AI image models, explained

Adobe Firefly is Adobe’s commercially safe AI image model — trained on licensed and public-domain content, with IP indemnification and Generative Fill built into Photoshop.

FLUX.2 is Black Forest Labs’ open-weight image family — one 32B checkpoint for generation and editing, strong prompt adherence, and legible typography you can self-host. What it does and how to use it.

GPT Image is OpenAI’s image family, led by GPT Image 2 — a model that plans layout and renders legible text before it draws. What it does, how to prompt it, and when to use it.

Ideogram 3.0 is the image model built for legible in-image text — logos, posters, packaging, signage. What it does, how to prompt it, and where it fits versus Midjourney and Flux.

Midjourney v7 is the release that made a repeatable house style realistic: stabler --sref, Omni Reference, Personalization v2 and Draft Mode. What it does and when to still reach for it.

Nano Banana is Google’s Gemini image generation and editing model, known for keeping the same face across edits. What it does, the model family, and how to prompt it.

Nano Banana 2 is Google’s Gemini 3.1 Flash Image model, launched February 2026 — Pro-level quality at Flash speed. What changed, and which banana to pick.

Recraft V3 is the design-first AI image model — brand style sets, editable vector/SVG output, and controllable text. What it does, how to use it, and where it fits.

Seedream 5.0 is ByteDance’s photoreal AI image and editing model — relight a shot, swap a surface, keep the product itself real. What it does, how to edit with it, and where it fits.

Stable Diffusion 3.5 is Stability AI’s open-weight image family — Large, Turbo and Medium you run locally or by API, with the biggest LoRA and ControlNet ecosystem. What it does and how to use it.