Meshy 7: the 3D asset platform that finishes what comes after generation, from topology and PBR to rigging, printability, and MCP
meshy
Meshy headlines Meshy 7 (2026) and its "true-to-reference" claim, meaning alignment between the generated model and your input image, with image-to-3D stated at under a minute. The real differentiator is not generation but finishing the engineering constraints that follow it: Smart Topology returns clean topology with a controllable poly count in about ten seconds, one-click PBR delivers the full albedo, normal, metallic, and roughness set at HD 4K with no shader setup, auto-rigging covers humanoid and quadruped with 600+ motion presets, and on the print side meshes come out watertight with auto-repair before export. It exports GLB, FBX, OBJ, STL, USDZ, and 3MF straight into OrcaSlicer and Bambu Studio. The agent surface is complete too: a REST endpoint per generation capability, an MCP server that connects into Claude or Cursor, and a CLI that turns one command into a finished model. Stated scale is 100M+ models and 12M+ users, with 100 free credits a month. Boundaries: no public quantitative benchmark (the quality stars are self-assigned), credit pricing per tier undisclosed, and "under a minute" excludes post-processing.
- CONFIDENCE
- Vendor Claim
- Official model card or keynote only, no independent re-test
- KEY METRIC
- 官方宣称:单图 → 3D 模型
- Vendor Claim · 2026
- MATURITY
- Product
- research → demo → product → production
Our take<p>By 2026 the differentiator in 3D generation is no longer whether a tool can generate, but <strong>whether the output is usable as-is</strong>. Meshy covers that stretch more completely than the rest of the field: Smart Topology returns clean topology with a poly count you control in about ten seconds, one-click PBR gives the full set with no shader setup, auto-rigging handles humanoid and quadruped with 600+ motion presets, and on the printing side meshes come out watertight with an auto-repair pass before export. Those are the actual reasons AI meshes fail to make it into a project, and it answers each one. That is why it sits at the top of our world3d ladder.</p> <p>The second point is its stance toward agents. It is not only a human tool: every generation capability is a REST endpoint, there is an MCP server that connects into Claude, Cursor, or any AI agent, and a CLI that turns one command into a finished model. <strong>Whether a 3D tool can enter an automated pipeline depends on having MCP and API access, not on having an attractive web UI</strong>, and Meshy has both.</p> <p>Three honest caveats. First, the headline claim of being "true-to-reference" targets the right problem, since the most common reason for rework in 3D generation is not ugliness but failing to match the concept art, yet <strong>there is no public quantitative benchmark</strong>. The quality stars in its own docs are self-assigned, so the card is graded C (vendor claim). Second, "under a minute" covers generation only and excludes topology, texturing, and rigging; the end-to-end total is not published and we have not measured it. Third, the relationship between credit cost and model tier is undisclosed, so the free 100 credits a month are enough to try it but not enough to derive a real unit cost. The claim of being the only 3D AI trained to print is a marketing assertion: the printability check and auto-repair are real, but "only" cannot be verified.</p>
What it actually delivers
The official documentation defines Meshy as "an AI-powered 3D modeling platform that transforms text descriptions, 2D images, and conversational prompts into production-ready 3D assets," and it stresses coverage of the full chain from generation through post-processing to export rather than a single endpoint that returns a mesh. The web workspace is organized into six modules: Image, 3D Model, 3D Printing, Animate, Scene, and Video. There are four entry points sharing one set of generation models and one account: the Web App, the API (REST, with an endpoint for every generation capability), plugins (Unity, Blender, Maya, and more), and a dedicated 3D Printing path. On top of that, 3D Agent turns "chat from idea, photo, or sketch to a 3D model" into a conversational flow, and Meshy App lets you take a photo on your phone and get a model from it.
The value here is not "can generate 3D," which is table stakes in 2026, but that it wires up the engineering constraints that come after generation: topology, PBR materials, rigging, printability, engine compatibility, and an API/MCP/CLI surface for agents. Those are the real gates between a 3D asset and a project that can actually use it.
Model tiers, as the vendor states them, with release years
| Model | Release | Positioning | Stated speed |
|---|---|---|---|
| Meshy 7 | 2026 | Most detail and highest precision; the headline is "true-to-reference," meaning alignment with your input image | Under a minute |
| Meshy 6 | 2025 | High quality with stable structure | About 1 minute |
| Smart Topology | 2026 | Topology optimized for game development and real-time rendering | About 10 seconds |
The docs state that Image to 3D runs on Meshy 7, "focused on alignment between your input image and the generated result." Reference alignment is the headline metric for this generation, which lands squarely on the most painful part of production: the usual problem with generated 3D is not that it looks bad but that it does not look like the concept art you supplied, which sends the art team back to rework.
Wiring up what comes after generation, item by item
| Stage | Stated capability | Why it matters |
|---|---|---|
| Topology | Smart Topology in about 10 seconds; Low Poly Mode gives clean topology with a poly count you control, light enough to drop into a project without a performance hit | Uncontrolled face counts and non-deformable topology are the common causes of death for AI meshes |
| Materials | One-click full PBR set (albedo, normal, metallic, roughness) with no shader setup; an HD Texture switch for 4K detail | Removes a manual material step |
| Rigging and animation | Auto-rigging for both humanoid and quadruped characters; 600+ motion presets covering walk cycles, idles, jumps, and attacks, with skeleton and weights ready to animate | The largest manual cost between "has a mesh" and "can appear on screen" |
| Printability | Stated as "The Only 3D AI Trained to Print": models generate watertight and manifold, ready to slice; a printability check still scans every export and auto-repairs edge cases. The example readout shows watertight Yes, volume 702.19 cm³, 0 holes, 0 non-manifold edges | Most AI models need manual hole-filling before they will slice |
| Multi-color printing | Multi-color printing maps each texture to the right filament with clean color boundaries; when a model is too complex, Auto Split breaks it into watertight parts pre-arranged on the build plate | Splitting and plate layout used to be manual work |
| Direct to slicer | Export STL, OBJ, or 3MF and send straight to OrcaSlicer, Bambu Studio, and others; plugins cover Bambu Studio, Creality Print, OrcaSlicer, Ultimaker Cura, Snapmaker, Flash Studio, Elegoo Slicer, and LycheeSlicer; publishing to MakerWorld, Printables, and Thingiverse; FDM and resin both supported | No manual file moving after generation |
| Engines and DCC | Compatible with Unity, Unreal Engine, Godot, Blender, and Maya, plus plugins for Unreal Engine, Roblox Studio, and 3ds Max | Covers both real-time and offline pipelines |
| For agents | REST API covering Text to 3D, Image to 3D, texturing, remeshing, and animation; an API Playground for no-code testing; an MCP server that connects into Claude, Cursor, and any AI agent; and a CLI where one command turns a prompt or image into a finished model | This is what lets it enter an automated pipeline rather than only a human workstation |
Export formats and how each handles textures deserve their own table, because they determine whether you need another step:
| Format | Use case | Textures |
|---|---|---|
| GLB | Web, AR, universal exchange | Embedded |
| FBX | Unity, Unreal, DCC tools | Separate files |
| OBJ | Legacy workflows, DCC tools | MTL plus textures |
| STL | 3D printing | Geometry only |
| USDZ | Apple AR Quick Look | Embedded |
| 3MF | 3D printing with color | Embedded |
Scale and commercial surface
The homepage states three figures: 100M+ generated models, 12M+ users, and 10M+ monthly visits, with G2 and Trustpilot ratings of 4.8 each. The free plan provides 100 credits a month with no credit card required. The operating entity is MESHY LLC, and the site runs Affiliate, Contributor, Fellowship, and Education programs alongside free tools such as a 3D File Converter, Online 3D Viewer, 3D Text Generator, 3D File Compressor, and STL Repair that serve as traffic entry points.
It also publishes three comparison pages: Meshy vs. Tripo, Meshy vs. Trellis 2, and Meshy vs. Hunyuan3D. These are written by the vendor, so read them with that in mind. They do at least show that Meshy positions itself against Tripo (a closed product) and TRELLIS and Hunyuan3D (open-weight research) on the same track, and our 3D ladder deliberately carries all three.
Boundaries and failure modes
- The speed figure is the vendor's and excludes post-processing. "Under a minute" refers to generation. Topology optimization is a separate roughly ten seconds, and PBR, rigging, and the pre-slice check each take more time. The true total from one image to a project-ready asset is not published and we have not measured it.
- The quality stars are self-assigned. Meshy 7, Meshy 6, and Smart Topology are all marked five stars in its own table, which is not a comparable quantitative measure. There are no public 3D reconstruction benchmark readings such as Chamfer distance or F-score, so the card is graded C (vendor claim).
- "The only 3D AI trained to print" is a marketing assertion. The printability check and auto-repair are real features, but "only" cannot be verified, and competitors are closing the same gap.
- Credit pricing per model tier is not published. The free plan gives 100 credits a month, but the docs contain no table showing what Meshy 7 versus Meshy 6 costs, or what HD 4K texturing and auto-rigging deduct. Estimating cost requires trial.
- Auto-rigging has a defined scope. The 600+ presets target humanoid and quadruped characters. Mechanical, multi-limbed, and non-biological structures fall outside it and still need manual rigging.
- STL carries geometry only. Multi-color printing requires 3MF. Choosing the wrong format silently discards color, and this is the easiest step to get wrong in the chain.
- The Scene and Video modules are new. Of the six modules, Scene and Video lack documentation at the depth given to 3D Model, so we do not know their capability level and we are not vouching for them.
Our reproduction plan
Four runs would move this from grade C to grade A. First, feed one concept image to Meshy 7, Tripo, and Hunyuan3D 2.1 and compare reference alignment, topology quality, and poly-count control, turning "true-to-reference" into a readable number. Second, time it end to end: real wall-clock time and credit consumption from image upload to a finished GLB including PBR, Smart Topology, and auto-rigging, to derive a unit cost. Third, test the printing chain by sending the exported 3MF straight into OrcaSlicer and checking whether the stated watertight, hole, and non-manifold-edge readouts and the auto-repair behave as claimed. Fourth, run its MCP server from Claude or Cursor to have an agent generate an asset and write it to disk unattended, verifying that agent access really does reach an automated pipeline. Until those run, read the number on the card as what the vendor says.