What is HY-3D-Texture?
HY-3D-Texture is Tencent's dedicated model for texturing existing 3D geometry. Its job is to add visual appearance to a mesh rather than create the mesh itself. That appearance can include base color, surface detail, and, when enabled, physically based rendering (PBR) material information.
This distinction is important. A mesh-generation model attempts to produce the shape and structure of an object. HY-3D-Texture assumes that the shape is already available in an OBJ or GLB file and focuses on making that shape look like a supplied reference. A suitable input might be a scanned object, a CAD-derived asset, a game prop, or a basic untextured model that needs materials before it is used in a scene.
The model is available through Tencent Cloud's TokenHub API under the canonical identifier hy-3d-texture. TokenHub processes requests as asynchronous jobs: a client submits the mesh and configuration, receives a job identifier, and then checks the job until the service reports completion or failure.
Where it fits in Tencent's 3D model lineup
HY-3D-Texture occupies a focused position in Tencent's HY-3D family. It is not presented as a general-purpose language model, image generator, or complete 3D production system. Its role is narrower and more practical: convert an existing untextured or under-textured mesh into a textured 3D asset.
That positioning makes it complementary to geometry-generation tools rather than a replacement for them. If the main problem is creating an object's shape from scratch, HY-3D-Texture is the wrong starting point. If the shape is already correct but lacks convincing materials or surface detail, the model addresses the relevant stage of the workflow.
Supported inputs and outputs
The documented API accepts OBJ and GLB input meshes. A reference image can be supplied by URL or as base64-encoded image data. The reference provides visual guidance for the texture-generation process, so the result depends partly on how clearly the reference communicates the intended colors, materials, and visible details.
Supported 3.1 configurations also document multi-view image conditioning. These inputs can represent views such as left, right, back, top, bottom, and angled front perspectives. Multiple views are useful when a single image does not show enough of the object. For example, a front-only image may provide little information about the back of a product or the sides of a game asset.
Completed jobs can provide textured 3D assets and related files, including OBJ, GLB, MTL, and texture-image outputs. The exact result set can depend on the request and the service configuration. The model's output is therefore an asset package rather than a text response, JSON explanation, or conventional image-only answer.
Texture resolution, PBR materials, and UV preservation
HY-3D-Texture exposes several controls that matter in production workflows. The texture_size setting supports square texture dimensions from 720 to 4096 pixels. Smaller textures can be more practical for previews or bandwidth-sensitive workflows, while larger textures may be preferable when the asset will be viewed closely. The available range is a documented API limit, not a guarantee that every asset will benefit from the maximum setting.
An optional PBR mode can generate physically based material information. PBR is a rendering approach that separates material properties such as color and surface response so that an asset behaves more consistently under different lighting conditions. Tencent's related Hunyuan3D-Paint 2.1 implementation provides technical context for this area, including albedo and metallic-roughness processing, multi-view rendering, texture baking, and texture inpainting. That related implementation should not be read as a claim that every feature in it is independently exposed by the HY-3D-Texture TokenHub endpoint.
The API also provides an option to preserve the input model's UV layout. UV coordinates describe how a flat texture is mapped onto a three-dimensional surface. Keeping the original UVs can be important when downstream tools, existing materials, or hand-authored texture coordinates rely on that layout. Users who do not need to retain the source mapping can evaluate the default behavior and generated results against their pipeline requirements.
How a typical API workflow works
A normal workflow begins with an OBJ or GLB mesh hosted at a location accessible to the service. The client submits a request to the Tencent Cloud TokenHub 3D endpoint, identifies the model as hy-3d-texture, and includes an optional reference image. The request can also set PBR generation, UV preservation, and the desired square texture resolution.
The service returns a job identifier rather than immediately returning a finished model in the same request. The client then polls the corresponding job status or otherwise follows the documented asynchronous process. Once the job completes, the client retrieves the available asset files. Applications should account for this job-based design when building user interfaces, queues, retries, storage, and error handling.
For a simple product-preview pipeline, the sequence could be: upload an untextured GLB, attach a product reference image, request a 2048-pixel texture with PBR enabled, wait for the job to complete, and pass the resulting GLB and texture files to a viewer. For a larger asset library, the same pattern can be placed behind a batch queue, although the supplied documentation does not identify a dedicated batch API feature.
What HY-3D-Texture does well
- Focused task definition: The model is specialized for texturing existing geometry, which makes its purpose clearer than a general 3D generation system.
- Reference-guided appearance: A supplied image can steer the generated colors and surface appearance toward a particular design.
- Multi-view support on documented 3.1 configurations: Additional views can provide information that is missing from a single reference image.
- Production-oriented controls: PBR generation, UV preservation, and configurable texture size give users more control than a simple one-click preview tool.
- Common 3D asset formats: OBJ and GLB inputs, together with possible OBJ, GLB, MTL, and texture outputs, fit common asset workflows.
- Cloud API access: The TokenHub interface makes the model suitable for applications and automated pipelines rather than only interactive experimentation.
These are capability-based strengths, not a claim that every generated result will be production-ready without review. Mesh topology, UV quality, reference clarity, view coverage, and resolution can all affect the outcome.
Important limitations
The most fundamental limitation is that HY-3D-Texture requires existing geometry. It does not replace a mesh generator, retopology tool, rigging system, or animation solution. A user who has only a text prompt or a single image and needs a complete animated character must use other tools for the missing stages.
Reference coverage is another practical constraint. A single image may clearly show one side of an object while leaving the remaining surfaces ambiguous. Multi-view inputs can reduce that ambiguity where the supported 3.1 configuration is available, but they do not eliminate the need for consistent and useful references.
Preserving UVs can help a pipeline, but the quality and suitability of the original UV layout still matter. A poorly unwrapped mesh may remain difficult to texture even when UV preservation is disabled. Similarly, selecting a 4096-pixel texture does not automatically create more useful detail if the source reference or mesh does not support it.
Tencent's supplied documentation does not identify a conventional context window, maximum output-token limit, knowledge cutoff, fine-tuning capability, streaming mode, caching feature, or batch API for this specialized 3D model. Those concepts are primarily associated with language-model services and should not be assumed here. The API's documented output constraints are expressed in terms of asset formats, job results, and texture dimensions instead.
Pricing and availability
Tencent Cloud's TokenHub pricing documentation lists HY-3D-Texture at 30 points per generation. The same documentation states that one point corresponds to 0.12 yuan under the documented HY-series 3D billing system. On that basis, 30 points correspond to 3.60 yuan per generation, although the applicable account, billing configuration, or Tencent Cloud pricing conditions should be checked before budgeting production usage.
This is point-based postpaid pricing rather than a conventional monthly subscription price. The supplied research does not identify separate input and output token prices, a recurring plan price, or a confirmed volume-discount schedule. Since each generation can produce a different asset package and may use different controls, teams should test representative meshes and references before estimating total operating cost.
Capability profile and trade-offs
HY-3D-Texture is not a text model, so conventional reasoning and coding capabilities are not relevant to its primary task. It does not provide tool or function calling in the language-model sense, and the documented interface is not a streaming text-generation API. Its useful output is a textured 3D asset and related files.
From a speed and cost perspective, the model is best understood as a specialized asset-processing service. Its focused scope can be more appropriate than paying for or adapting a broader 3D system when the mesh already exists. However, the supplied research does not provide a verified latency benchmark, so no fixed completion time should be assumed. The asynchronous job design means applications need to tolerate processing time rather than expecting an immediate response.
The editorial ratings associated with this entry score speed at 6 out of 10 and cost at 7 out of 10. These are internal comparative assessments, not Tencent-published benchmark results or official quality guarantees. They suggest a reasonably practical balance for a specialized cloud workflow, but actual performance and value depend on mesh complexity, output settings, queue conditions, and the quality standard of the project.
When to choose HY-3D-Texture
Choose HY-3D-Texture when the geometry is already available and the main requirement is to automate or accelerate texturing. It is a good candidate for:
- Applying a product's visual design to an existing 3D model for e-commerce or visualization.
- Adding materials to scanned objects or CAD-derived assets.
- Preparing game, virtual-world, or simulation assets that need texture maps before integration.
- Generating textured previews for large asset libraries.
- Building a cloud workflow that accepts OBJ or GLB files and returns downloadable textured assets.
- Preserving an existing UV layout or requesting PBR-oriented material output.
Another option is more appropriate when the required result is new geometry, accurate mechanical reconstruction, retopology, animation, rigging, or a complete character-production pipeline. A general image-generation system is also not a direct substitute: it can create a picture of an object but does not necessarily produce a usable textured OBJ or GLB asset.
What to evaluate before production use
Before committing a large asset set, test representative meshes rather than relying on a simple demonstration object. Include assets with different topology, UV layouts, material types, and levels of geometric detail. Compare single-view and multi-view references where supported, and inspect hidden or less-visible surfaces instead of judging only the front-facing result.
Also decide whether UV preservation or regenerated mapping better serves the downstream application. Test the smallest texture size that meets the visual requirement, because a larger file can increase storage and delivery costs without improving unclear source detail. Finally, validate the downloaded OBJ, GLB, MTL, and texture files in the actual target software, since successful job completion does not by itself confirm that every downstream import workflow will interpret the asset exactly as expected.
Bottom line
Tencent HY-3D-Texture is a narrowly focused solution for turning existing OBJ or GLB geometry into reference-guided textured assets. Its most useful differentiators are optional multi-view conditioning on supported versions, PBR material generation, UV preservation, square texture sizes from 720 to 4096 pixels, and cloud-based asynchronous processing. At 30 Tencent Cloud points per generation, it is positioned for developers and production pipelines that need automated 3D texturing rather than a general-purpose AI model.
Its limits are equally clear: it does not create the mesh from scratch, and the supplied research does not establish a fixed latency, context-style limit, or broad set of language-model features. For the specific problem of texturing geometry that already exists, however, its focused API and asset-oriented outputs make it a practical option to evaluate.