An editorial ranking of ten hosted services and open-weight engines, with the criteria stated up front, two comparison tables, and a plain account of the cleanup every generated mesh still needs.
A text prompt or a single photograph now gets you a usable mesh from any of a dozen services, and a handful of open-weight engines will do the same on your own GPU. They do not produce the same kind of mesh, they do not charge the same way, and only some of them hand you a file that a slicer or a game engine accepts without argument.
This is an editorial ranking, not a benchmark report. It rests on the criteria below, on the documentation each vendor publishes, and on ordinary hands-on use of the hosted tiers. Where a number appears it is a typical industry range or a price the vendor lists, and both move.
Read it as a buying guide: which engine suits hard-surface props, which suits organic characters, which one is worth self-hosting, and what you will still have to fix in Blender before an asset ships.
Five criteria decided the order. Geometry was judged on the same three prompts across every hosted service: a stylised character bust, a hard-surface prop, and a photograph of a household object. Texturing was judged on whether the vendor returns PBR maps or only vertex colours; topology on how much retopology a mesh needs before it can deform. Export breadth came from each vendor's format list, price from the public pricing page. No service received a numeric score, because a score would imply measurements this article does not have.
| Service | Inputs | Vendor-listed exports | Where it fits |
|---|---|---|---|
| Meshy | text, image | GLB, FBX, OBJ, USDZ, STL | props and characters with PBR maps |
| Tripo | text, image, multiview | GLB, FBX, OBJ, USDZ, STL | fast iteration, API batches |
| Rodin (Hyper3D) | text, image, multiview | GLB, FBX, OBJ, USDZ, STL | higher-fidelity assets, Blender plugin |
| Hunyuan3D weights | image, multiview | OBJ, GLB from a local pipeline | self-hosted work, licence per release |
| TRELLIS 2 weights | image, multiview | GLB, OBJ via ComfyUI nodes | local pipelines, batch jobs |
| TripoSR weights | image | OBJ, GLB | blockouts in seconds, no PBR |
| Sloyd | text, parameter presets | FBX, OBJ, GLB | modular low-poly props |
| CSM | image, video | GLB, FBX, OBJ, USDZ | scenes and asset libraries |
| Kaedim | images, sketches | FBX, OBJ, GLB after artist cleanup | studio work needing clean topology |
| Alpha3D | text, image | GLB, FBX | product and AR previews |
Meshy is the safe default for general work. It handles text-to-3D and image-to-3D in the browser, runs PBR texturing in the same flow, and exports a long format list including GLB, FBX, OBJ, USDZ and STL for printing. The vendor also sells an API, which matters once you need fifty props rather than one hero asset. The trade-offs are ordinary: credits vanish while you iterate on textures, and organic characters come back dense and uneven, wanting decimation before a game engine sees them.
Tripo competes on iteration speed and input flexibility: text, a single image, or a set of multiview photographs, with fast previews that make it cheap to try six directions before committing. The export list is comparable, the API is documented for batch work, and the vendor's own open release, TripoSR, gives you a local fallback. The weak spot is fine mechanical detail: thin struts, threads and small holes tend to fuse together, so hard-surface parts usually need rebuilding in a DCC tool.
Rodin, sold by Deemos under the Hyper3D brand, targets higher-fidelity single assets rather than volume. The vendor emphasises cleaner surfaces and stronger geometry, accepts multiview input, and publishes a Blender plugin so a generation can land in a scene without a manual export step. Try it when a client judges a render rather than a prototype; skip it when you need two hundred variations by Friday.
The practical question is what the asset is for. For game props and printing, pick the service whose export list matches your next tool and whose credits last longest, which usually means Meshy or Tripo. For a single marketing render, Rodin's geometry justifies the slower loop. If your pipeline already lives inside Blender, plugin support can settle the decision on its own.
Open weights are a different purchase: you download a model instead of renting generations. Tencent's Hunyuan3D line splits the job into a shape stage and a texture stage and ships weights that run behind ComfyUI nodes or a plain Python script. Microsoft's TRELLIS 2 takes image and multiview input and produces structured geometry that holds up on objects with clean silhouettes. Both drop per-generation fees and move the cost into hardware and upkeep, and licence terms differ between releases, so read the licence file for the version you download.
TripoSR, released by Stability AI and Tripo, is the speed option: a feed-forward image-to-3D model that returns a coarse mesh in seconds on a modern card, with no PBR texture pass. It is a good blockout generator and a poor final-asset generator. Budget hardware honestly: Hunyuan3D 2.x and TRELLIS-class pipelines typically want 12-24 GB of VRAM depending on resolution and batch size, and one asset usually takes tens of seconds rather than minutes.
Self-hosting trades credits for maintenance: the ComfyUI graph or inference script, the Python environment, multi-gigabyte model downloads per version, and the debugging when an update breaks a node. In exchange, output is reproducible, nothing is metered, and nothing leaves your machine. If nobody on the team enjoys that work, hosted services are cheaper than they look, and managed alternatives exist for the same reason: Sloyd generates modular low-poly props from parameters, CSM targets scene and asset libraries, Kaedim puts human artists on retopology.
| Service | Free tier | Typical paid entry | What burns credits |
|---|---|---|---|
| Meshy | limited monthly credits | individual plans around 16-30 USD per month | generations, texture passes, private assets |
| Tripo | starter credits | similar band, annual lower | generations, high-res meshes, API calls |
| Rodin (Hyper3D) | daily free credits | tiered subscriptions | generations, fidelity, downloads |
| CSM | trial credits | mid-range subscription | image-to-3D runs, exports |
| Sloyd | limited trial | per-seat subscription | generations, API calls |
| Kaedim | quote only | priced per delivered asset | artist hours, revisions |
| Self-hosted engines | weights free to download | your GPU time and power | VRAM hours, storage, maintenance |
One category the tables do not cover is the comparison front end. Open3DAI runs one local procedural generation engine and can connect third-party engines through API adapters you configure with your own keys, so the same prompt or image can go to several engines and be compared side by side before export. It covers text-to-3D, image-to-3D, multiview input and AI texturing, and exports GLB, glTF, OBJ, STL, PLY and 3MF, alongside 48 format converters, 12 browser tools and a REST API. It does not train or host the third-party models, and it does not do rigging, animation, retopology, UV unwrapping, photogrammetry or CAD-grade parametric solids; those jobs still belong to a desktop tool, and a generated mesh should be inspected and repaired before printing.
Engine availability, credit pricing and third-party terms change; verify them on the vendor page before committing a budget.
Any of the hosted services can produce the base shape, so the deciding factor is the export and the cleanup, not the generator. Prefer a service that exports STL or 3MF without forcing a conversion, then check the mesh in Blender or Meshmixer for wall thickness, non-manifold edges and a hollow interior before slicing. None of these tools guarantees a watertight, print-ready result, and generated meshes often need holes drilled, thin walls thickened and the model rescaled to real-world millimetres.
Most hosted services have a free tier capped by monthly credits, which is enough to evaluate output quality but not to finish a project. Paid individual plans commonly sit somewhere in the 16-30 USD per month band, with annual billing cheaper and studio seats priced higher. Open-weight engines such as Hunyuan3D, TRELLIS 2 and TripoSR carry no per-generation fee, but you pay for the GPU, the electricity and the time spent maintaining the environment. Treat every figure as a list price that can change.
It depends on the plan and on the licence of the specific engine or version. Hosted vendors usually grant commercial rights on paid tiers and restrict or watermark them on free tiers, while open-weight releases differ from version to version, so read the licence file that ships with the weights you download. If a project has a legal review step, record the engine name, version and plan at the time of generation, because terms are updated more often than most pipelines track.
No. They return a static mesh, sometimes with PBR textures and sometimes with vertex colours only. Rigging, skinning, keyframes and animation are done afterwards in Blender, Maya or another DCC tool, and the generated topology usually needs retopology first because a dense triangle mesh deforms badly. Plan the character pipeline as generation, retopology, UV work, rigging, animation, and budget most of the hours for the middle three steps.
A CUDA GPU with roughly 12-24 GB of VRAM is the comfortable range for the full shape-plus-texture pipelines at useful resolution, and a single asset typically takes tens of seconds rather than minutes at that tier. Cards with 8 GB can run reduced settings, often by lowering resolution or running the shape and texture stages separately, with more out-of-memory failures along the way. Storage matters too: model weights and caches run to several gigabytes per version.
It varies by vendor, engine version and quality setting, but a common pattern is a handful of credits for a draft generation and noticeably more for a high-fidelity or retextured result, with private or commercial-use modes costing extra. Retexturing an existing mesh is usually cheaper than regenerating it. Check the current plan page before budgeting, because credit costs and engine availability change between releases, and free daily credits on some services are time-limited rather than accumulative.
Send the same prompt or image to several engines, compare the meshes side by side, then export GLB, glTF, OBJ, STL, PLY or 3MF. Creating an account grants free credits and needs no card.
The short answer is that Meshy covers the most pipeline steps and Tripo returns the geometry most people keep. The longer answer is that both picks rest on three criteria you can test yourself in one afternoon.
Rodin, Meshy, Tripo, Sloyd, Hunyuan3D 2.x and TRELLIS 2 judged on the four things a game team cares about: edge flow, retopology cost, engine handoff and texture output.
3D AI Studio sells one credit balance that covers image-to-3D, text-to-3D, texturing, rigging and a shelf of mesh utilities, and it routes part of that work to Meshy, Tripo and Rodin. This review works through the published credit schedule, the list prices behind it, and the pipelines where a subscription is the wrong purchase.