Nano Banana 2 / Pro (Gemini Image)

Nano Banana produces photorealistic images from a text prompt or edits an existing photo while preserving what shouldn't change — lighting, pose, background — and altering only what the prompt asks for. It's particularly strong at edits that need to look like they were shot in-camera rather than composited.

On AutorunX it's tuned for commercial imagery: clean product photography and marketing banner art where accurate text rendering, believable reflections/shadows, and brand-consistent color all matter.

Nano Banana is Google's public nickname for its Gemini-family image model, now offered on AutorunX in two tiers: Nano Banana 2 for fast everyday generations and Nano Banana Pro for the highest-fidelity, most complex compositions. Both generate images inside the same multimodal Gemini architecture that handles Google's text and reasoning models, rather than bolting a diffusion model onto a separate text encoder.

Because image generation shares the model's reasoning pathway, Nano Banana can apply a "thinking" step before rendering — parsing a dense prompt's constraints (exact object counts, specific poses, layout instructions, legible on-image text) and holding onto them more reliably than diffusion-only competitors typically manage.

It runs in two modes on AutorunX: text-to-image generation from a written prompt, and image-to-image editing, where an existing photo is transformed while the untouched regions stay pixel-consistent with the source. Output can reach a 4K-capable resolution tier, and the model is known for strong semantic/world knowledge — it tends to get real-world details (brand logos, correct text spelling, plausible physics) right where older diffusion models guess.