# What is AI image generation?

Japanese: 画像生成AI (がぞうせいせいエーアイ)

**In one line:** AI image generation is artificial intelligence that creates new images from text prompts and other input, based on patterns learned from data.

AI image generation is AI that makes pictures from words. Type something like “a cat swimming in a sunset sea.” It makes a drawing or a photo-like image to match. It learned from lots of pictures paired with descriptions. That is how it knows how words and looks go together. A common method starts with something like TV static. Then it removes the noise bit by bit, until a picture appears. Anyone can make a picture in seconds. But rights to the pictures used for learning are being debated. It is a friend to people who can’t draw. And a headache for people who can.

**Common mix-up:** It is not cutting and pasting photos. It creates the image from scratch, based on features it has learned.

**How a picture comes out of noise**

Most tools today use a “diffusion model.” In training, clear pictures are slowly buried in noise, like TV static, and the AI learns the reverse: how to remove a little noise at a time. To make a new picture, it starts from pure static and removes noise dozens of times, steering toward what your words describe. Many models do this in a squeezed-down version of the image, which is why some can even run on a home computer.

**The 2022 boom**

Research on turning text into pictures is older, but in 2022 OpenAI’s DALL·E 2, Google’s Imagen, Stability AI’s Stable Diffusion and Midjourney arrived one after another, and results began to look like real photos and hand-drawn art. Stable Diffusion was released openly, so people around the world built their own versions and add-ons, and it spread very fast.

**What about copyright?**

In Japan, Article 30-4 of the Copyright Act, added in 2018, generally lets works be used without permission for purposes like AI training, where nobody is “enjoying” the work itself. It does not apply when it would unfairly harm the rights holder. And if a finished image looks like an existing work and was based on it, it can infringe, just like a hand-drawn copy. Japan’s Agency for Cultural Affairs published its official view on AI and copyright in March 2024.

**How people use it**

- “I made three versions of the slide picture with an image generator.”
- “That picture looks AI-made. The hand has six fingers.”

**FAQ**

- Does AI image generation just cut and paste? — No. It builds the image from scratch, based on features it has learned.
- Is an AI-made picture protected by copyright? — A picture made fully automatically by AI is generally thought to have no copyright. If a person shaped it creatively, through detailed direction or editing, it may. Rules differ by country, so check official guidance.
- Can I use AI images for business? — It depends on each service’s terms. Some allow it, some only on paid plans. Images that look like existing characters can still break copyright.
- How can you tell an AI picture? — Odd fingers, broken text and strange shadows are clues, but many images are now almost impossible to spot. Some companies now attach “content credentials” that record how an image was made.

Topic: Tech
Published: 2026-11-15
Source URL: https://eli5.net/en/w/gazo-seisei
日本語: https://eli5.net/w/gazo-seisei
