Text-to-Image Generation
Creating a new image from a written description, by combining a text encoder that understands the prompt with a generative image model.
Also image generation · prompt to image · DALL-E
You describe a picture in words, and the computer paints it for you, even something that has never existed, like a cat astronaut on the moon.
Where it fits
+All relationships as a list
Is a kind of / part of
Kinds and parts
Siblings
Uses / built on
Used by
Evolved into
How it works
A text-to-image system has two jobs: understand the prompt, and paint a matching picture.
- Understand. A text encoder turns the prompt into embeddings. Early systems used CLIP’s text encoder; newer ones often use larger language models, which follow complex prompts better.
- Paint. A generator creates the image while “looking at” those embeddings. Today this is almost always a diffusion model, usually working in a compressed latent space as in Stable Diffusion. Each denoising step attends to the text through cross-attention.
Classifier-free guidance turns the prompt’s influence up or down: higher guidance follows the prompt more literally but reduces variety. You can see exactly this trade-off in the diffusion widget on the diffusion model page.
Everyday use cases
- Case 01
Illustration and concept art
Quick drafts of scenes, products or characters to iterate on.
- Case 02
Marketing and presentations
Custom images instead of stock photos.
- Case 03
Photo editing
Describe a change ("add snow", "remove the car") and fill it in.
Famous examples
- 2022Stable Diffusion
An open-weights text-to-image model based on latent diffusion. Its public release in 2022 let anyone generate images on their own computer.
- 2021DALL-E
OpenAI's text-to-image models. The first (January 2021) generated images token by token with a Transformer; DALL-E 2 (April 2022) switched to diffusion guided by CLIP.
- 2022Midjourney
A popular text-to-image service from an independent San Francisco lab, launched in open beta in July 2022 and first used entirely through a Discord bot.
Strengths, weaknesses and when to use it
Strengths
- Turns ideas into images in seconds, no drawing skills needed.
Weaknesses
- Struggles with exact text, counts and spatial instructions.
- Copyright, consent and deepfake concerns around training data and outputs.
When to use itFor drafts, illustration and creative exploration; be careful with realistic images of real people.
Compare with
Check yourself
- Q01
How does the prompt steer the image in a diffusion-based generator?
Show answer
A: The text embedding conditions every denoising step, usually through cross-attention