Nano Banana 2 AI Image Generator

Every image below was generated on this page on 18 August 2026, first take, with the exact prompt shown.

8:1 banners that are composed, not cropped

This is the reason to come to this page. An 8:1 frame is where most image models fall apart into repeated tiles, because they are really generating a normal image and stretching it. Ours came back as a genuine panorama: a single ridgeline running from one edge to the other, mist pooling in the valleys, warm light on the crests, everything scaled consistently across nearly six thousand pixels of width. That is a usable site header straight out of the model, not a crop from something squarer.

1:8 vertical strips, in one continuous piece

The tall direction is harder than the wide one, because the subject has to stay coherent over an enormous vertical run. We asked for a single ancient sequoia photographed from its base looking straight up. What came back holds together the whole way: continuous bark texture with no seam where a lesser model would restart, the trunk narrowing correctly with height, and forest light breaking through the canopy at the top. Useful for sidebar rails, tall mobile banners and print strips — formats that normally mean commissioning a photograph.

Fourteen references — and what happens to the small print

It takes 14 reference images in one pass. We gave it four — a labelled tumbler, a skillet cornbread, a lighthouse photograph and a poster — and asked for a single overhead table scene under one consistent morning light. All four arrived, recognisable, correctly lit and correctly scaled. The large label lines came through spelled correctly. The smaller line beneath them did not: it degraded into marks that read as text but are not. So plan on it holding your headline-sized text through a composite, and on setting anything smaller yourself.

• Web and product designers

  Site headers, hero strips and sidebar rails at proportions that normally mean cropping a stock photo and losing most of it. 8:1 and 1:8 come out composed for that shape.

• Anyone merging many references

  Fourteen images in one pass. Our four-reference composite kept all four subjects recognisable under a single consistent light.

• Teams producing print strips and banners

  Long formats — bookmarks, roll-ups, shelf strips, letterbox title cards — plus a 4K tier when the output has to go to press.

• Not for: small print inside a composite

  Headline-sized text survives a multi-reference merge; the small secondary line in our test did not. Set fine type yourself afterwards.

Credits, reference-image limits, resolutions and ratio counts are read from our own model registry on 18 August 2026 — the ratio counts exclude "auto", which is a setting rather than a ratio. Speeds are our own measurements with the sample size stated.

1. 1

   Open the workspace

   Create a free account — no API key, no setup. The generator on this page is already set to Nano Banana 2.

2. 2

   Pick the shape first

   If you need a banner or a strip, choose 8:1, 4:1, 1:4 or 1:8 before you write the prompt, and describe the scene as continuous — "a ridgeline stretching from one end of the frame to the other" rather than "a mountain".

3. 3

   Attach up to 14 references

   For a composite, say what each reference is for. Headline-sized text on a reference survives the merge; set anything smaller yourself afterwards.

4. 4

   Choose a quality tier

   1K to iterate, 2K for most finals, 4K when it is going to print. The exact credit cost shows before you run.