UVgami unwraps UVs automatically and for free

Developer Daniel Boxer has released UVgami — a free automatic UV unwrapping tool that runs right in the browser and also exists as a Blender add-on and a desktop application. The code is open and published on GitHub. The user drags a model into the window, clicks Unwrap and downloads it with a finished UV map; the computation runs on the user's machine, and files are never sent to a server. No account is needed, there is no limit on the number of models, and the result may be used in commercial projects.

The web version accepts OBJ, GLB, GLTF and FBX. OBJ comes back in the same format, everything else as GLB. Required browsers are Chrome or Edge 116+, Firefox 124+ or Safari 17.4+. The add-on installs into Blender 4.3 and newer by dragging in the zip archive, after which the engine is downloaded with the Download Engine button in the N-panel; auto-updates are enabled by adding a remote extension repository in Blender's preferences.

Three engines — three different trade-offs

Two engines are available in the browser, OptCuts and xatlas; the add-on adds PartUV to them. The choice comes down to what you pay for the result with: time, the number of seams or the number of islands.

• OptCuts — the default engine, computes on the CPU. Delivers the best quality: seams are placed where stretching is minimal. The price is speed — on dense meshes the process drags on.
• xatlas — also CPU, but noticeably faster. You end up with more seams and islands, but that is enough for baking lightmaps and quick texture painting.
• PartUV — computes on the GPU, requires CUDA and is available only on Windows and Linux. Produces fewer islands and can be faster on dense meshes.

The engines are built on published algorithms: OptCuts is work by Minchen Li and co-authors, xatlas is a library by Jonathan Young.

What this gives you in a real pipeline

Local computation removes the issue that usually gets in the way of using cloud services in studio work: the model does not leave the workstation. For a freelancer that is convenience; for an outsourcing studio it is the condition under which the tool can be brought into the pipeline at all.

The add-on has a wider set of settings than the browser. Priority switches between a balanced mode, minimum stretching (the slowest) and minimum seams (the fastest). Hard Surface mode places seams along sharp edges and bevels — appropriate for mechanical objects, not for organics. With weights the artist paints in areas where seams are unacceptable or where stretching needs to be reduced. Proxy mode unwraps a simplified copy and transfers the seams onto the original geometry — a way to actually get a result on a dense mesh. Stack Similar unwraps one of the repeating elements and copies the UVs onto the rest, overlapping the islands. A timeout is set separately: when it triggers, OptCuts saves the intermediate result. There is also a panel in the UV editor for refining a finished map — recomputing and relaxing individual islands or selected polygons.

Where the catch is

The developer lists the limitations outright. Non-manifold geometry and zero-area polygons break the unwrap — the mesh will have to be cleaned up beforehand in your own 3D package. Seam avoidance mode slows the computation, which is why it is recommended to combine it with proxy, while stretch reduction does not work with proxy at all. PartUV has two segmentation options: the neural one weighs around 5 GB and produces fewer seams, but NVIDIA's license forbids using it in commercial work; the geometric one takes about 200 MB and runs fast. The browser version is slower than the add-on and the desktop application, which run the engine natively. The web version, the add-on and the desktop application are available at UVgami.

Why tools like this are appearing right now

UV unwrapping remained for years one of the most manual operations in the 3D pipeline: automatic parameterization algorithms had existed for a long time, but they lived in academic repositories, had to be built from source and barely reached modelers. Now the situation is changing from two directions. The browser has learned to run a heavy native solver without installation and without sending data to a server, and the stream of new work on geometry processing provides ready-made methods that only need to be wrapped in an interface.

Against this backdrop, automatic unwrapping is probably ceasing to be a standalone paid feature and becoming a baseline capability: paid solutions will start competing not on the fact of automation itself, but on speed, predictability of seams and island packing. A separate signal for studios is the licensing caveat around neural segmentation: as trained models get embedded into graphics tools, the legal status of a particular mode becomes as much of a production parameter as computation speed, and it will have to be checked before the asset goes into a project.

Which unwrapping engine would you pick first?

• OptCuts
• xatlas