From 2D to 3D: Bring Characters and Objects to Life with AI

Generating 3D models with AI typically requires creators to develop a character concept or product design in one image-generation platform, download the asset, run it through a separate AI 3D generator, and then import it into a 3D tool for final modifications. And things only get worse as the project becomes more complex. Moving dozens of assets across different platforms turns into an organizational nightmare.

To help creators focus on creating, we brought 3D generation directly into Leonardo. Creators can now generate their initial 2D concepts using the industry’s best AI image models and turn them into 3D models all in one place.

Whether building background props for an indie game or product mockups for an interactive web viewer, the process remains unified. Assets can be exported as a .glb file and dropped directly into downstream rigging, slicing, or engine pipelines.

Can You Generate a 3D Model with AI?

Modern generative AI architectures can now synthesize functional, production-ready 3D models. These tools can create 3D assets that are ready for rendering, animation, or physical fabrication. Let’s explore three of the most common applications.

Indie Game Development

For independent game studios, 3D generation acts as an engine for building large volumes of secondary assets. Developers use AI to populate open worlds with background non-playable characters (NPCs), generate environmental scatter props like barrels and crates, and iterate on different variants for enemy mobs. This allows teams to build expansive environments while focusing their manual modeling efforts on primary characters and hero assets.

E-Commerce and Product Visualization

Retail and e-commerce designers use 3D visualization to build interactive web configurators and augmented reality experiences. These digital assets allow customers to rotate and inspect products such as clothing, footwear, furniture, and jewelry directly on a screen. AI generation makes it possible to rapidly prototype these digital twins, bringing physical inventory into a virtual space for marketing and customer interaction.

3D Printing and Fabrication

For the fabrication and maker community, generative 3D serves as a direct ideation tool to translate concepts into physical objects. Hobbyists and fabricators use it to generate tabletop gaming miniatures, custom hardware brackets, and complex decorative sculptural pieces. It offers a fast way to move from a basic visual idea to a structural model ready to be prepared for a 3D printer.

How to Make 3D Video Game Characters

We’ll demonstrate how to create a 3D enemy mob model using AI from start to finish. To begin, open the Leonardo app and navigate to the Image panel:

On the image generation page:

1. Choose an image model.
2. Choose your generation settings (we recommend the square format for this task).
3. Write your prompt and click Generate.

Once you are happy with the generated image, head to the 3D tab:

For the best results, the 3D generation model needs to view your character from different angles. For this reason, use the 3D Reference View Creator blueprint, which you can access directly from the prompt bar:

Select your character image and click Generate:

Once the Blueprint generation finishes, add your reference views (up to five) using the Image Reference button:

Next, select the 3D generation settings. While we will explain these settings in detail in a later section, you can select a Quad mesh at an 18k quality, and choose All as the material:

You can now begin the generation! To get a 360-degree view of the result, click the View 360° button:

In the final step, download your model as a .glb file:

How to Make a 3D Model from an Object

Turning an object into a 3D model follows a process almost identical to the one we just covered for creating a character. For this reason, you can refer to the screenshots above if you have trouble navigating the app menu while following these steps:

1. Generate or upload an image of the object you want to turn into a 3D model.
2. Generate multiple reference views of the object using the 3D Reference View Creator blueprint.
3. Add the reference views using the Image Reference button.
4. Set your 3D generation settings to a Triangle mesh at a 20k quality, and choose All as the material.
5. Click View 360° to inspect the model and download it.

If you want to try out this process, you can use this prompt in the first step:

Here’s the result we got by following the steps above:

Tips for Getting the Best Results with AI 3D Generation

An efficient process starts with a clean 2D image and requires the right structural settings for the specific project. You can control the settings of the 3D generation from:

1. The main settings menu.
2. The advanced settings menu.

Let’s see how to set up the input and configure the generation for professional results.

Start with a Clean 2D Input

The AI infers the entire shape of an object based on its 2D outline. If the input image has a cluttered background or overlapping elements, the AI becomes confused about where the object ends and where the background begins. Always use a solid white, neutral, or transparent background and focus on a clear, readable silhouette.

Lighting is equally important. Use soft, even studio lighting in the image prompt. Harsh directional lighting or deep shadows will be baked into the 3D texture as permanent dark spots. This looks unnatural when the asset is placed in a new 3D environment with different lighting.

Once you’re satisfied with the quality of the 2D input, make sure to generate multiple views using the 3D Reference View Creator blueprint.

Choosing the Right Mesh and Quality

A 3D model is built from a mesh of connected points. You have two options for this structure:

1. Triangle geometry is the standard choice for static objects like furniture, weapons, or decorative props. It handles complex shapes well and works reliably across web viewers and 3D printers. Your quality options range from 2k up to 500k. Start at 20k polygons to keep generation times fast while testing different shapes. Once the form is correct, increase the quality to 150k or 500k for the final export.
2. Quad geometry builds the mesh using four-sided squares. This is mandatory if the model is a character that will be animated. A quad structure provides logical edge loops, allowing joints to bend without breaking the visual texture. The quality options range from 4k up to 50k. Start at 18k to evaluate the overall structure before committing to a denser 50k final render.

Selecting the Right Material

The material setting determines how the surface of your 3D model renders and reacts to light.

PBR (Physically Based Rendering) simulates how light behaves in the physical world. A metal armor plate will reflect light differently from a cotton shirt. If your model needs to look realistic in a lit scene, a game engine, or a product mock-up, PBR is the correct choice.

The Shaded option provides a stylized surface without complex lighting physics. This is useful for stylized games, illustrations, or workflows where you plan to control the final look manually once the model is in your scene.

If you are still deciding on the final render environment, choose the All setting. This generates both PBR and Shaded textures simultaneously, allowing you to compare options later without regenerating the geometry.

T-Pose Vs. A-Pose

If you are generating a character for animation, the starting posture dictates how well the model will move. The T-pose positions the character with arms sticking straight out to the sides. While this aligns with global grid axes, it forces severe stretching on the shoulder geometry.

The A-pose is the modern industry standard for production models. It places the character in a relaxed stance with arms angled down. This reduces topological stress on the mesh, making the rigging process cleaner and resulting in natural shoulder deformations during movement.

Fine-Tuning with Advanced Settings

If your starting 2D image has a transparent background, turn on the Use Original Alpha setting. This prevents background noise from bleeding into the 3D geometry and preserves sharp edges.

The Bounding Box setting helps the model understand the physical limits of the object, ensuring it does not cut off important details.

Finally, use the Seed number to iterate predictably. If you use the same seed number with the same input and settings, you’ll always get the same result regardless of the stochastic nature of AI models.

A Unified Workflow Protects Creativity

If generating 3D assets with AI means jumping between multiple applications, creativity is lost to organization. On Leonardo, you can go from a 2D concept to a production-ready 3D model and keep your entire creative pipeline in one place.

Whether you need a simple prop for a mobile game, a quad-mesh character, or a model for a 3D print, you can head to the Leonardo app and go from an idea to a 3D asset without leaving the platform.

FAQ

How to bring a character to life in 3D?

To bring a character to life in 3D, start by generating a 2D concept image with a clean silhouette and neutral background. Next, use the 3D Reference View Creator blueprint to generate multiple angles of the character. Process these reference images through a 3D model like Rodin V2, selecting a Quad mesh if the character needs to be animated. You can complete this entire unified workflow directly within the Leonardo platform without exporting files between steps.

How to create a 3D figurine with Nano Banana?

You can create the initial 2D concept of a figurine using AI image models like Nano Banana or GPT Image. However, to turn that flat 2D concept into a 3D physical asset, the image must be processed through a 3D generation model like Rodin V2 to create a structural file. Instead of bouncing between different tools, you can use Leonardo to generate the 2D concept and convert it into a 3D file in a single platform.

How to make a 3D model from photos?

Making a 3D model from photos requires a clear subject with a neutral or transparent background, so make sure to fix that before anything else. Next, upload your photo to an AI generation platform like Leonardo (where you can also adjust the background) and create multiple reference views of the subject from different angles to help the AI understand the object's overall volume. Run the reference images through the 3D generation settings, typically using a Triangle mesh for static objects. You can manage this entire process from photo upload to final 3D export inside Leonardo.