PolyMedium vs Hunyuan3D: Hosted vs Open-Source AI 3D
Updated August 30, 2026 · Published August 29, 2026
Hunyuan3D is Tencent's open-source 3D generation family: genuinely strong models whose weights are free, if you bring the GPU and the setup time. PolyMedium runs in the browser for $0.25 a model. This is the classic hosted-versus-self-hosted decision, so here is what actually trades off.
Feature comparison
| Feature | PolyMedium | Hunyuan3D |
|---|---|---|
| Setup | Browser, none | Local GPU, model download, Python environment |
| PBR textures | Yes, baked | Yes |
| Reconstruction style | Scan-faithful on objects and drawings | Prior-heavy: excellent on organic characters, weaker on flat drawings |
| Game-engine output | GLB with mobile/game/web presets | Generic mesh, manual cleanup |
| Cost | $0.25 per object, characters $0.50 to $3.00, no subscription | Free weights; you supply GPU hardware or cloud GPU time |
| Control | Hosted pipeline, no knobs to maintain | Full control: fine-tune, modify, run offline |
PolyMedium pros
- • Zero setup: from photo to exported GLB in the browser in minutes
- • Subject detection routes each upload to a pipeline benchmarked for it, including a drawing pipeline that prior-heavy engines handle poorly
- • Game-preset exports arrive runtime-ready instead of needing cleanup
Trade-offs
- • Hosted and closed: you cannot fine-tune the pipeline, inspect the weights, or run it offline the way open weights allow
- • Per-model fees: at very high volume on a GPU you already own, self-hosting Hunyuan3D costs less per mesh
- • Your images are processed by a hosted service, which some studios' policies rule out
Verdict
PolyMedium if you want to ship assets today with no infrastructure. Hunyuan3D if you have the GPU, want full control of the pipeline, and your subjects are the organic characters it is strongest on.
The structural difference behind every row above: most tools on this page sell you one engine. PolyMedium tested 41 engines and routes each upload to whichever wins on that subject — including, at some tiers, engines from the companies it is compared against.
FAQ
Is Hunyuan3D really free?
The model weights are, as open-source releases from Tencent (github.com/Tencent-Hunyuan). The real costs are hardware and time: a GPU with substantial VRAM, environment setup, and per-generation compute. If you rent cloud GPUs instead, each generation has a real dollar cost that varies with the instance. PolyMedium's $0.25 per object includes all compute, hosting, and free exports.
Which produces better models?
It depends on the subject, honestly. Hunyuan3D's prior-heavy approach is strong on organic characters, where a learned shape prior fills in convincing anatomy. On flat drawings and doodles, depth priors work against you: adding imagined depth distorts the art, which is why PolyMedium routes drawings to a scan-faithful pipeline that preserves the marks. For mechanical objects, faithful reconstruction of the actual reference tends to beat a prior's guess.
Which is better for a game team?
For most teams, hosted wins on total cost: no ML engineer time, no GPU queue, and exports arrive at engine-ready polycounts in GLB with PBR. Self-hosting Hunyuan3D makes sense when volume is very high, the pipeline needs customization, or content policy requires everything to stay on your own metal.