Explore enterprise products and AI capabilities built for industry.
Let’s talk about your businessProducts, integrations and partnershipsBring AI into the enterprise.
Automated data-warehouse builder
Automated high-quality datasets
AI appliance for business decisions
Opens in a new tabDedicated offline AI bidding appliance
Connect AI to products and operations.
Explore the interfaces and integration paths for our AI capabilities.
Models, foundational frameworks and the engineering behind AI systems.
Explore collaboration across products, technology and delivery.
Frameworks, tools and examples for developers to understand and reuse.
Updates, technical thinking and industry observations from IDENIFE.
Our focus on education, talent and sharing knowledge.
HuanYu · Image-to-3D API
Integrate the capabilityBring a visual idea into 3D work.
Condition on an image. Build a spatial representation. Bring geometry and materials into design software. HuanYu connects the journey from visual concept to 3D asset through an API.
Explore the technical pathIllustration of spatial representations
For industrial design platforms, 3D tools and content pipelines
Delivered through an APIA concept should lead to more than a presentation. Image-to-3D gives designers a spatial starting point to inspect, modify and organise. Its value is the connection between design stages, not simply another file.
Identify silhouette, proportion, material and part cues.
Image conditionsUse generative priors to infer views and occlusion.
Spatial representationPrepare geometry, materials and hierarchy for editing.
Editable assetHow generation works
Image conditions, spatial priors and geometric constraints work together: starting from visible surfaces, towards an object that can be viewed from different angles and prepared for editing.
Subject separation and image normalisation define the object. Silhouette, local boundaries, perspective and material cues constrain spatial inference; generative priors supply candidate explanations for the back and occluded regions.
Unseen structure is inferred. Additional views can provide constraints; a complete-looking render is not evidence of the hidden geometry.
Cross-view consistency is central to image-to-3D. A form that works from the front should preserve its proportions, part relationships and surface continuity from the side. Multi-view conditions help organise a coherent representation instead of a collection of unrelated pictures.
Appearance consistency is considered alongside silhouette, depth and surface-normal relationships.
Gaussians represent appearance; editing tools need explicit surfaces. Surface constraints and geometry extraction produce a mesh, followed by work on disconnected fragments, normals, topology and polygon budgets. Texture baking and UV organisation connect appearance to conventional material workflows.
Surface extraction, retopology and material preparation are separate steps, not a format conversion.
Single-image generation relies on image priors. Classical 3D Gaussian Splatting reconstructs a radiance field from multiple observed views. They address different problems; Gaussian representations provide a spatial basis for generation and rendering.
Gaussian splatting
Each Gaussian has a position, scale, orientation and appearance. Projected primitives contribute pixels through visibility-aware alpha compositing. Differentiable rendering lets image-space errors inform the spatial representation.
A renderable representation is not an editable surface.Gaussians and meshes serve different roles in appearance and geometric editing. Shape, topology and materials still need independent checks during conversion.
Structure for industrial design
A structured object makes editing addressable.
Our industrial-design research explores part-level generation. Semantics, geometric boundaries and part relationships help organise housings, functional elements and supports into addressable objects, giving local shape changes, material substitutions and design reviews a defined scope.
Silhouette · shell material
Local mesh · independent material
Transform · hierarchy
Meshes, materials, hierarchy and transforms provide a starting point for editing, review and scene composition in 3D software.
Parametric solids, precise surfaces, dimensional tolerances and assembly constraints must be established and validated in CAD and engineering workflows. Part separation is not engineering-solid generation.
An asset for form exploration, visualisation or further modelling has different acceptance criteria. Translate its intended use into checks across appearance, geometry, materials and scene organisation.
Silhouette drift / part proportions / cross-view consistency
The shape should remain coherent across viewpoints.Normals / non-manifold edges / floating fragments / face count
Check the mesh for its next use, beyond the rendered image.UV overlap / seams / texture dependencies / colour space
Carry the appearance into the destination tool.Axes / unit conventions / naming / hierarchy / version
Make each delivered object identifiable, traceable and editable.Without a real-world scale reference, model dimensions depend on an agreed convention. Manufacturing accuracy and feasibility require engineering validation.
A transition in an automotive design workflow: choose a concept direction, then inspect its shape and proportions in 3D software. The image establishes intent; the model carries the next stage of work.
Design workflow example · The model is shown in an editing-tool screenshot, not a measurement of generation accuracy or manufacturing capability.
API integration
Keep your projects, accounts and design workflow. Integrate image-to-3D as an asynchronous capability: submit a task, follow its stages, receive assets and return them to the original environment.
A task identifier connects inputs, stages, failure reasons and artifact versions. Execution completion and quality outcomes remain separate.
Separate submission from result retrieval. Agree on polling or callbacks, timeout handling and duplicate-request policies so a long generation process does not depend on one page request.
Specify model formats, texture references, coordinate systems, units and versions, then validate import in the target software. Available formats, stages and error fields follow the actual service agreement and integration documentation.
Published research explains the methods; it does not establish HuanYu’s implementation or performance.
From design intent, to a visual direction, to space.
Explore HuanYu Design Generation