How to generate good assets with AI Generation

AI Generation might seem hard to do, well at least when trying to get a perfect result. The quality of models are increasingly improving, but there still seems to have a level of complexity in getting the model to work properly. This guide is exactly for this purpose.

Fundamentals

AI Generation , oversimplified, boils down to three things:

• Model Medium
• Input Data
• Expected level of quality in output

Model Medium

There are different types of AI Models that perform generation. Some of them accept a different kind of input data, such as text, image or sound. Some models are even multimodal , which means they accept two mediums, (eg. FLUX Klein 2 can receive an image and a text prompt, to create a new image or edit the current image). The medium is important, because it determines not only the type of input data, but also in the case of text-to-anything models, the type of prompting required for good results.

Input Data

Input Data, once you understand the model medium, is the data you provide to the model to generate a result. For example, if you are using a text-to-image model, the input data would be a text prompt describing the image you want to generate.

But the input data may differ from model to model, even if they share the same input medium. For example, if you want a specific style of image from a retro camera, some models will prefer the camera Name to be included in the input data, whilst others might prefer camera settings such as aperture, filters, etc. These characteristics are model-specific and usually happens during model training, when the model learns to associate input data with the desired output (labelling).

Expected Level of Quality

Sometimes, you might want the AI to create a structure of the work rather than a detailed project. Even in the opposite case, understanding the previous points will lead you to refine the final text prompt to what you need, or in the case of image-to-anything models, the proper images to inspire your model.

Why bigger creations need more layers

There are multiple ways to create layering in VoxelAI Studio. The layering system is compatible with MagicaVoxel, and as your model grows, the layering system will not only help you get properly organized for your own work, but it will also help the agents work within your project to edit anything.

Agents within VoxelAI Studio can see, edit and tweak layers. You can also use the AI to help you layer your own work.

Color and material naming conventions

At generation time, our AI will generate a sensible, named palette for your model. This palette is a way to constrain the colors, give a proper stylistic approach to the models and make it easier to work with. See in the docs for the existing material id/name/hex explanation.

Wrap-up

Bigger Models will take more resources to generate, not only in amounts of AI tokens, but also colors, shape, etc. That means smaller projects and or iterations might be needed, if you want to build a city for example, that will have multiple colors within the project, possibly from differing palettes.

Want to experiment it yourself? Try it out on the /generate page.