March 12, 2025
5 min read

How to Create 3D Models Using AI: Step-by-Step Guide with TRELLIS

Written by [object Object]

By Kevin Kern

Creating 3D models traditionally involves complex tools and significant manual effort. AI solutions like TRELLIS from Microsoft Research make the process significantly easier, letting you transform simple images into functional 3D models. In this post, we’ll walk through each step - complete with examples, practical tips, and video demonstrations. Whether you’re a seasoned 3D artist or a curious beginner, you’ll find insights here on how to bring AI-driven modeling into your own workflow.

1. Example: Meet "Turion" - AI-generated 3D Model

To give you a clear picture of the entire process, i created “Turion,” a fully AI-generated 3D model. Turion demonstrates what’s possible when cutting-edge image-to-3D technology meets creative exploration. By studying how Turion was formed, you’ll get a firsthand look at how straightforward it can be to transform an idea or image into a tangible 3D asset.

2. Generate or Find Images for Your 3D Scene

Your journey begins with the images you want to convert into 3D. You can use high-resolution photos, art illustrations, or even entirely AI-generated images. The key is to choose pictures that clearly depict the object or scene you want to model. Crisp, well-lit images help TRELLIS identify important details and produce more accurate 3D shapes.

Whether you decide to create fresh images with AI or use existing ones from your library, this step lays the foundation for a successful 3D conversion.

3. Clean the Image and Remove the Background (Optional)

Although it’s optional, removing the background can greatly improve clarity. For instance, if your image has a busy background or extra elements, TRELLIS might capture unwanted geometry. By isolating the subject - whether it’s a character, product, or decorative item - you guide the AI to focus on what really matters. This often results in a cleaner, more accurate final model.

4. Convert Image to 3D Using TRELLIS – Where It Gets Interesting

Now for the core step: using TRELLIS to convert your prepared image into 3D. TRELLIS is a powerful asset generator developed by Microsoft Research. Once you feed in the cleaned or original image, the AI identifies shapes and textures, then translates those details into a workable 3D model. This step might feel a bit magical, but it’s underpinned by advanced machine learning algorithms that have been trained on diverse datasets.

5. How to Access TRELLIS for Image-to-3D Generation

TRELLIS can be accessed in several ways, making it flexible for different needs and technical skill levels:

  • Self-hosted: Install TRELLIS locally on your machine (make sure your hardware meets the requirements). This gives you complete control but requires more setup.

  • Online demo: Use the TRELLIS demo on Hugging Face to experiment quickly without worrying about local hardware or installation.

  • Third-party services: Platforms like Fal.ai or Replicate host TRELLIS for you, letting you focus on creating 3D models without the hassle of infrastructure management.

Choose the route that aligns best with your resources and project scope.

6. Clean Your 3D Model in Blender (Optional)

Even if your initial AI-generated model looks good, it might still have some imperfections. For example, you may see small artifacts, holes in the mesh, or overly dense geometry. Using a 3D tool like Blender can help you clean up and refine the model:

  • Remove any unwanted geometry.

  • Smooth or simplify the mesh to optimize it for real-time use.

  • Adjust the scale, rotation, or positioning of the model as needed.

This step ensures your 3D model is polished and ready for integration into games, VR/AR applications, or other design projects.

7. Prompt Example for AI 3D Scene Creation

If you’re new to writing prompts for AI-generated scenes, you might wonder how to phrase your request. Below is a sample prompt you can adapt to your own needs when testing AI 3D modeling tools. Remember to be as specific as possible about shapes, textures, and desired style to get better results.

AI-generated 3D Models: Current Limitations and Future Potential

Like other AI-generated content - such as images or code - 3D models still require some human oversight. You may need to invest time in post-processing to fix minor errors or enhance details. Additionally, if you’re using a remote service, you might encounter usage fees or limitations.

It’s worth noting that AI technology evolves rapidly. Consider how quickly AI image generation advanced over the past year - from obviously synthetic visuals to near-photorealistic results. It’s very likely we’ll see similar leaps in AI-driven 3D modeling, paving the way for more efficient workflows, greater realism, and broader accessibility.

Bonus: Generate 3D Meshes Directly in Blender with AI

Another valuable technique is creating 3D meshes directly within Blender using the “MeshGen” addon. This addon taps into AI-driven modeling features right inside Blender’s interface, saving you the extra step of exporting and importing models between different platforms. For anyone who works heavily in Blender, MeshGen can be a convenient way to experiment with AI-based model generation.

Practical Example: AI-generated Christmas Scene

To wrap things up, here’s a festive example: a Christmas-themed scene made entirely with AI using Bolt. Every element, from the models and shaders to the underlying code and final environment, was generated by AI. This showcases how an entire holiday scene - complete with decorations and atmosphere - can be produced using the right set of AI tools.