How AI Image Generation Works
AI image generators turn text (and optionally images) into new pictures. Most current tools are diffusion models: they learn to remove noise from images, then generate by starting from random noise and denoising step by step, guided by your prompt.
The main tools
| Tool | Known for |
|---|---|
| Midjourney | strong aesthetics and style, accessed via its web app |
| OpenAI image models (in ChatGPT and the API) | following detailed instructions, text in images |
| Stable Diffusion family (SDXL, SD3, and community models) | open weights, local use, deep control |
| FLUX models | high quality, open and hosted variants |
| Adobe Firefly | integration with Photoshop, commercial-safety focus |
| Ideogram | text and typography in images |
Features, pricing and terms change often; check each tool's current documentation and licence.
Strengths
- Concept art, mood boards, illustrations and backgrounds.
- Product scenes and lifestyle mock-ups.
- Variations and rapid exploration of ideas.
- Editing: extending images, replacing objects, removing backgrounds.
Weaknesses
- Exact details: hands, fine text, logos, specific product features can be wrong.
- Consistency of the same character or product across many images (improving with reference features).
- Accurate depictions of real people and trademarks raise legal and ethical problems.
Commercial use
Before using images commercially, check:
- the tool's terms of service and the plan you are on (some require paid plans for commercial use),
- whether outputs can include recognisable trademarks or styles you should avoid,
- local law on copyright for AI-generated images, which is still developing and may limit protection of purely generated work.
Check yourself
- How does a diffusion model generate an image?
- Name two weaknesses of current image models.
- What should you check before using generated images commercially?
進捗はこのブラウザーにのみ保存されます。
レビューに提出（10 ポイント）