Generate Images Side by Side

Image models have specialized: the photorealism leader is not the typography leader is not the editing leader. The practical consequence: for any image that matters, run the brief on two models and keep the winner. In idapt that is one composer and a model switch.

The jobs and their models

Photorealism: the FLUX line remains the reference for photographic light and detail.
Text in the image: posters, logos, UI mockups want a typography specialist (the Ideogram line).
Editing an existing image: instruction-faithful editors change what you asked and leave the rest intact (the Nano Banana line leads here).
Iterating in conversation: chat-native models refine over turns: "warmer light, keep everything else".
Exploration volume: fast, cheap models buy you ten compositions for a flagship's single render.

The image models directory carries live prices per model; the best image generators roundup keeps the picks current with a stated methodology.

The workflow that works

Explore cheap: five variations of the concept on a fast model.
Commit premium: the winning composition, re-rendered on the right specialist.
Edit precisely: fixes via an editing model with targeted instructions, not re-rolls.
File it: images land in your Drive next to the copy or deck they serve; reference them in any later chat.

Because generation lives inside chats, the model that wrote your copy can critique the image against the brief in the same thread: a loop no standalone generator has.

Good to know

Per-image prices are shown before you generate; exploration on a fast model costs cents.
Reference images steer style and identity where the model supports them.
Agents generate images under the same budgets as any tool: a content pipeline can produce its own art.

Run yesterday's image need through two models side by side: the image generation feature page covers the controls, and the comparison surface does the judging layout for you.