Trellis 2 is an image-to-3D service built on TRELLIS.2, a 3D generation model released by Microsoft as a research project. All you need is a single image: from it the service produces a textured mesh you can download as GLB. Everything runs in the browser, so no local GPU is required, and the output is intended for import into Blender, Unity, or Unreal Engine. The underlying TRELLIS.2 model is a 4B-parameter flow-matching transformer that supports resolutions up to 1536³ voxels. The service is operated by trellis2.com, and the catalog lists its release date as December 24, 2025.
Key Features
- A 3D mesh from one image: You do not need multi-angle photos or a 3D scan ─ a single image is enough to produce a mesh. That suits early-stage asset work such as game prototypes or product mockups, where having the shape is what matters
- Topology preserved via O-Voxel: The sparse voxel representation used by TRELLIS.2 (O-Voxel) handles open surfaces, non-manifold geometry, and enclosed interior structures without breaking them. It helps with thin shapes such as cloth or leaves, and with shapes that have hollow interiors
- PBR material support: Output includes base color, roughness, metallic, and opacity. Metallic and semi-transparent surfaces carry over into DCC tools more faithfully, which cuts down on repainting textures
- In-browser GLB preview: You can inspect the geometry and materials of a generated GLB in the browser without uploading it, checking for breakage before importing into Blender or elsewhere
- Adjustable generation parameters: Seed, decimation (polygon reduction), and texture size can be specified, so you can produce either lightweight or high-detail meshes depending on the use case
- A free demo alongside paid plans: The free demo runs on Hugging Face infrastructure, so there is a queue. Paid plans generate without waiting in line and store your results in the cloud
Pricing
| Plan | Monthly price | What you get |
|---|---|---|
| Free demo | $0 | Try it on Hugging Face; queue applies |
| Basic | $15.9 | 700 credits/month (~23 models), all resolutions, standard queue priority, GLB download and cloud storage, email support |
| Standard | $29.9 | 1,500 credits/month (~50 models), all resolutions, high queue priority, priority support |
| Premium | $49.9 | 3,000 credits/month (~100 models), all resolutions, highest queue priority, dedicated support, early access to new features |
Annual billing comes with a 25% discount. All paid plans state that you own the content you generate.
Pricing is as of August 2026. Check the official site for the latest details.
Pros & Cons
✅ Pros
- You can try high-resolution 3D generation from the browser alone, with no local GPU
- The underlying TRELLIS.2 is a Microsoft research project with published code, so the model internals can be inspected
- Output ships with PBR materials, so assets look reasonable right after import into a DCC tool
- Strong on open surfaces and non-manifold geometry ─ thin shapes that trip up typical image-to-3D tools hold together better
- A free demo lets you check quality on your own images before paying
⚠️ Cons
- The official site itself notes that mesh artifacts can occur; there is no guarantee the output is production-ready as-is
- The credit system means heavy trial and error burns through your allowance faster than expected, and costs are hard to predict when one image is not enough
- Output centers on GLB, so pipelines that require FBX or USD need a conversion step
- The free demo runs on shared infrastructure, so waits can be long at busy times
- There is no rigging or animation feature, so animating a character requires a separate tool
Comparison with Similar Services
| Criteria | Trellis 2 | Meshy | Tripo | Hunyuan3D |
|---|---|---|---|---|
| Base model | Microsoft TRELLIS.2 (code published) | In-house | In-house | Tencent (model published) |
| Main input | Single image | Image, text | Image, text | Image, text |
| PBR materials | Yes | Yes | Yes | Yes |
| Published max resolution | 1536³ voxels | Not published | Not published | Not published |
| Rigging / animation | No | Yes | Yes | Limited |
| Free trial | Free demo | Free tier | Free tier | Free environments available |
Meshy and Tripo are all-round tools covering the whole 3D pipeline (text-to-3D generation, retopology, rigging), whereas Trellis 2 concentrates on the quality of shape generation from a single image. If what you want is character animation, the former; if you want to raise the shape of a prop or product quickly, Trellis 2 is a candidate.
Who Is It For
- Solo developers who need a large number of placeholder assets for a game or app prototype
- E-commerce and manufacturing staff who want to turn product photos into 3D mockups for AR display or presentation material
- People with little 3D modeling experience who want to turn a hand-drawn illustration or photo into a solid model
- Anyone frustrated by breakage in existing image-to-3D tools on thin shapes or models with interior structure
- Anyone without a high-end local GPU who wants to run 3D generation in the cloud
Summary
Trellis 2 packages Microsoft’s TRELLIS.2 research model into a usable service, and its strength is pulling a PBR-textured mesh out of a single image. The sparse voxel representation holds up well on open surfaces and non-manifold geometry, where typical image-to-3D tools struggle. That said, the service itself acknowledges that mesh artifacts can occur, so final quality still assumes human cleanup. A practical path is to run your own images through the free demo first, then consider a paid plan once wait times or generation counts become the limiting factor.