The image-gen landscape, July 2026

If you last looked at AI image generation a year ago, the map has been redrawn:

• The chatbots got genuinely good. ChatGPT's image model currently tops the blind-vote arenas, and Google's Nano Banana models render legible text at 4K. For most people, the assistant you already use is now a competent image tool — no new account needed.
• Sora is gone. OpenAI shut its video app down in April; the API follows in September. It's a useful reminder that in this field, tools are weather, and craft is climate — skills transfer, subscriptions don't.
• Video crossed the "startling" threshold. Veo through Google's Flow, Kling 3.0, and Grok's Imagine Video produce short clips with native audio that would have been unbelievable eighteen months ago. Free tiers are stingy, but they're real.
• Local tooling matured. ComfyUI has an actual desktop installer now, and training a personal LoRA takes an evening on a consumer GPU. The gap between "using a website" and "owning your pipeline" has never been easier to cross.

The AI Art page organizes all of this into three tiers — start at no code, and climb only when a tool stops being enough.