ, PLY, GLB
Multicolor: No, single color
Account: Not needed
Open source: Yes
Skill level: Beginner: fill in a form
Best for: Free single-color figurines and sculpts from one photo
Selling prints: Allowed The Stable3DGen code and the trellis-normal-v0-1 weights are MIT licensed and the StableNormal estimator it uses is Apache 2.0; the GitHub README says NVIDIA libraries were removed so this version can be used commercially.
How we checked: Opened the generator Space loads without an account: image upload with automatic background removal, Normal Bridge panel, seed and two-stage guidance sliders, and an export menu for obj, glb, ply and stl
Last checked: 26 September 2026

Open generator

Sources: Hi3DGen demo on Hugging Face, Stable3DGen code on GitHub (MIT), Hugging Face ZeroGPU usage quotas (checked 26 September 2026)

Something out of date? Let us know

Made Hi3DGen? Get the listing badge

What is Hi3DGen?

Hi3DGen is a free browser tool that turns a single picture of an object into a detailed 3D mesh you can download as STL. It is a research demo from the GAP Lab at CUHK Shenzhen and ByteDance's Game-AIGC team, hosted as a Hugging Face Space, and its model is fine-tuned from Microsoft's TRELLIS.

You upload an image, the app strips the background, and Generate Shape starts a GPU job. Along the way it estimates a normal map of your picture, shown as the Normal Bridge, and uses it to steer the shape toward sharper surface detail. When the job finishes you get eight preview renders and a 3D Model tab with a format menu for GLB, STL, OBJ or PLY. Advanced Settings holds the seed plus guidance and step sliders for both generation stages, and you can leave it closed on a first try.

The result is shape only with no color, which suits single-color prints anyway. The app itself warns that each mesh is about 40 MB, so plan on scaling and simplifying before you slice. The code and weights are MIT licensed if you would rather run it on your own GPU.

What you can make

Single-color figurines from a photo
Character and creature sculpts
Props and decorative objects
Base meshes to clean up, paint or remix

How to use Hi3DGen

Open the Space and upload one clear image of a single object
Press Generate Shape and wait for the preview renders
Optionally change the seed or guidance under Advanced Settings and run it again
Open the 3D Model tab, pick stl in the File Format menu and click Export Mesh
Scale, repair or simplify the mesh, then slice and print

What you'll need to print it

Filament and parts that suit what you make with Hi3DGen.

Matte PLA

Resin

Siraya Tech Fast ABS-like Resin (affiliate link) Siraya Tech
Siraya Tech Blu Tough Resin (affiliate link) Siraya Tech

Multicolor PLA

Some of these are affiliate links: we may earn a small commission, at no extra cost to you. More about that

Hi3DGen pros

Free in the browser with no sign-up, and the format menu exports STL, OBJ, PLY or GLB directly
Removes the photo background for you, then uses a normal-map step designed to keep folds, edges and small details
MIT-licensed code and weights (about 1.3k GitHub stars) that you can also run on your own NVIDIA GPU

Hi3DGen cons

Shared ZeroGPU time: only 2 minutes a day without an account, queues at busy times, and a research demo can go offline
Geometry only with no color, and from a single photo the unseen back of the object is guessed
Heavy meshes of about 40 MB with no size or base options, so scaling and repair happen in your slicer or a mesh tool