AI 3D Model Generator

Transform reference photos or text descriptions into textured 3D models. Supports multi-view geometry reconstruction, PBR material maps, real-time 3D preview, and multi-format export.

Workspace

Model

Hunyuan 3D Pro
ImageText

Multi-view reconstruction with up to 8 reference angles. Image or text input; PBR is optional.

Polygon budget

500K

Default face budget

PBR Materials

Required *

FrontChoose Image

Optional

Back
Left
Right
Top
Bottom
Left Front
Right Front

Supports JPG, PNG, and WebP up to 8 MB. Single subjects on plain backgrounds deliver optimal results.

AI services upload the images or text you provide to our servers for processing.

Generation Showcase

About This Tool

This tool calls cloud AI 3D generation models to convert single front photos, multi-angle reference images, or descriptive text prompts into textured 3D triangle meshes in standard GLB format. Uploaded images are retained in private storage for up to 14 days and then deleted, and generated assets can be inspected interactively in your browser or exported directly to FBX, OBJ, or STL.

Credit Consumption Rules

Free accounts receive 0 credits; Plus or Pro includes monthly credits, and credits already in your balance remain usable. Additional credit packs are one-time purchases and do not auto-renew.

Generation OptionCredit Cost and Notes
Base Model GenerationConsumes 100 to 300 credits depending on compute requirements. The current price is displayed next to the model selector.
Multi-View ReconstructionAdds a flat 100 credits on compatible models (such as Hunyuan 3D Pro), regardless of the number of extra angles supplied.
PBR Material GenerationAdds 100 credits when enabling metallic, roughness, and normal map generation. Included by default on select models.
Quality SettingsAdds 100 credits for a Hunyuan 3D Pro custom polygon budget (300K or 1M) and for Tripo H3.1 detailed geometry or texture quality. Trellis 2 resolution and Meshy 7.1 geometry resolution are included in the base price.

How to Use

  1. 1

    Import Source Reference

    The front orientation is required in image mode, with optional complementary angles. Alternatively, switch to text mode for prompt-driven creation.

  2. 2

    Refine Angles and Materials

    Add side or rear angles to eliminate hidden-surface distortions, or enable PBR textures for physically grounded surface reflections.

  3. 3

    Cloud Inference and Inspection

    Submit the task for cloud inference. Once complete, inspect surface geometry and texture seams from every angle in the 3D viewport.

  4. 4

    Export and Format Conversion

    Download the production GLB asset, or convert it immediately to FBX, OBJ, or STL formats. Large files can be optimized via our 3D Compression tool.

Generation Models Compared

All four models return a textured GLB. They differ in accepted inputs, geometry density, and how much of the surface is reconstructed from reference images rather than inferred by the model.

ModelBaseQuality optionsPBRMulti-viewMaximum
Hunyuan 3D Pro100 credits+100 for custom face count (300K or 1M)+100+100400 credits
Trellis 2100 creditsIncluded at any resolution and texture sizeNo PBRNot supported100 credits
Tripo H3.1100 credits+100 detailed geometry; +100 detailed textureIncludedNot supported300 credits
Meshy 7.1300 creditsIncluded at any geometry resolutionIncludedNot supported300 credits

Hunyuan 3D Pro

Tencent's Hunyuan 3D Pro accepts text or a front image, plus up to eight reference angles of the same object. Multi-view input constrains surfaces the camera never saw, so the back and top are reconstructed from your photos instead of invented. PBR maps are optional.

Trellis 2

Microsoft's open 4B-parameter image-to-3D model. It works from a single photograph and produces dense geometry with sharp 2K textures. With one view, hidden surfaces are inferred, so choose subjects whose defining shape is visible from the front.

Tripo H3.1

Tripo's flagship model, aimed at high-fidelity assets with PBR materials included. Upstream it supports up to two million faces and quad meshes; this site generates from one image or a text prompt with selectable standard or detailed geometry and texture quality.

Meshy 7.1

Meshy 7.1 targets fine surface detail such as fabric weave, stitching, and scales, and returns base color, metallic, roughness, and normal maps. This site runs the full quality mode with textures and PBR enabled.

Frequently Asked Questions

View all FAQs

When should I use image-to-3D versus text-to-3D generation?

Use image mode when you have existing physical items or concept art. The front view is mandatory (supporting JPG, PNG, and WebP up to 8 MB) and yields high geometric fidelity. Text mode offers creative freedom when concepting original shapes from scratch by describing geometry, material finishes, and styles.

What are the advantages of multi-view input, and how is it billed?

With a single front image, unseen surfaces are hallucinated based on model priors. Supplying back, side, or top and bottom photos grounds the geometry and eliminates blind-spot distortion. On supported architectures such as Hunyuan 3D Pro, enabling multi-view adds a flat 100 credits, regardless of how many extra angles you upload.

When should PBR material maps be enabled?

Enable PBR when the model is destined for game engines, real-time WebGL rendering, or photorealistic visualization requiring metallic, roughness, and normal maps. If you only require a geometric mesh for 3D printing or prototyping, keeping PBR off saves 100 credits and allows direct STL export without intermediate conversion.

What image characteristics produce the best 3D reconstruction?

A single centered subject filling roughly 70% or more of the frame against a neutral, uncluttered background with even diffuse lighting yields the cleanest meshes. Avoid collage layouts, heavy occlusion, transparent or reflective glass, and embedded logos or watermarks. Multi-view inputs must depict the exact same physical object.

How should I structure text prompts for 3D modeling?

Combine the primary subject name with spatial geometry, surface material details, and design aesthetics. For instance, a weathered terracotta jug with braided hemp handles in realistic style provides actionable geometric cues. Avoid vague subjective buzzwords such as high quality or photorealistic.

Do generated 3D meshes have physical dimensions for 3D printing?

The output is a normalized polygonal triangle mesh without real-world millimeter or inch measurements, and is not a parametric CAD solid. For 3D printing, you can export STL directly when PBR is disabled, or import the GLB into slicers like Bambu Studio or Cura to rescale to target dimensions and verify wall thickness.

How long does generation take, and what happens if a task fails?

Depending on cloud queue volume and the selected model, processing generally completes within 1 to 3 minutes. If a run fails due to network interruptions, upstream downtime, or exceeds 30 minutes, the system marks it failed and automatically refunds all deducted credits. Free accounts receive no credits; Plus or Pro includes monthly credits, and credits already in your balance remain usable. Plus runs up to two, and Pro runs up to four concurrently.

How long are model download links retained?

Uploaded input images and generated 3D files are stored in private cloud storage for 14 days. Download links are valid for 14 days; after that, the files are permanently deleted and cannot be reissued. Download and save your models within the retention window.