What is Hunyuan3D 3.5?

Hunyuan3D 3.5 is part of Tencent Hunyuan3D series — an open-source diffusion-based 3D generation system that turns text and images into textured 3D assets. This page covers the 3.5 series; the live generator launches from the Hunyuan3D tool page.

How do I use it online?

Upload a single reference image, pick a generation mode, and click generate. The job runs on fal.ai via Trigger.dev — you watch progress in real time, then preview the result in the interactive 3D viewer.

How much does each generation cost in credits?

Credits are tiered by output. Geometry (bare mesh, no texture) is 2 credits. Normal (with texture) is 3 credits. Normal with PBR materials is 4 credits. New users receive free credits to get started.

How long does generation take?

Most jobs finish in well under a minute on fal.ai. Exact time depends on the polygon count you select and whether PBR textures are enabled.

What file format is the download?

GLB. The generated mesh — with PBR textures baked in when enabled — is delivered as a single .glb file you can drop into Blender, Unity, Unreal, or any glTF-aware viewer.

What polygon counts can I choose?

From 40,000 to 1,500,000 faces, with 500,000 as the default. Lower counts are lighter for real-time engines; higher counts preserve fine detail for rendering and 3D printing.

Can I try it without signing up?

Yes. Anonymous generation is enabled, so you can upload an image and generate a model before creating an account. Sign in to keep your generations and manage credits.

Do I need a GPU or any installation?

No. All inference runs in the cloud on fal.ai and Trigger.dev. You only need a modern browser — the open-source Hunyuan3D also exists for local use, but this tool does not require it.

Built on Tencent Hunyuan3D

Hunyuan3D 3.5: open-source text & image to 3D generation

Hunyuan3D 3.5 is Tencent open-source 3D generation: a two-stage diffusion pipeline that turns text and images into textured 3D models. It pairs a DiT shape model with a Paint texture model and ships in Pro and Rapid editions with up to 8K PBR textures.

Start Generating View Architecture

Pro & Rapid editions·8192² PBR textures·Open-source lineage

Two-stage diffusion pipeline

Shape and texture are decoupled, so each stage specializes — and you can re-texture any mesh.

Stage 1

Hunyuan3D-DiT — Shape

A scalable flow-based diffusion transformer. Takes a conditioned image (or text-rendered image) and produces a bare mesh that aligns with the prompt. The 2.1 release ships a 3.0B DiT and a 0.6B mini variant.

6 GB VRAM for shape-only
0.6B mini / 3.0B full / 1.1B multiview
Turbo and Fast distillations cut steps

Stage 2

Hunyuan3D-Paint — Texture

A 1.3B texture synthesis model that uses strong geometric and diffusion priors to bake high-resolution PBR maps onto the generated — or hand-crafted — mesh.

Albedo, normal, metallic, roughness
Up to 8192×8192 resolution
Re-texture any existing mesh

bare mesh → textured asset

Input modalities × editions

Four input types meet two editions. Pro unlocks everything; Rapid optimizes the common path.

Input	Pro edition	Rapid edition
Text Up to 1024 characters of natural language	Full support	Full support
Image Single reference photo or concept art	Full support	Full support
Multi-View Up to 4 angle photos for precise geometry	Full support	Not available
Sketch Line drawing + attribute description	Full support	Not available

Sketch and multi-view are Pro-only. Rapid focuses on text and single-image for sub-60s turnaround.

Generated with Hunyuan3D 3.5

Real outputs from text and image input. No post-processing.

Text → 3D

Describe it, get a fully textured model. No reference needed.

anime-style female engineer in orange and gray power armor

retro sci-fi ray gun, turquoise, red and gold stylized weapon

stylized treasure chest with gold coins, padlock, vines and moss

Image → 3D

Upload a photo or concept art; get a matching textured mesh.

Technical specifications

Open-source reference numbers from Hunyuan3D 2.1, plus platform-layer 3.5 additions.

Models

Shape (DiT-v2-1): 3.0B params
Texture (Paint-v2-1): 1.3B params
Mini shape: 0.6B params
Multiview shape: 1.1B params

Runtime

VRAM — shape only: 6 GB
VRAM — shape + texture: 16 GB
Pro edition time: 1–3 min
Rapid edition time: <60s

Output

Polygon range: 40K – 2M
Topology: Triangle or quad
PBR resolution: up to 8192²
Export formats: GLB, OBJ

Open-source since November 2024

Tencent released the first 3D models in Nov 2024; the community made it one of the most downloaded 3D generative model families on Hugging Face.

3M+

Hugging Face downloads

150+

Enterprises on Tencent Cloud

20/day

Free generations (official engine)

Tencent released the first 3D models in Nov 2024; the community made it one of the most downloaded 3D generative model families on Hugging Face

Nov 2024
First Hunyuan3D models open-sourced
Jan 2025
Hunyuan3D 2.0 inference code + weights
Mar 2025
Turbo, mini, multiview, and FlashVDM released
Jun 2025
Hunyuan3D-2.1 fully open-sourced (PBR + VAE + training code)
Jun 2025
Hunyuan3D 2.5 system technical report
Jul 2025
HunyuanWorld-1.0 — immersive 3D world generation

Open-source companions

FlashVDM

Inference accelerator that preserves mesh quality

Blender Addon

Run Hunyuan3D 2.0 inside Blender (Jan 2025)

Hunyuan3D-Studio

Production platform for mesh manipulation and animation

From idea to downloadable asset

Four steps — the same path whether you use Pro for quality or Rapid for speed.

Choose input

Text, single image, up to 4 multi-view photos, or a sketch with attribute description.

Generate

DiT reconstructs geometry; Paint bakes PBR textures. Pro for control, Rapid for speed.

Preview

Inspect in the interactive 3D viewer. Rotate, zoom, check every angle.

Export

Download GLB or OBJ — ready for Blender, Unity, Unreal, or 3D printing.

Generation modes

Pro exposes four modes; Rapid runs a streamlined subset.

Normal

Standard textured mesh — the default for most assets.

LowPoly

Game-ready low-polygon mesh with tri or quad topology, 40K–2M faces.

Sketch

Line-drawing input with color, category, and material attributes. Pro only.

Who builds on Hunyuan3D

Over 150 enterprises adopted Hunyuan3D via Tencent Cloud within the first year.

Unity China

Real-time 3D engine integration

Bambu Lab

Consumer 3D printing pipelines

Liblib

AI content creation platform

Industries

Game developmentE-commerce & product vizFilm & VFXAR / VR / spatialArchitecture3D printing

Hunyuan3D 3.5 Pro vs Rapid vs 3.0

Two 3.5 editions plus the prior 3.0 baseline. Pick the right fit for quality, speed, and budget.

Feature	Hunyuan3D 3.5 Pro	Hunyuan3D 3.5 Rapid	Hunyuan3D 3.0
Edition Focus	Max fidelity	Speed-first	Baseline
Generation Time	1–3 min	<60s	1–2 min
Max Texture	8192² PBR	8192² PBR (+20 cr)	4096² PBR
Polygon Range	40K – 2M	40K – 2M	up to 1.5M
Multi-View Input	Yes (up to 4)	No	Yes (basic)
Sketch Mode	Yes	No	No
Best For	Production assets	High-volume / prototyping	General use
Credits	60 (max 100)	35 (max 55)	12

Frequently asked questions

Try Hunyuan3D 3.5 now

Generate your first 3D model in under a minute. Free credits for new users.

Image to 3D See pricing

Hunyuan3D 3.5: open-source text & image to 3D generation

Pro & Rapid editions·8192² PBR textures·Open-source lineage

Input

Pro edition

Rapid edition

Text

Up to 1024 characters of natural language

Full support

Image

Single reference photo or concept art

Full support

Multi-View

Up to 4 angle photos for precise geometry

Full support

Not available

Sketch

Line drawing + attribute description

Full support

Not available

Feature

Hunyuan3D 3.5 Pro

Hunyuan3D 3.5 Rapid

Hunyuan3D 3.0

Edition Focus

Max fidelity

Speed-first

Baseline

Generation Time

1–3 min

<60s

1–2 min

Max Texture

8192² PBR

8192² PBR (+20 cr)

4096² PBR

Polygon Range

40K – 2M

up to 1.5M

Multi-View Input

Yes (up to 4)

Yes (basic)

Sketch Mode

Yes

Best For

Production assets

High-volume / prototyping

General use

Credits

60 (max 100)

35 (max 55)