Home Creators Posts Import Register Favorites Logout

Downloads

Content

Hi3DGen turns a single image into a detailed, UV-unwrapped 3D mesh on your own PC; on an RTX 5090 our dwarf example became a 344,622-face mesh in 1 minute.

SOTA, ultra advanced app that is 1-click to install on Windows, RunPod and Massed Compute, with single image, multi-view and batch processing (2.4 times faster in V19) and auto-save as OBJ, GLB, PLY and STL. Delivers better meshes than Trellis, Hunyuan3D-2.0 and TripoSG.

1-Click to install newest state of the art image to 3D mesh generator Hi3DGen with amazing ultra advanced app on Windows, RunPod, Massed Com

Please visit Patreon exclusive posts index to find all of our Patreon scripts, posts, apps and more.

Join discord to get help, chat, discuss and also tell me your discord username to get your special rank : SECourses Discord

Please also Star, Watch and Fork our Stable Diffusion & Generative AI  GitHub repository and join our Reddit subreddit and follow me on LinkedIn (my real profile)

=======

Latest zip file : Hi3DGen_v19.zip

Last Updated 19 October 2026

Step by step tutorial of Hi3DGen : https://www.youtube.com/watch?v=HjbD20B2C1g

Official Project Page : https://stable-x.github.io/Hi3DGen/

Windows Requirements

Hi3DGen SECourses App turns a single image into a detailed 3D mesh on your own PC. Hi3DGen first predicts a sharp normal map of your image and then uses it as a bridge to the 3D shape, so the fine details of the image end up in the mesh. Our app adds 1-click installers, single image, multi-view and batch generation, automatic saving in four 3D formats, presets and a live status for every stage. Below you will see every feature with real screenshots from the app.

Why Hi3DGen SECourses App

  • Detailed 3D meshes from one image: Hi3DGen's normal bridging keeps sharp edges and fine surface details. On an RTX 5090, our dwarf example became a UV-unwrapped mesh of 344,622 faces in 1 minute.

  • Three ways to generate: a single image, 2 to 8 views of one object, or a whole folder in one batch.

  • Batches 2.4 times faster: V19 works on the next image on the GPU while earlier meshes are unwrapped on the CPU. Four images took 1 minute 42 seconds in one batch, against 4 minutes 8 seconds one at a time.

  • Models stay on the GPU: on cards with 20 GB or more, the models stay loaded between generations, so the normal map of the next image takes under a second.

  • Live status and early previews: the stage, progress, seed and time of every stage. The normal map appears right away, and the mesh is shown while its UVs are still being computed.

  • Cancel at any time: a single image, several views or a whole batch, even in the middle of UV unwrapping.

  • Auto-save in four formats: OBJ, GLB, PLY and STL with the normal map, every run in its own numbered folder.

  • Ready for your 3D apps: UV-unwrapped meshes, with a polygon count slider from 5 to 100 %.

  • Presets: Default and High Quality, or save your own.

  • 1-click installers: Windows, RunPod and Massed Compute, with Python 3.12, PyTorch 2.13, CUDA 13, precompiled spconv and xFormers for Windows, and automatic model downloads that resume.

  • Works offline: after the install, every model loads from the app folder.

  • Frequent updates: V19 came out on 18 October 2026 with a new interface and faster processing.

Here is the app right after a generation. The dwarf below became a detailed, UV-unwrapped 3D mesh in 1 minute on an RTX 5090.

1: Hi3DGen predicts a normal map of your image first, the bridge to a detailed shape. 2: the status card shows the time of every stage. 3: rotate and zoom the mesh, then download it. 4: every run is saved as OBJ, GLB, PLY and STL automatically.

1-Click Installation

What You Download

You get one small zip file with the installers for every platform.

Extract it into a short folder path without spaces, double-click Windows_Install.bat, then start the app with Windows_Start.bat. Windows_Update.bat pulls the latest app code at any time.

Latest PyTorch, CUDA 13 and Precompiled Libraries

The installer makes its own Python 3.12 virtual environment, so your other apps stay untouched. It installs PyTorch 2.13 with CUDA 13 and every other package with uv, a very fast package installer.

spconv for the sparse 3D convolutions and xFormers for attention come precompiled for Windows, so nothing has to be compiled on your PC, and the final check found no broken requirements. Our fresh install took 5 minutes 13 seconds on our PC, models included.

Automatic Model Downloads

At the end, the installer downloads every model by itself, about 7 GB in all.

It fetches Hi3DGen, the StableNormal normal map model, BiRefNet for background removal and DINOv2 from pinned versions, then verifies every model. If a download stops, run Windows_Resume_Model_Download.bat and it continues where it stopped. After that, the app never downloads anything at start.

Start the App

Double-click Windows_Start.bat. The app checks every model, loads the pipeline and opens in your browser.

The very first start prepares the new files and takes about two minutes. After that, the app was ready 20 seconds after the double-click on our PC, with the models kept on the RTX 5090 between generations.

Cloud GPUs: RunPod and Massed Compute

You can also run the app on a cloud GPU. The cloud installers set up everything with one command and download the models for you, and the app starts with a Gradio share link you can open on any device.

The step-by-step commands are in the instruction files inside the zip.

Requirements

On Windows you need Python 3.12, Git, FFmpeg, CUDA 13, cuDNN and Visual Studio with the C++ tools. This tutorial shows every step, and the same setup runs all our AI apps.

Single Image to 3D Mesh

Drop an image and click Generate 3D Mesh. The app removes the background, predicts the normal map, generates the 3D shape, simplifies it, unwraps the UVs and saves the files.

Where to find it: the Single Image tab, open when the app starts.

Here is a real run of the mech example, caught during UV unwrapping:

1: the normal map appears as soon as it is ready. 2: the status card shows the stage, the elapsed time and the seed. 3: the mesh is already in the viewer while xatlas computes its UVs, the longest step. Cancel stops the run at any step, and the previous mesh comes back in the viewer.

Download and Auto-Save

When the run finishes, pick OBJ, GLB, PLY or STL in File Format, and the Download button converts the mesh to it.

Display Mode switches the viewer between solid, wireframe and point cloud. Every run is also saved automatically in its own numbered folder:

Open Outputs Folder opens it in one click, and the four Auto-save boxes choose the formats. The OBJ and GLB files keep their UV coordinates, so you can paint or bake textures on them in Blender or any other 3D app.

Multiple Images of One Object

Give Hi3DGen 2 to 8 views of the same object, and it uses all of them for one mesh.

Where to find it: click the Multiple Images tab.

Here is a real run with three views of one figure:

1: upload the views, and the previews and the first view without its background appear right away. 2: MultiDiffusion uses every view and is the recommended mode; Stochastic cycles through the views. 3: three views became one mesh in 52 seconds. Seven ready multi-view example sets come with the app.

Batch Processing Whole Folders

Point the app at a folder of images, and it makes a 3D mesh of every image, each in its own subfolder.

Where to find it: click the Batch Processing tab.

Here is a batch of four example images, caught while three of them are in progress at once:

V19 works on the next image on the GPU while up to three earlier meshes are simplified and unwrapped in separate processes. Four images took 1 minute 42 seconds in one batch, against 4 minutes 8 seconds one at a time: 2.4 times faster, with the same meshes. Skip Existing Files skips the images that already have their files, so an interrupted batch continues where it stopped, and the summary lists every skipped and failed image.

Advanced Settings

Both generation stages, the mesh simplification, the UV unwrapping and the normal map each have their own sliders, and single images, multiple images and batches all use them.

Where to find it: Advanced Settings is under the tabs and open when the app starts.

Here are the settings of the Default preset:

A fixed seed reproduces a result, -1 picks a new seed for every run, and the status card shows the seed used. Polygon Count Percentage keeps 5 to 100 % of the faces, and the UV atlas goes up to 4096 px. The Parameter Guide tab explains every setting with its range and tips.

Presets

A preset stores every setting: Default for balanced speed and quality, High Quality for more detail.

Where to find it: Preset Management is in the right column, under the viewer.

Pick a preset, and it loads at once:

High Quality raises the guidance of both stages to 4, the second stage to 8 steps, the polygon count to 80 %, the UV atlas to 2048 px and the normal map to 1024 px. Type a name and click Save Preset to keep your own, and the app opens with your last used preset next time.

Performance: Models on the GPU

Choose where the models live and how many meshes a batch unwraps at once.

Where to find it: click Performance, under Advanced Settings.

Here is the Performance section on an RTX 5090:

Auto keeps the models on the GPU between generations when the card has 20 GB or more, so the next image starts without loading anything, and moves them to the CPU after each stage on smaller cards. If another program fills the GPU and memory runs out, auto switches to moving them by itself. Keep and Offload force either way, Free GPU memory clears the GPU in one click, and the batch slider sets how many meshes are unwrapped at the same time.

Light and Dark Theme

The app starts in dark mode, and one click switches to the light theme.

The switch is instant, even while a generation runs, and the app remembers your choice.

Latest Updates

V19 came out on 18 October 2026:

  • Python 3.12, CUDA 13, PyTorch 2.13 and Gradio 6.30, installed from one requirements file.

  • A new interface with colored buttons, a light and dark theme and a live status card with the time of every stage.

  • The normal map appears right away, and the mesh is shown while its UVs are computed.

  • Cancel works for single images, several views and batches, even during UV unwrapping.

  • The models stay on the GPU between generations on cards with 20 GB or more, and StableNormal loads once instead of for every image.

  • Batches are two to three times faster: the next image runs on the GPU while earlier meshes are unwrapped.

  • StableNormal and BiRefNet ship with the app, and every model loads from the app folder with no download at start.

  • The results match V17: identical background masks and meshes within 0.2 % of the model size, in the developer's checks.

Get Hi3DGen SECourses App

Download the latest zip file attached to this post, extract it and run the installer for your platform. To update, get the newest zip, overwrite the old files and run Windows_Install.bat again.

Related posts

All posts in this topic: Image to 3D Generation Apps · New to SECourses? Start Here: Best and Most Popular SECourses Posts

Files

Hi3DGen Full Tutorial With Ultra Advanced App to Generate the Very Best 3D Meshes from Static Images

Hi3DGen is the newest state of the art image to 3D mesh generation model. In this tutorial I will show you step by step how to install and use this amazing open source AI model to generate the very best 3D meshes from static images and use in your projects. 🔗Follow below link to download the zip file that contains App installer - the one used in the tutorial ⤵️ ▶️ https://www.patreon.com/posts/The-App-Installer-130766890 🔗 Requirements - Python, Git, CUDA, C++, FFMPEG, MSVC installation tutorial ⤵️ ▶️ https://youtu.be/DrhUHnYfwC0 🔗 SECourses Official Discord 10500+ Members ⤵️ ▶️ https://discord.com/servers/software-engineering-courses-secourses-772774097734074388 🔗 Stable Diffusion, FLUX, Generative AI Tutorials and Resources GitHub ⤵️ ▶️ https://github.com/FurkanGozukara/Stable-Diffusion 🔗 SECourses Official Reddit - Stay Subscribed To Learn All The News and More ⤵️ ▶️ https://www.reddit.com/r/SECourses/ 🔗 SUPIR Image Upscaling App Tutorial for Upscaling Images With Really Adding New Details ⤵️ ▶️ https://youtu.be/OYxVEvDf284 DetailVerse: The Fuel for High-Fidelity High-quality models require high-quality data. The researchers found that existing 3D datasets like Objaverse, while large, are dominated by objects with simple geometry and plain surfaces. Training on such data inherently limits a model's ability to generate complex details. To solve this, they built DetailVerse, a massive synthetic dataset of 700,000 high-quality 3D assets specifically designed to be rich in geometric complexity. The creation pipeline is a feat of engineering in itself: Text Prompt Curation: Starting with 14 million text prompts, they used LLMs to filter, classify, and standardize them to describe diverse, single-objects with clear geometric attributes. High-Quality Image Generation: The state-of-the-art text-to-image model Flux.1-Dev was used to generate images from these prompts, with specific instructions for isometric views and high detail. Robust 3D Synthesis: The powerful image-to-3D model Trellis was used to convert these images into 3D meshes. Rigorous Quality Control: The resulting 3D assets underwent a strict filtering process, using both automated checks (a DINOv2-based quality classifier) and human expert evaluation to ensure only the highest-quality, artifact-free models made it into the final DetailVerse dataset. Results: Setting a New State of the Art Hi3DGen's performance was rigorously evaluated, and the results speak for themselves. Qualitative Superiority: As seen in numerous visual comparisons, Hi3DGen consistently generates 3D models with superior detail and fidelity to the input image. It excels at capturing complex textures, sharp edges, and subtle surface variations that other models miss entirely. Quantitative Dominance: In quantitative benchmarks for normal estimation, NiRNE significantly outperforms both regression-based and diffusion-based methods on key metrics like Normal Error (NE) and Sharp Normal Error (SNE). User-Approved: In a user study involving both amateur 3D users and professional 3D artists, Hi3DGen's results were overwhelmingly preferred over five other leading models, including Hunyuan3D-2.0, Dora, Clay, Tripo-2.5, and Trellis. Hi3DGen represents a significant leap forward in single-image 3D generation. By rejecting the direct 2D-to-3D approach and instead adopting a "normal bridging" strategy, the framework effectively decomposes a complex problem into solvable parts. The combination of a specialized normal estimator (NiRNE), a regularized geometry generator (NoRLD), and a purpose-built, detail-rich dataset (DetailVerse) creates a synergistic system that achieves unprecedented fidelity. Video Chapters 0:00 Introducing Hi3DGen: State-of-the-Art Image-to-3D Model 0:34 Advanced Gradio App Features & One-Click Installation 1:17 In-Depth App UI Features: Presets & 3D Result Analysis 2:14 How to Download & Install Hi3DGen App on Windows 2:55 Starting the Fully Automatic Windows Installation Process 3:38 Pro Tip: Upscale Input Images with SUPIR for Best Quality 4:25 App Parameter Guide and Batch Processing Folder Setup 5:12 How to Install and Use on Cloud GPUs (RunPod & Massed Compute) 5:46 Installation Complete & How to Resume Failed Model Downloads 6:11 Starting The App & Running Your First 3D Generation 6:54 Understanding the Generation Process & The Slow xatlas Step 8:04 Analyzing The First Generated 3D Mesh in The UI 8:26 Locating & Checking The Automatically Saved 3D Files (OBJ, GLB, STL) 8:54 How to Set Up & Start Batch Processing Multiple Images 9:39 Monitoring Batch Progress & Using The Cancel Feature 10:15 Detailed Step-by-Step Guide for RunPod Installation 10:55 Detailed Step-by-Step Guide for Massed Compute Installation 11:41 Teaser: Ultimate Video Upscaling App Coming Soon Some background music by NoCopyrightSounds : https://gist.github.com/FurkanGozukara/681667e5d7051b073f2e795794c46170

Comments

Markus

Is the output model solid or hollow?

SECourses

this generates 3d mesh only but the best one right now among open source models

SECourses

i didnt try yet. let me know once they published 1.2

Markus

Yeah, but only a mesh, right? It's not solid?

Darryl Walk

C:\Hi3DGen_v15>cd Hi3DGen C:\Hi3DGen_v15\Hi3DGen>call .\venv\Scripts\activate.bat Traceback (most recent call last): File "C:\Hi3DGen_v15\Hi3DGen\secourses_app.py", line 1, in import gradio as gr ^^^^^^^^^^^^^^^^^^^ ModuleNotFoundError: No module named 'gradio' Error: Python script failed. Press any key to continue . . .

Manuel Leiton

Uninstall Python. Clean environment variables. Reinstall python 3.10.11 solved me.

SECourses

wow nice ty will cover this too hopefully

SECourses

install error. make a fresh install and send me logs ( email me monstermmorpg@gmail.com ) so i can find your error. did you follow requirements? https://youtu.be/DrhUHnYfwC0

SECourses

install error. make a fresh install and send me logs ( email me monstermmorpg@gmail.com ) so i can find your error. did you follow requirements? https://youtu.be/DrhUHnYfwC0

SECourses

it is mesh. i dont know 3d world good so i dont know what you mean by solid :D

SECourses

yes we need python 3.10.11 true glad you solved

Darryl Walk

any hints or solutions? I do have gradio hosting Hunyuan though, but I thought is self contained and isolated from that?

SECourses

I can help if you just delete venv and run installer and send me logs so i can see install error. but 99.99% related to following requirements tutorial : https://youtu.be/DrhUHnYfwC0

truzter

I start the tool but the link page localhost is empty any ideas?

SECourses

yes stupid gradio broken and still didnt fix. fixed with v16 please run install again

Martí Garcia

For what you've tried, which is the best configuration to get the most similar result as the input image? So Imagine I need a 3D model that is exact as the picture I have, you know, having the most consistency possible. I know this tools never get the shape 100% as the input, but what's your reccomendation for achieving the best accurate model from the original picture? Let's say we talk about a statue, where anatomy, proportions, shape and details are really important, Thanks!

SECourses

sadly i dont have any configuration for this model. but hopefully i will make an app for better model soon

Josh S

Do you have a updated installer and guide for 2.1?

SECourses

Hi what do you mean by 2.1? our installer installs latest improved version and 1 click to install already

Marco PG

I think he's talking about HUNYUAN3D 2.1?

100Sachen80

Hi Furkan, does the installer need an update? When installing: Installing collected packages: huggingface_hub Attempting uninstall: huggingface_hub Found existing installation: huggingface-hub 0.36.0 Uninstalling huggingface-hub-0.36.0: Successfully uninstalled huggingface-hub-0.36.0 ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. tokenizers 0.20.3 requires huggingface-hub<1.0,>=0.16.4, but you have huggingface-hub 1.1.4 which is incompatible. transformers 4.46.3 requires huggingface-hub<1.0,>=0.23.2, but you have huggingface-hub 1.1.4 which is incompatible. When starting: File "G:\Hi3DGen\Hi3DGen\hi3dgen\pipelines\hi3dgen.py", line 96, in _init_image_cond_model dinov2_model = torch.hub.load(os.path.join(torch.hub.get_dir(), 'facebookresearch_dinov2_main'), name, source='local',pretrained=True) NameError: name 'os' is not defined During handling of the above exception, another exception occurred: Traceback (most recent call last): File "G:\Hi3DGen\Hi3DGen\secourses_app.py", line 2236, in hi3dgen_pipeline =Hi3DGenPipeline .from_pretrained (trellis_local_path ) File "G:\Hi3DGen\Hi3DGen\hi3dgen\pipelines\hi3dgen.py", line 87, in from_pretrained new_pipeline._init_image_cond_model(args['image_cond_model']) File "G:\Hi3DGen\Hi3DGen\hi3dgen\pipelines\hi3dgen.py", line 98, in _init_image_cond_model dinov2_model = torch.hub.load('facebookresearch/dinov2', name, pretrained=True) File "G:\Hi3DGen\Hi3DGen\venv\lib\site-packages\torch\hub.py", line 647, in load model = _load_local(repo_or_dir, model, *args, **kwargs) File "G:\Hi3DGen\Hi3DGen\venv\lib\site-packages\torch\hub.py", line 677, in _load_local hub_module = _import_module(MODULE_HUBCONF, hubconf_path) File "G:\Hi3DGen\Hi3DGen\venv\lib\site-packages\torch\hub.py", line 115, in _import_module spec.loader.exec_module(module) File "", line 879, in exec_module File "", line 1016, in get_code File "", line 1073, in get_data FileNotFoundError: [Errno 2] No such file or directory: 'C:\\Users\\user/.cache\\torch\\hub\\facebookresearch_dinov2_main\\hubconf.py'

SECourses

thanks i did a big update with v17 please make a fresh install

Bartłomiej Golenia

Hi! Works great when using Single Image But i do have a problem when using Multiple Images: Processing 2 images in multidiffusion mode... Multi-image generation error: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same Am i doing something wrong? Is there a tutorial how to use it?

Bartłomiej Golenia

I restarted the batch file and i started generating! But Time and time again I have another issue: Multi-image generation error: local variable 'first_preprocessed' referenced before assignment I got it - this happends when you de-select all auto-saves. If at least one is selected then it works :)

SECourses

did you make a fresh install. i cant reproduce the error. i need to reproduce to be able to fix. can you make a quick video of how you testing? also make a fresh install and send logs to monstermmorpg@gmail.com

Chương

Tôi xin phép nói tiếng Việt Nam. Tui có mua tài khoản thành viên của tác giả, tôi đã tải những file tác giả tạo, mà tác giả làm file nào cũng không cài đặt và sử dụng được. Tôi không hiểu nỗi ông làm cho đa số thành viên tải về file nào cũng không cài đặt và sử dụng được thì ông nên nghỉ làm. Vì thực tế nếu 1 người tác giả giỏi thì sẽ tạo làm sao cho thành viên dễ cài đặt và sử dụng được. Tôi không vui vì mua thành viên xong cài đặt và sử dụng không thành công tất cả những bài viết của tác giả.