Neural networks for 3D modeling are gaining popularity because they simplify the work of designers. AI takes on complex, routine processes, increases the accuracy of models, and saves time. We will tell you how AI can be used in 3D modeling and whether it can replace graphic editors and special software.
Operating principle
For 3D graphics, generative adversarial neural networks (GAN) are usually used, consisting of:
-
generator - studying the basic distribution of data, generating images;
-
discriminator - evaluating the quality of the generator’s output data.
In a very simplified form, the process looks like this. Generator creates a picture based on the description, and the discriminator analyzes how similar it is to a conventional apple, cat, or umbrella. If the result is identified as fake, the generation-verification process will continue until the discriminator is satisfied with the result.
Using AI in 3D modeling
Models with integrated artificial intelligence have greatly simplified the lives of designers, taking on the most difficult part of the work. AI can automatically:
- convert 2D images into 3D;
- create textures, effects;
- reproduce characters, scenes for animation, games.
In addition, it can speed up rendering processes, adapt images to specified parameters (lighting, style, effect), interact with the user, and give recommendations for improving content.
Tools
The main methods of modeling:
- animation - key, procedural, rigging (imitation of realistic movements);
- rendering - creating a raster image from a three-dimensional scene, used in games, video creation;
- texturing - applying textures to the base, repeating real prototypes;
- lighting - to add depth, highlight elements, create the desired atmosphere of the scene.
The importance of choosing lighting can be explained with an example. Let’s say the user has modeled a three-dimensional figure of an athlete. Carefully worked out the muscles, muscles. What happens if you “set” the light in front of the model. The relief press, which took so much time to draw, will visually smooth out. And if you light the figure from the upper left corner, the viewer will be able to evaluate the quality of the object’s elaboration.
As for special programs, the most popular are:
- Autodesk Maya - an editor loved by designers, artists, with an open source code, which expands the possibilities for optimizing the software for the tasks of a specific studio;
- ZBrush - a convenient tool for sculpting;
- Cinema 4D - supports a set of plugins, is compatible with other software, has an intuitive interface with convenient icons, due to a large set of options it is used for complex scenes, animated characters;
- ChatGPT - it has many limitations, but the bot copes well with basic tasks, creates files for printing on a 3D printer.
All software has free limited access. You can use the full functionality only after paying for a subscription.
Examples
Let’s start with the most popular - Blender. This is open source software with tools for rendering, sculpting, and converting 2-D content. The service is free, and due to its open source code, it can be easily optimized for specific tasks.
The list also includes AI:
-
Tripo3D - can do the simplest rigging animation, generate an alternative version of a character, supports GLB, USD, FBX formats;
-
Meshy - understands a request in the format of text or a reference picture, generates textures based on the description, supports animation;
-
Meshcapade - used for animation, will cope with the task of generating a realistic human figure, automatically changing the proportions of the body depending on the specified height and weight.
After generation, you can load it into a 3D editor, since most likely you will not be able to get the desired result without artifacts that does not require editing.
Recommendations
To get a predictable result, you need to compose prompts correctly. The main rules:
-
the first hint for AI should start with a precise command, for example create, draw image (create, draw an image), otherwise the AI may decide to chat, give a link to an article or a joke;
-
the more precise and specific the description, the better - Microsoft recommends using at least six descriptive words in a prompt for its Copilot;
-
accurately formulate a description of the style - pixel, realistic, anime;
-
do not use negative particles, introductory words - the AI does not perceive them, automatically “throwing them out” from the request;
-
the background, surrounding objects, colors, effects also need to be described in detail, but briefly, otherwise the AI may get confused.
To generate an object from a photo, picture, take high-resolution images, cut out the background, it will be easier for the neural network to understand what is wanted from it. Do not use samples with several elements. If you need to get a conventional cabinet from a prototype picture, which shows a room with furniture, it is better to remove unnecessary details and leave the main item.
You can evaluate what AI is capable of for free on the website. Access to working with the neural network will open immediately after authorization. To use the full functionality without limits on requests, you can choose one of three subscription options.
Results
It is difficult to call artificial intelligence an independent tool yet, since the generated content often requires additional adjustments. But according to forecasts, the technology will develop, opening up new possibilities for 3D graphics.
