Name: Image to 3D
Brand: Milad_Kambari
Price: 18.00 USD
Availability: InStock
Rating: 4 (18 reviews)

Image to 3D

Milad_Kambari in Modeling

BLENDER ADDON · AI 3D GENERATION

Drop in a picture or write a sentence and get a real, textured 3D model — generated entirely on your own graphics card, right inside Blender. No cloud. No credits. Nothing to upload.

4 AI Engines for 3D Models

2 for Textures

5 for Canvas

Runs 100% Locally

No Cloud · No Credits

VERSION 2

Image to 3D 2

Turn a picture or a sentence into a textured 3D model inside Blender. Everything runs on your own NVIDIA graphics card, with no monthly fees and no internet needed after setup.

WATCH IMAGE TO 3D 2

NEW IN VERSION 2

Canvas

A node board where you wire nodes together — from a sentence or a picture all the way to a finished, enhanced 3D model.

	CANVAS · 01 A Node Board Inside Blender Canvas is new in version 2. Wire nodes together on one board, from a sentence or a picture all the way to a textured 3D model. Every step stays visible, so you can see and change each one.
CANVAS · 02 Multi View Give it one picture and Multi View makes the back, left and right views for you. Hunyuan can use these extra views to build the model from every side.
	CANVAS · 03 Upscale Bring in a picture with the Upload or Image node and run it through Upscale to get a bigger, sharper picture before you turn it into 3D.
CANVAS · 04 From an Idea to a Game Asset Write a Prompt with the 3D Ready preset for a clean studio background, then Generate 3D, Enhance Detail and Enhance Texture. Auto Run runs the whole chain in one go, and every model is placed side by side in your scene.

Canvas Nodes

●	Prompt: a picture from text, with a 3D Ready preset for a clean studio background.
●	Upload, Image and Upscale.
●	Multi View: back, left and right views from one picture.
●	Find Parts: finds the separate parts of a model and makes a picture of each one.
●	Generate 3D, Enhance Detail and Enhance Texture.
●	Auto Run: runs the whole chain in one go.
●	Every model you make is placed side by side in the Blender scene.

Everything in Image to 3D 2

Canvas comes on top of everything you already know from Image to 3D.

Engines & Enhance

●	Three 3D engines: TRELLIS, Pixal3D and Hunyuan. Hunyuan can also take back, left and right photos.
●	Main settings: Quality, Faces up to 2,000,000, and Texture on or off.
●	Enhance Detail: uses Hi3D to add more surface detail to a model’s shape.
●	Enhance Texture: uses FLUX.2 Klein to make the texture sharper and more realistic.
●	Both Enhance tools work on models from all three engines.

Downloads, Updates & Support

●	Downloads: each model downloads on its own, straight from its maker’s official page, and every file is checked against the official one.
●	Update: checks Superhive for a new version. Auto Update is optional and off by default.
●	Save Report: saves a report file for support.
●	Minimum system: NVIDIA GPU with 8 GB VRAM, and 16 GB RAM.
●	Licences: every Canvas model allows commercial use. The details are in THIRD_PARTY_LICENSES.md.

WATCH THE TEASER

WATCH THE REVIEW

01	ONE IMAGE → ONE MODEL

Drag, Drop, Generate

Drag a picture onto the card and press Generate 3D — a few minutes later you have a real mesh with a proper texture sitting in your scene. You don’t even need to remove the background: the addon does it for you.

Made with Image to 3D

Real results, straight out of the add-on.

Made on an RTX 4070 Ti View post on X →	Hard-surface sample View post on X →
4K texture · 500K faces View post on X →	4K texture · 500K faces View post on X →
Image to 3D + Smart Remesh · 1.5M to 5K faces View post on X →	1.5M faces View post on X →
GOD OF DETAILS · Image to 3D + Smart Remesh View post on X →	1.5M faces View post on X →

4 AI Engines for 3D Models

Three generation engines to pick from — plus Hi3D, the fourth engine, powering Enhance Detail.

TRELLIS

Fast, with lots of fine detail — the default engine and the quickest way from picture to mesh.

Pixal3D

Stays really close to the exact view in your picture — what you see is what you get.

Hunyuan

The cleanest shapes and the best textures — and the only engine that accepts extra views of your object.

11 AI Models, All Running on Your GPU

Each one is optional and downloads only when you need it.

AI Engines for 3D Models

TRELLIS

Pixal3D

Hunyuan

Hi3D · Enhance Detail

For Textures

FLUX.2 Klein · Enhance Texture

Marigold · Enhance Texture

For Canvas

Qwen-Image-Edit · Multi View, Prompt, Find Parts

Qwen-Image · Prompt

SAM 3 · Find Parts

BiRefNet · 3D Ready

Real-ESRGAN · Upscale

Canvas also uses FLUX.2 Klein for part pictures.

02	REAL GENERATION QUALITY

Detail That Holds Up Close

A single photo becomes a full 3D head with clean topology and a painted texture — generated locally, at the face count you choose, with 1K, 2K or 4K textures.

03	MORE VIEWS · HUNYUAN

Show It More Angles, Get a Better Model

With the Hunyuan engine you can add extra photos of the same object — the big slot is the front, and labelled slots for Left, Right and Back show exactly which picture goes where. More views, more accurate shape and texture. (TRELLIS and Pixal3D work from a single picture.)

04	ENHANCE DETAIL (HI3D) & ENHANCE TEXTURE

Upgrade a Model You Already Made

Two one-click upgrades for any generated model: Enhance Detail runs the Hi3D engine directly on your existing model — adding fine surface detail and repainting the texture without regenerating the shape from scratch — while Enhance Texture sharpens and enriches the texture only.

How Long Does It Take?

Same networks, same resolution, same result on every system — your hardware only changes the time, never the quality. All times are minutes:seconds, measured with a 4K texture and 200,000 faces.

	TRELLIS		PIXAL3D		HUNYUAN		ENHANCE (ON AN EXISTING MODEL)
SYSTEM	FAST	HIGH	FAST	HIGH	FAST	HIGH	ENHANCE DETAIL	ENHANCE TEXTURE
MINIMUM RECOMMENDED 8 GB VRAM · 16 GB RAM RTX 4060 class · estimated	~2:25	~4:45	~5:25	~4:50	~4:10	~5:50	~9:00	~3:40
MY SYSTEM 12 GB VRAM · 64 GB RAM RTX 4070 Ti · measured	1:21	2:32	2:48	2:30	2:18	3:12	4:32	2:09

PLEASE NOTE

These times are approximate and will vary with your hardware — they can be faster or slower on your system. The "My System" row is my own experience, measured on my own machine. The "Recommended" row is an estimate for an RTX 4060-class 8 GB card, based on my tests simulating a limited system (my card capped to 8 GB VRAM and 16 GB RAM): the steps that run on the graphics card take a little over twice as long as on an RTX 4070 Ti, while the steps that run on the processor take about the same. What never changes is the result — every system produces exactly the same model and texture, only the waiting time differs.

Download Only What You Use

The AI models are big files — so the addon has a built-in Downloads tab where every engine has its own row with its name and size. You only download the engines you actually want to use, into any folder on any drive you choose.

One-Click Download

Press Download and the addon fetches every file straight from the official model pages — with a progress bar, speed and time left. Stop any time; it resumes exactly where it left off.

Or Download By Hand

Prefer your own download manager? By Hand shows every part a model needs, opens its official page and the exact folder it goes into, then Check Files verifies every file against the official fingerprint.

Never Get Lost

Pick an engine you haven’t downloaded yet and the addon tells you — one press on Go to Downloads takes you straight to the right row. Broken files are detected and re-downloaded.

Image to 3D Models

Each engine is a separate download — get only the ones you need.

DOWNLOADS ROW	USED FOR	SIZE
TRELLIS	Main engine	16.8 GB
Pixal3D	Pixal3D engine	25.2 GB
Hunyuan	Hunyuan engine	18.6 GB
Hunyuan More Views	Several photos with Hunyuan (needs Hunyuan)	4.6 GB
Enhance Detail	Hi3D (needs Hunyuan)	10.1 GB
Enhance Texture	FLUX.2 Klein + Marigold	17.3 GB

Canvas Models

Downloads > Canvas. Each Canvas node needs only its own models.

DOWNLOADS ROW	USED FOR	SIZE
Qwen-Image-Edit (text model)	Multi View, Prompt, Find Parts	15.7 GB
Qwen-Image-Edit 4-bit	Multi View, Prompt	11.1 GB
Qwen-Image 4-bit	Prompt	11.1 GB
SAM 3	Find Parts	3.2 GB
FLUX.2 Klein	Part pictures	14.9 GB
BiRefNet	Prompt with 3D Ready	0.4 GB
Real-ESRGAN	Upscale	0.1 GB

NOTES

●	You don’t need everything. One engine is enough to start — for example TRELLIS alone is 16.8 GB. Each Canvas node needs only its own models.
●	Some models are shared and download only once: Klein (Enhance Texture and Canvas), BiRefNet (TRELLIS, Pixal3D and Canvas), DINOv3 (TRELLIS and Pixal3D), Real-ESRGAN (Hunyuan and Canvas).
●	Everything together, without the shared repeats, is about 132 GB.
●	These sizes are only the model files. Each engine also has a one-time software install (Python and its libraries).
●	Qwen comes in two builds: FP4 for RTX 50 cards and INT4 for older cards. The add-on picks the right one by itself, so on older cards each Qwen 4-bit download is a bit smaller (10.7 GB instead of 11.1 GB).

ENGINES & LICENCES

Image to 3D is a platform for running 3D engines. It does not include or distribute any AI model or third-party engine. Every engine is optional and is downloaded by you, only when you choose it, directly from its makers’ official pages. You can see which engines are available in the Downloads section on this page.

Each engine has its own licence, so please review the licence of any engine before you use it. All licences are listed in THIRD_PARTY_LICENSES.md inside the add-on.

Everything Inside

Three Quality Levels Fast for quick tests, High for most work, Ultra for maximum detail (Ultra needs a 12 GB card — on 8 GB it automatically uses High).	Your Face Count Set exactly how detailed the mesh should be, from 5,000 up to 2,000,000 faces. 200,000 is a great start.	Texture Control Texture on or off (shape-only is much faster), at 1K, 2K or 4K — and a Seed to get a different version of the same picture.
Blender Never Freezes The engines run as separate background processes — you keep working while the model generates, with live progress on the card.	Clean Card UI Everything lives on one clean card in the viewport, in the same style as Auto Unwrap and Modeling Toolkit — only the settings that actually matter.	Private by Design Your images never leave your computer — everything runs on your own GPU, offline after the models are downloaded.

Recommended System

8 GB

VRAM · NVIDIA GPU

16 GB

RAM

WINDOWS · LINUX

OPERATING SYSTEM

IMPORTANT · NVIDIA GPU REQUIRED

Image to 3D runs only on NVIDIA graphics cards. If your computer has an AMD graphics card (GPU), the addon will not work.

An AMD processor (CPU) is completely fine — my own system runs on an AMD CPU. The limitation is only about AMD graphics cards.

Release Notes

Every update is free for everyone who owns Image to 3D.

v2.0.6

LATEST

●	Fixed the engine install stopping with a certificate error (“UnknownIssuer”) on some PCs and networks, like Shadow cloud PCs: if the installer can’t check a site’s certificate, packages are installed with pip, which uses the system’s own check.

v2.0.5

●	Fixed the OK / Export button of file windows (File > Export, Choose Image) not working while the add-on’s tab is open.
●	Canvas Prompt and Multi View now say clearly when the graphics card is too old (RTX 20 / GTX 16) instead of failing.
●	3D Ready now tells you when the main 3D engine needs to be installed, instead of showing a Python error.

v2.0.4

●	Fixed the install stopping at 3% on some PCs with a certificate error: if the installer can’t download Python itself, Blender downloads it instead.

Showing the latest updates. Image to 3D has had regular free updates since version 1.0.