Image generation
A model that produces images from a text prompt or other conditioning signal by learning the statistical structure of large image datasets.
Learn first
An image generation model takes a text description — or another image, a sketch, a layout — and produces pixels that match it. Most current systems are trained to reverse a process of adding noise: the model learns how noise degrades real images, then runs that process backwards at inference time. The hard part is coherence at scale: preserving geometry, respecting fine-grained instructions, and keeping outputs consistent across seeds and styles. Quality now rivals studio photography for many subjects; the gap that remains is reliable compositional control.
Where it came from
In megatrends
Related players
Finds citing this concept
Overall = breadth + depth + substance + ½·novelty + freshness (a small boost that halves each week)
Nano Banana 2
Google's cheap, fast image model briefly beat its own premium model — and a rival — on a blind quality leaderboard, at about half the price.
SynthID's truce
OpenAI now stamps Google's invisible watermark onto every image it generates — the moment a vendor silo became a cross-industry convention.
How this connects
Tap a node to open it