An AI model is a trained mathematical algorithm that has learned from data and can generate new content on this basis. In the field of image generation, this means that the model calculates a new image from a text prompt and optional reference images.

Examples of AI models are Nano Banana from Google and FLUX from Black Forest Labs. These models are the actual “engine” that generates the images.

Important: An AI model is not a finished application, but rather the computational logic in the background. It always requires software or a platform in order to be used.

An API (application programming interface) is a technical interface that allows software to communicate automatically with a service (in our case, an AI service).

In terms of AI image generation, this means that a system sends a reference image and a text prompt to an AI service, for example, and receives the generated image directly in return. An API therefore enables the automated, system-integrated creation of AI material.

Fine-tuning means retraining an existing AI model in order to adapt it to specific content or tasks. This involves further training the basic model with a certain number of photos – of certain people, products or styles, for example – so that it delivers more precise and customised results. Fine-tuning requires more computing power than training a LoRA, but changes the model directly and permanently.

Inpainting refers to the targeted replacement or addition of image areas within an existing photo, for example to retouch parts or insert new content in a specific area.

LoRA (Low-Rank Adaptation) is a compact extension for AI models. It makes it possible to add new content such as specific people, products or styles without having to retrain the entire model at great expense. This saves time and resources and makes AI adaptations much easier.

Outpainting extends a picture beyond its original edges in order to continue the picture content in a logical and stylistically appropriate way.

A reference image is an existing image that serves as a visual template for the AI. The AI uses this image to adopt or draw on certain features such as face, pose, style, lighting, or perspective. The reference image therefore determines the appearance and direction of the newly generated AI material.

SaaS stands for “Software as a Service.” This means that software is not installed locally, but is provided via the cloud and used in a browser.

One example is Freepik. Freepik provides design and AI functions online and accesses AI models in the background with the help of an API. This API allows Freepik to communicate automatically with AI models. The AI model generates an image and sends the result back.

In short:
• The AI model generates the image.
• The SaaS service provides the user interface and infrastructure.
• The API connects the platform and model technically.