What is this?
AI image generation creates and edits pictures from text descriptions. It was generative AI's first viral breakthrough. The frontier has moved from "can it draw?" to precise editing: change one object and keep everything else. Readable text and consistency across a series are the current tests.
Key tools & players
- Midjourney — the aesthetic benchmark
- FLUX (Black Forest Labs) — strongest open-weight family, powers many apps (including ours)
- Google's image models — conversational editing that went viral as "Nano Banana"
- DALL·E / GPT image (OpenAI) — image generation inside ChatGPT
- Stable Diffusion — the open-source project that democratized everything
- Adobe Firefly — commercially-safe generation inside Photoshop
Milestones
- 2021 — DALL·E shows text-to-image is possible
- 2022 — Stable Diffusion open-sources it; Midjourney makes it beautiful; an AI image wins an art fair
- 2023 — Photorealism arrives; the "AI hands" problem fades
- 2024 — FLUX raises the open-weight bar; in-image text becomes readable
- 2025 — Conversational editing goes viral; provenance labels (C2PA) spread
- 2026 — Identical-character series and exact layout control mature
- Aug 2026 — Provenance becomes law: the EU and California require machine-readable AI labels (C2PA) in the same week
- Aug 2026 — Google makes the visible AI badge optional outside the EU and South Korea; invisible SynthID and C2PA stay
- Sep 2026 — Apple's iPhone signs the photo at the sensor and adds a verifiable original (our coverage)
- Sep 2026 — Alibaba's Qwen-Image drops Apache 2.0 for a research-only licence, splitting open weights from open use (our coverage)
Mini glossary
- Diffusion: generating an image by progressively denoising random pixels
- Inpainting: regenerating only a selected part of an image
- LoRA: a small add-on that teaches a model a specific style or face