
Oct 8, 2026
Best Free Online 3D Viewers in 2026: GLB, STL, OBJ, FBX
Six free ways to open a 3D file without installing much, compared on formats, upload, sign-in and limits, including the TRELLIS.2 browser viewer.
TRELLIS.2 and Tripo are not the same kind of product, so a single winner table would mislead. TRELLIS.2 is Microsoft's open-weight image-to-3D model; trellis-2.com is an independent hosted workflow built around it. Tripo is a commercial, API-first 3D generation platform with its own hosted studio and a documented developer API.
Choose trellis-2.com when you want to upload one image in a browser, pay for a prepaid credit pack, and download a textured GLB without writing code. Choose Tripo when you need an API, asynchronous task handling, rigging or animation retargeting, non-GLB export formats, or higher parallel throughput. If you need open weights you can run yourself, use Microsoft's TRELLIS.2 repository directly; neither hosted service gives you the model weights.
This page compares the two on facts a buyer can check: published prices, output formats, documented limits, licence and workflow. It does not invent quality scores, benchmark results or user counts. Every price is dated 2026-10-21 and linked to the vendor page it came from.
Microsoft Research publishes TRELLIS.2 as an open-source image-to-3D project. The upstream repository describes a 4-billion-parameter model built on an O-Voxel sparse voxel representation, with PBR materials and an MIT licence. The same repository states that the code needs an NVIDIA GPU with at least 24 GB of memory, is currently tested only on Linux, and has been verified on A100 and H100 GPUs. It also publishes model-runtime figures of about 3 seconds at 512³, 17 seconds at 1024³ and 60 seconds at 1536³ on an H100. Those are upstream measurements on named hardware, not a promise for a shared queue or a consumer GPU.
trellis-2.com is an independent third-party service around that model. It is not Microsoft and not the official project site. Its documented workflow is one image per generation, returned as a GLB with embedded PBR textures, at 512³, 1024³ or 1536³. The credit cost is 150, 300 or 500 credits respectively, shown before you generate. A standard browser run often finishes in roughly one to three minutes, depending on provider load and input complexity. Failed or canceled jobs return their reserved credits, and the service states a maximum of two active jobs per account.
The public pages reviewed for this comparison describe a browser workflow; they do not document a developer API. If you need to automate trellis-2.com, confirm the current integration options with its support before planning a pipeline.
Tripo is a commercial 3D generation platform with a public developer API. Its documentation describes a base URL of https://openapi.tripo3d.ai/v3, asynchronous task creation followed by polling or webhooks, and endpoints for text-to-3D, image-to-3D, multiview-to-3D, texturing, segmentation, retopology, format conversion, rig checking, auto-rigging and animation retargeting. The quick-start guide says a typical generation takes 10–120 seconds, recommends polling every 2 seconds and asks callers not to exceed 1 request per second.
The API is credit-based. On the pricing page checked 2026-10-21, 1 credit = $0.01 USD, credits can be bought from $1, and payment runs through Stripe. That page is the source of every Tripo price in this article. Tripo also operates a consumer studio, but this comparison uses the published API prices and limits because those are the numbers a developer can verify.
| Question | trellis-2.com (hosted TRELLIS.2) | Tripo API |
|---|---|---|
| Delivery | Browser workflow, hosted by an independent service | Hosted API plus a separate studio product |
| What you upload | One image per generation; JPEG, PNG, WebP, GIF or AVIF | Image or image URL for image-to-3D; text prompts and multiview inputs are also documented |
| Default output | GLB with embedded PBR textures | GLB by default |
| Other formats | Free browser converter for glTF, OBJ, STL, PLY and VOX | Convert endpoint lists GLTF, FBX, USDZ, OBJ, STL and 3MF |
| Resolution choices | 512³, 1024³, 1536³ | Mesh and texture controls such as face limit, quad mesh, texture size and texture format; model versions are selectable |
| Published generation cost | 150 / 300 / 500 credits at 512³ / 1024³ / 1536³ | 20 credits without texture or 30 with standard texture for image-to-3D; add-ons cost extra |
| Rigging and animation | Not advertised in the documented workflow | Rig check (free), auto-rig (25 credits), retarget (10 credits per animation) |
| API and automation | No public API documented on the reviewed pages | Documented REST API, asynchronous tasks, polling and webhooks |
| Concurrency | Maximum two active jobs per account | Per-category account limits: 10 H-series, 5 P-series, 1 image, 10 animation, 5 model processing, 10 mesh |
| Self-hosting | No; the site is a hosted service | No; Tripo is a hosted commercial platform |
| Open weights | Yes, from Microsoft's separate TRELLIS.2 repository | No |
| Best fit | A no-code, one-image-to-GLB workflow with prepaid packs | Developers and pipelines that need an API, rigging, format control or throughput |
The headline price is not the cost of a usable asset. The useful number is cost per accepted model after failures, retries and any extra processing. The tables below use only the prices published by each vendor on 2026-10-21.
| Pack | Price | Credits included | 512³ drafts covered | Effective price per 512³ draft |
|---|---|---|---|---|
| Mini | $4.99 | 150 | 1 | $4.99 |
| Starter | $12.90 | 300 | 2 | $6.45 |
| Pro | $38 | 3,000 | 20 | $1.90 |
| Studio | $129 | 22,500 | 150 | $0.86 |
The same pricing page also shows higher list prices of $18, $55 and $189 for the three larger packs next to the current amounts. One-time credits never expire, failed jobs return their reserved credits, and unused credits can be refunded within three days under the stated policy. Higher resolutions consume more credits: a 1024³ job uses 300 and a 1536³ job uses 500, so the “drafts covered” column above is specifically for 512³.
| Operation | Credits | Cost at 1 credit = $0.01 |
|---|---|---|
| Image to 3D, no texture | 20 | $0.20 |
| Image to 3D, standard texture | 30 | $0.30 |
| Text to 3D, no texture / standard texture | 10 / 20 | $0.10 / $0.20 |
| Multiview to 3D, no texture / standard texture | 20 / 30 | $0.20 / $0.30 |
| HD texture add-on | +10 | +$0.10 |
| 8K ultra texture add-on | +20 | +$0.20 |
| HD geometry quality add-on | +20 | +$0.20 |
| Quad mesh add-on | +5 | +$0.05 |
| Smart low-poly add-on | +10 | +$0.10 |
| Auto-rig | 25 | $0.25 |
| Animation retarget | 10 per animation | $0.10 |
| Advanced format-conversion options | 10 | $0.10 |
| Texture an existing model | 50 | $0.50 |
Two worked examples show why scope matters more than the sticker:
Those numbers compare generation credits, not quality or cleanup time. They also do not include rigging, animation or retopology, which Tripo sells as separate operations and trellis-2.com does not document. A fair comparison adds the cost of the missing steps, including the DCC tool or service you would use to finish the asset.
trellis-2.com deliberately narrows the choices: three resolution levels, one output format, and a free browser converter for glTF, OBJ, STL, PLY and VOX. The texture is part of the GLB, so the handoff to Blender, Unity, Unreal or a print slicer is one download. The trade-off is control. If you need a specific triangle budget, a quad mesh, a particular texture size or a USDZ file directly from the API, that is not the documented workflow here.
Tripo exposes more of those controls. Its convert endpoint lists a face_limit, a quad flag (which forces FBX output), texture size up to the published default of 4096 pixels, texture formats including JPEG, PNG and WebP, pivot and scale options, and FBX presets for Blender, 3ds Max and Mixamo. Its rigging endpoint accepts GLB, FBX, OBJ or STL input up to 150 MB and returns GLB or FBX. Those are documented parameters, not quality guarantees, but they decide whether a generated asset can drop into an existing pipeline without a second tool.
For 3D printing, the shortest path differs. On trellis-2.com you generate a GLB and convert it to STL in the free browser tool, then run your own mesh checks. On Tripo you can request STL or 3MF from the convert endpoint, paying the conversion operation. Either way, check wall thickness, manifold edges and scale before slicing; neither vendor promises a print-ready solid.
This is the clearest dividing line. Tripo is built to be called by software: create a task, receive a task ID, poll or wait for a webhook, download the result. Its rate-limit page says limits apply per API key, generation endpoints are stricter than query endpoints, and a throttled request returns HTTP 429 with error code 1007. Concurrency is separate and applies per account and per task category. The published defaults are 10 H-series 3D generation tasks, 5 P-series 3D generation tasks, 1 image-generation task, 10 animation tasks, 5 model-processing tasks and 10 mesh-operation tasks. When a category is full, new tasks in that category return 429 with code 2000 and a Retry-After header; other categories keep working.
trellis-2.com is a manual browser workflow with a stated maximum of two active jobs per account. That is enough for a designer working through a folder of product photos; it is not a batch API. If your plan involves a queue, retries, per-tenant isolation or server-side integration, Tripo's documented model is the closer fit, and you should budget for its rate limits and concurrency rather than assuming unlimited parallel jobs.
The TRELLIS.2 model is published under the MIT licence in Microsoft's repository. That is the licence to read if you intend to self-host, fine-tune or redistribute the model. Self-hosting is not free operationally: the upstream notes ask for at least 24 GB of NVIDIA memory, state Linux-only testing, and name A100 and H100 hardware for the published timings. Add electricity, storage, setup time, monitoring and upgrades before comparing it with a hosted bill.
trellis-2.com is an independent hosted service, not Microsoft, and its own terms govern the service. Tripo is a commercial hosted platform; its terms govern output rights, data handling and account use. This page does not reproduce either contract. If commercial redistribution matters, read the vendor terms and the MIT licence text directly before you rely on a summary.
Neither vendor publishes a measured success rate on a defined image cohort, and this page will not invent one. A vendor demo is a curated input; your product photos, character art or scanned objects are not. The only defensible comparison is your own test set.
A reproducible protocol:
Use this formula rather than the per-generation price:
cost per accepted asset =
(generation spend + variable infrastructure + operator time) / accepted assetsA cheaper generation can lose to a more expensive one if it fails more often or needs more cleanup. Equally, a fixed pack can be simpler than a usage-based API even when the API's unit price is lower, because the pack removes billing and integration work.
trellis-2.com is a good fit when:
Tripo is a good fit when:
Neither hosted service is the right answer when:
If the browser path fits, start with one 512³ job, inspect the GLB from several angles, and only then spend credits on a higher resolution:
Review current TRELLIS.2 credit packs
If you are still shortlisting, the broader five-way comparison covers Meshy, Hunyuan3D, Tripo and Rodin as well:
All pages were opened on 2026-10-21. Prices and limits on vendor pages can change; recheck before purchase.
It depends on volume and scope. On the pages checked 2026-10-21, trellis-2.com sold 150 credits for $4.99 (one 512³ draft), 3,000 credits for $38 (20 drafts) and 22,500 credits for $129 (150 drafts). Tripo's API priced image-to-3D at 20 credits without texture or 30 credits with standard texture, at 1 credit = $0.01, so one textured draft cost $0.30 before add-ons. At low volume the fixed pack is simple; at API volume Tripo's per-task price is much lower. The outputs and included steps differ, so compare cost per accepted asset, not the headline number.
The public pages reviewed for this comparison (llms.txt, /pricing and /image-to-3d) document a browser workflow and do not describe a developer API. Tripo publishes an API with asynchronous tasks, polling and webhooks. If you need automation, confirm the current trellis-2.com integration options with its support before you plan a pipeline.
Tripo's API documents rig-check, auto-rig and animation retargeting as paid operations, with output in GLB or FBX. The trellis-2.com workflow reviewed here returns a textured GLB draft and does not advertise rigging or animation. For a rigged character, Tripo's API or a separate DCC step is the documented path.
trellis-2.com returns GLB with embedded PBR textures; its free browser converter can then produce glTF, OBJ, STL, PLY or VOX. Tripo returns GLB by default and its convert endpoint lists GLTF, FBX, USDZ, OBJ, STL and 3MF, with texture size, texture format, face limit, quad mesh and pivot controls.
trellis-2.com states a maximum of two active jobs per account, and failed jobs return their reserved credits. Tripo applies per-category concurrency at account level: 10 for H-series 3D generation, 5 for P-series 3D generation, 1 for image generation, 10 for animation, 5 for model processing and 10 for mesh operations, with 429 responses and a Retry-After header when a pool is full.
Yes. Microsoft publishes the TRELLIS.2 repository under the MIT licence. The upstream notes require an NVIDIA GPU with at least 24 GB of memory and state that the code is currently tested only on Linux, with verification on A100 and H100 GPUs. Self-hosting also carries hardware, electricity, setup and operator cost; trellis-2.com is the hosted alternative, not the model itself.
Start with the workflow you will actually run. If you need a browser upload and a GLB today, test trellis-2.com at 512³. If you need an API, batch throughput, rigging or non-GLB exports, test Tripo with a small credit purchase. In both cases use 20–50 images from your real workload, keep every failure, and record accepted outputs, spend and operator time before choosing.