The Hi3d/Texture Model: How it Can Create Textured 3D Assets

Overviewhi3d/texture is maintained by hitem3d. Hi3D Texture applies a reference image to an existing geometry mesh and returns a textured 3D asset. It fits workflows that already have a GLB mesh and need image-guided surface appearance rather than new geometry generation. The key decision is input compatibility: the mesh must be supplied as a GLB, while the reference image can be PNG, JPEG, or WebP up to 20MB. The endpoint supports the hitem3dv1.5 and scene-portraitv1.5 texture models, with resolution choices constrained by the selected model.Best Use CasesTexture an untextured GLB produced by another 3D-generation workflow, including geometry from an image-to-3D endpoint with texturing disabled.Prototype image-guided asset appearance for games, visualization, product concepts, or digital scenes when the geometry already exists.Reuse one mesh with different reference images to explore surface treatments without regenerating geometry.Export a generated asset in a downstream-friendly format such as GLB, OBJ, STL, FBX, or USDZ.LimitationsThe endpoint textures existing geometry; it is not described as a text-to-3D or image-to-3D generation endpoint.Only GLB mesh inputs are supported.Reference images must be PNG, JPEG, or WebP and no larger than 20MB.Resolution support depends on the selected model: hitem3dv1.5 accepts 512, 1024, 1536, and 1536pro; scene-portraitv1.5 accepts 1536 only.Only v1.5 models support the staged texture task.The safety checker is enabled by default and checks input images before processing when enabled.The supplied documentation does not specify pricing, latency, training data, hardware, or a license.How it comparesh3.1/image-to-3d — h3.1/image-to-3d generates a high-quality 3D model from a single image, whereas Hi3D Texture starts with an existing GLB mesh and focuses on applying a reference image as texture. Choose Candidate 1 when geometry generation is the requirement; choose Hi3D Texture when geometry is already available.hunyuan-3d-3.1 — hunyuan-3d-3.1 is described as producing 3D models with texture fidelity and geometry precision. Hi3D Texture exposes a narrower staged texturing workflow with explicit GLB input and selectable export formats.h3.1/text-to-3d — h3.1/text-to-3d generates 3D models from text descriptions. Hi3D Texture requires both a reference image and an existing GLB mesh, so it is not a text-only asset-generation alternative.hunyuan3d-v3/text-to-3d — hunyuan3d-v3/text-to-3d turns simple sketches into detailed, fully textured 3D models and targets Unity, Unreal, and Blender-ready formats. Hi3D Texture instead modifies an existing GLB using an image reference and offers explicit GLB, OBJ, STL, FBX, and USDZ export choices.pixal3d — pixal3d turns a single image into a high-fidelity 3D model with detailed geometry and realistic textures. Hi3D Texture does not claim to create geometry from the image; it textures supplied GLB geometry.Technical SpecificationsEndpoint: fal-ai/hitem3d/texture. Creator: hitem3d. Platform: fal. Category tag: Image-to-Image. The model description identifies Hi3D as the texturing system. Available model versions are hitem3dv1.5 and scene-portraitv1.5. hitem3dv1.5 supports texture resolutions 512, 1024, 1536, and 1536pro; scene-portraitv1.5 supports 1536. The default resolution is the selected model's recommended resolution. Generated model export formats are GLB, OBJ, STL, FBX, and USDZ, with GLB as the default. Input-image safety checking is enabled by default.Model Inputs and OutputsRequired inputs are image_url (string) and mesh_url (string). image_url must point to a PNG, JPEG, or WebP reference image up to 20MB. mesh_url must point to a GLB geometry model; the schema gives a DamagedHelmet GLB as an example and notes that geometry from an image-to-3D endpoint with enable_texture=false can be used. model is an enum with hitem3dv1.5 and scene-portraitv1.5, defaulting to hitem3dv1.5. resolution is nullable and accepts 512, 1024, 1536, or 1536pro as strings; compatibility depends on model. export_format is an enum of glb, obj, stl, fbx, and usdz, defaulting to glb. enable_safety_checker is boolean and defaults to true. The required output model_mesh is a file object containing a download url and optional content_type, file_name, and file_size. thumbnail is optional and may be null; when present, it has the same file metadata structure.Getting Startedimport { fal } from "@fal-ai/client"; const result = await fal.subscribe("fal-ai/hitem3d/texture", { input: { image_url: "https://storage.googleapis.com/falserverless/example_inputs/dog.png", mesh_url: "https://raw.githubusercontent.com/KhronosGroup/glTF-Sample-Models/master/2.0/DamagedHelmet/glTF-Binary/DamagedHelmet.glb", model: "hitem3dv1.5", resolution: "1024", export_format: "glb", enable_safety_checker: true }, logs: true, onQueueUpdate: (update) => { if (update.status === "IN_PROGRESS") { update.logs?.forEach((log) => console.log(log.message)); } }); console.log(result.data.model_mesh.url); console.log(result.data.thumbnail?.url); Hi3D: Pursuing High-Resolution Image-to-3D Generation with Video Diffusion Models — The supplied Hi3D paper, “Hi3D: Pursuing High-Resolution Image-to-3D Generation with Video Diffusion Models,” describes Hi3D as a video-diffusion-based system for high-resolution 3D content from a single image. This supports the model family’s image-to-3D context, but the endpoint schema here specifically documents the staged task of texturing an existing GLB mesh.Frequently asked questionsQ: What does this endpoint generate?A: It generates a textured 3D model from a reference image and an existing geometry mesh.Q: What mesh format can I upload?A: Only GLB is supported for mesh_url.Q: Which image formats and sizes are supported?A: PNG, JPEG, and WebP images are supported, up to 20MB.Q: Which model should I select?A: hitem3dv1.5 is the default and supports 512, 1024, 1536, and 1536pro. scene-portraitv1.5 supports 1536 only.Q: What happens if I omit resolution?A: The endpoint uses the selected model’s recommended resolution.Q: What output formats are available?A: You can request glb, obj, stl, fbx, or usdz. GLB is the default.Q: Is safety checking enabled automatically?A: Yes. enable_safety_checker defaults to true and checks input images before processing when enabled.Q: What output must my application handle?A: model_mesh is always required in the response and includes a download URL. An optional thumbnail may also be returned.This is a simplified guide to an AI model called hi3d/texture, maintained by hitem3d. If you like these kinds of analyses, join AIModels.fyi or follow us on Twitter.