What is HY-3D-Retopology?
HY-3D-Retopology is Tencent’s specialized service for retopology: reorganizing the surface structure of an existing 3D mesh so that it has a more regular topology and, when requested, fewer polygon faces. A polygon is one of the small geometric surfaces that make up a 3D object. Reducing unnecessary polygons can make an asset easier to edit, animate, simulate, store, or render in real time.
The model is part of Tencent’s Hunyuan 3D offering and is exposed through Tencent Cloud TokenHub. Tencent’s documentation identifies the underlying technology as the Polygon 1.5 model and gives the API model identifier as hy-3d-retopology. The service is therefore best understood as a production asset-processing tool, not as a general-purpose conversational or generative AI model.
What the model does
HY-3D-Retopology takes an existing 3D file as input and creates a revised mesh. Its purpose is to improve the organization of the mesh and reduce its face count according to the selected processing level. This can be useful when an original asset is too dense for a target application or has topology that is difficult to work with.
The available controls documented by Tencent are practical rather than language-oriented. Users can choose between triangular and quadrilateral surface topology, and can select a high, medium, or low reduction level. The choice affects the balance between geometric detail and mesh simplicity. The supplied documentation does not define exact polygon-count targets for those levels, so they should not be treated as fixed percentage reductions.
| Capability | Documented support |
|---|---|
| Input files | OBJ, GLB, and FBX |
| Topology choices | Triangle or quadrilateral |
| Reduction levels | High, medium, or low |
| Example output files | OBJ and GLB |
| Processing model | Asynchronous submission and status polling |
| API model identifier | hy-3d-retopology |
Supported inputs and outputs
The documented inputs are 3D files in OBJ, GLB, or FBX format. This makes HY-3D-Retopology suitable for workflows that already have a mesh and need a processing step before the asset moves into a game engine, renderer, animation package, simulation pipeline, or further modeling work.
The public example response includes generated OBJ and GLB files. A completed response can also include an image resource associated with the result. The documentation does not describe the service as a text, image-generation, audio, video, or speech model. It also does not document image-to-3D or text-to-3D input for this particular model. Its central input and output are 3D mesh assets.
Because the service works on uploaded files, it should not be evaluated using language-model measures such as context-window size, token output limits, reasoning benchmarks, or coding benchmarks. Tencent does not publish a context length or maximum output token limit for HY-3D-Retopology, and those limits are not directly applicable to its documented mesh-processing workflow.
How the API workflow works
HY-3D-Retopology uses an asynchronous task pattern. An application submits a source 3D file and the requested retopology settings through Tencent Cloud TokenHub. The service returns a task identifier, which the application uses to query the task status. The client waits until processing is complete and then retrieves the available result resources.
This workflow is different from a streaming text API. The supplied documentation does not describe token streaming or a continuously streamed mesh result. Applications should therefore be designed around job submission, polling, completion handling, and downloading the resulting files.
Tencent documents a default concurrency limit of one task per account. That restriction matters for batch processing: a pipeline that needs to process many assets may have to queue jobs rather than submit them all at once. Tencent also states that task identifiers remain valid for 24 hours, so applications should record the task ID and retrieve completed outputs within that period.
Strengths and practical value
- Focused asset processing: The service addresses a specific production problem instead of requiring a general 3D-generation workflow.
- Multiple common file formats: OBJ, GLB, and FBX support covers several widely used exchange and asset formats.
- Topology selection: Triangle and quadrilateral options allow the output to be aligned with different downstream modeling or rendering requirements.
- Adjustable reduction: High, medium, and low settings provide a simple way to choose between more aggressive simplification and greater retention of mesh detail.
- API integration: TokenHub access makes the service usable as one step in an automated asset pipeline rather than only as a manual desktop operation.
These are documented capabilities, not a guarantee that every source mesh will produce the same quality. Retopology results can depend on the shape, density, topology, and complexity of the uploaded asset. Tencent’s public guide does not provide independent benchmark results or quality scores.
Limitations and unknowns
The main limitation is that HY-3D-Retopology is a narrow-purpose service. It does not provide a documented interface for writing text, generating images, producing speech, answering general questions, writing code, or reasoning over arbitrary instructions. It is also not presented as a text-to-3D or image-to-3D model. If the starting point is a description or a picture rather than an existing mesh, a different type of 3D-generation system would be more appropriate.
The public documentation leaves several technical details unspecified. It does not state a model release date, context window, maximum output token count, knowledge cutoff, model-specific price, or exact polygon-reduction percentages. It also does not document fine-tuning, caching, batch API support, function calling, tool use, or real-time streaming for this model. These omissions should be treated as unknowns rather than as evidence that the features are available or unavailable in every Tencent environment.
The default one-task concurrency limit may also be a practical constraint for studios or services processing large asset collections. Although asynchronous processing is appropriate for jobs that may take time, it introduces more application complexity than an immediate request-and-response operation. A production integration needs status polling, retry handling, task expiration handling, and output download logic.
Pricing and availability
Tencent’s supplied model documentation does not identify a model-specific price for HY-3D-Retopology. No input or output token pricing applies to the documented mesh workflow, and no official per-file or per-task price was identified in the current research. Prospective users should check their Tencent Cloud account and the current TokenHub service terms before estimating operating costs.
The model is documented for API use through Tencent Cloud TokenHub. The supplied research identifies a calling guide updated September 9, 2026. Availability, account requirements, quotas, and commercial terms can change independently of the model’s functional description, so the live Tencent Cloud documentation remains the authoritative source for access details.
When to choose HY-3D-Retopology
Choose HY-3D-Retopology when you already have a 3D asset and need an automated retopology or polygon-reduction step. It is a logical option for preparing dense meshes for games, real-time rendering, animation, simulation, downstream editing, or asset libraries where file complexity needs to be controlled.
It is especially relevant when an automated API workflow is more useful than manually rebuilding topology for every asset. The triangle-versus-quadrilateral choice and the three reduction levels provide basic control without requiring the caller to implement its own mesh-processing algorithm.
Another option may be more appropriate when the task begins with text or images, when detailed manual control of the topology is essential, or when a high-volume pipeline requires more than the documented one-task default concurrency. A general 3D-generation model is better suited to creating an object from a prompt or image, while a dedicated modeling tool may be preferable when an artist must inspect and edit individual edge loops directly.
Bottom line
HY-3D-Retopology is a focused Tencent service for transforming existing high-polygon meshes into more manageable 3D assets. Its documented value lies in file-based retopology, configurable triangle or quadrilateral topology, selectable reduction levels, and asynchronous API integration. It should be judged as a specialized mesh-processing component rather than as a general AI model. The lack of published pricing, exact reduction targets, throughput beyond the default concurrency information, and formal quality benchmarks means that teams should test representative assets and verify current Tencent Cloud terms before committing to a production workflow.