本文へスキップ
AI図鑑

テキストから画像生成

一文の説明から画像を一枚生成する

画像の生成と編集初級 #18
入力テキスト画像

本ページの本文は英語で提供されています。タイトルと導入は日本語化されています。

この能力とは何か

Maps a natural-language description directly to an image: text in, pixels out, with no annotation or reference image in between. Unlike image-to-image it has no input image at all, and unlike image editing it does not modify an existing picture but synthesises a new one from scratch.

技術的にどう実現するか

The dominant route is latent diffusion: the prompt is encoded into a conditioning vector, injected into a denoising network through cross-attention, denoised step by step in a compressed latent space, and decoded to pixels. Conditioning was later extended by classifier-free guidance, which amplifies prompt adherence by contrasting the conditioned and unconditioned directions. Training uses vast image–text pairs, and DALL·E 3 coupled prompt rewriting with generation in one pipeline, markedly improving adherence to long prompts.

代表的な製品

7

関連する組織

代表的な用途

  • Concept design and storyboards
  • Marketing assets and illustration
  • Pre-visualisation for games and film
  • Personalised avatars and wallpapers

どう評価するか

FID
Distance to the real image distribution; lower is better
CLIP score
Semantic agreement between image and prompt
Human-preference Elo
Preference ranking from pairwise comparison

限界と難しさ

  • Counting, precise spatial relations and ordering remain unreliable
  • Letters inside the image are frequently garbled, especially in long strings
  • Hands, limbs and object-contact points break down, often needing several resamples

背景にある概念