AI Terminology Course
AI Terminology
/
Intermediate

Image Generation

Definition

The use of deep learning models (primarily Diffusion models) to generate novel visual imagery based on natural language text descriptions.

Explain Like I'm New

You type 'A cat riding a skateboard on Mars'. The AI starts with a screen of pure TV static, and slowly removes the static pixel-by-pixel until a perfect photorealistic image of a cat on Mars appears.

Real World Example

Midjourney, DALL-E 3, and Stable Diffusion.

Common Use Cases

  • •Graphic design
  • •Concept art
  • •Advertising

Interview Questions

basic

  • Can an Image Generation AI create a picture of something that has never existed in the real world?

intermediate

  • What happens if two users type the exact same prompt into Midjourney? Will they get the exact same image?

Flash Cards

Question

Create new things?

Click to reveal answer
Answer

Yes. It does not 'copy and paste' from Google Images. It understands the mathematical concepts of 'Cat', 'Skateboard', and 'Mars', and chemically blends them together to synthesize a brand new arrangement of pixels.

Question

Same image?

Click to reveal answer
Answer

No. The generation process starts with random mathematical noise (a random 'Seed'). Because the starting noise is different every time, the final resulting image will always be unique.