Rabatt auf Premium-Pläne, wenn du dich registrierstRabatt sichern

Blog/models·8. Okt. 2026·5 Min.·vom eroq-Team

Nano Banana 2.1 guide — fast drafts and plain-English prompts

Nano Banana 2.1 is Google's fast image model on eroq, typically about 8 seconds a take: how to prompt it in plain sentences, iterate fast and when to move up.


Nano Banana 2.1 is Google's fast multimodal image model and the quickest of the seven models that joined eroq in October 2026, typically about 8 seconds a take. It reads ordinary sentences well, moves easily between styles and costs 31 credits / image, a fraction of its big sibling Nano Banana Pro. That makes it the everyday engine for the part of the job nobody puts in the case study: the twenty attempts before you know what the picture is. Here is how to write for it, how to iterate without wasting credits, and when to hand the idea to a heavier model.

The numbers

  • Price: 31 credits / image, about $0.31 at the entry pack. Nano Banana Pro is 4 times that.
  • Speed: typically about 8 seconds a take, a planning estimate rather than a promise.
  • Takes: 1 to 4 per run; two takes cost 62 credits, four cost 124 credits. Failed takes refund themselves.
  • Formats: 1:1, 3:4, 4:3, 9:16 and 16:9, rendered natively at the same price.
  • Inputs on eroq: a text prompt. No reference pictures, no Edit mode, no negative-prompt field.
  • Access: open to any plan, including accounts that only ever buy credit packs.
  • Moderation: handled upstream by the provider. Refusals return content_blocked and are never billed.
  • API id: nano-banana-2-1.

Write sentences, not tag lists

Nano Banana 2.1 is positioned on eroq for natural-language understanding, so write the way you would describe a photo to a friend. Compare:

  • Tags: cat, laptop, cozy, warm light, bokeh, masterpiece, best quality
  • Sentence: "A ginger cat asleep on a closed silver laptop on a cluttered desk, its tail draped over the trackpad, late-afternoon sun through the blinds striping the scene, cozy and quiet."

The sentence carries relationships that a tag list cannot: asleep on, draped over, through the blinds. Those little words are where the composition lives. A few habits that help:

  • Use verbs. "Leaning", "spilling", "half-hidden behind" place things more precisely than adjectives.
  • One or two sentences. Describe the whole scene in one go; a paragraph of afterthoughts dilutes it.
  • Put the medium early. "A watercolor of…", "An editorial photograph of…". The model moves to match the style you name.
  • Say what you want, not what you don't. There is no negative prompt here, so "an empty beach at dawn" beats "a beach, no people". If you are used to negative prompts, negative prompts explained covers the engines that take one, such as Anima.

The one-clause loop

Speed only pays off with a method. The loop that works:

  1. Write a base sentence.
  2. Render two takes.
  3. Change one clause: the light, the lens, the medium, the time of day.
  4. Render two more and compare.

Changing one thing at a time is how you learn what each phrase does; change three and you will not know which one helped. Each loop costs 62 credits. Ten loops, twenty takes, come to 620 credits, about $6.20 at the entry pack. The library keeps every render with its prompt, and Remix reloads it, so the winning version is never lost.

Styles it moves between

Nano Banana 2.1 covers a broad range from one model. Phrases that change the result decisively:

  • "editorial photograph", "35mm street photo", "studio product shot"
  • "flat vector illustration", "isometric 3D render", "paper cut-out collage"
  • "watercolor on cold-press paper", "pencil sketch", "gouache poster"
  • "risograph print in two inks", "1970s travel poster", "storybook illustration"

Put the style in the first few words and keep it to one per prompt. "A watercolor 3D render" asks for two things and delivers neither.

Example prompts

  1. Studio apartment, 4:3. A complete scene in one sentence, medium first.

    An editorial interior photograph of a small sunlit studio apartment with a fold-down desk, a monstera in a terracotta pot and a striped rug, morning light through sheer curtains, calm and airy.

  2. Mountain town, 16:9. Flat style plus a strict palette keeps a busy scene readable.

    A flat vector illustration of a mountain town at night, snow on the roofs, warm yellow windows, a cable car crossing between two peaks, a limited palette of navy, white and amber.

  3. Harbor watercolor, 3:4. The medium and its quirks ("wet paper blooms") are part of the prompt.

    A loose watercolor of a fishing harbor at low tide, boats leaning on the mud, gulls on the posts, wet paper blooms in the sky, soft grey and ochre.

  4. Island coffee shop, 1:1. Isometric framing and a toy-like finish, described as relationships.

    An isometric 3D render of a tiny coffee shop on a floating island, an espresso machine on the counter, two stools under a striped awning, a little staircase down to a pond, pastel colors, soft shadows, toy-like.

  5. Ramen overhead, 4:3. Food photography vocabulary in plain words.

    An overhead photo of a ramen bowl on a dark wooden counter, a soft-boiled egg cut in half, spring onions, steam rising, chopsticks resting on the rim, warm side light, magazine food photography.

  6. Surfer print, 9:16. A print process named in full, with its imperfections.

    A risograph-style print of a lone surfer walking toward big waves at sunset, two inks only, pink and teal, visible grain and slight misregistration.

When to move up

The full method is in draft fast, finish high-fidelity. To start the loop now, open the image studio, pick Nano Banana 2.1 and set takes to two.

FAQ

How fast is Nano Banana 2.1?

Typically about 8 seconds a take, the quickest of the seven models that joined eroq in October 2026. Times are planning estimates: a busy moment can add queue time.

Is Nano Banana 2.1 good enough for final images?

Often, yes, for social posts, blog headers and anything seen at small or medium size. When a final needs maximum detail, accurate small text or many constraints at once, re-render the winning prompt on Nano Banana Pro.

Does Nano Banana 2.1 take negative prompts or reference images on eroq?

No. On eroq it renders from the prompt alone: no negative-prompt field, no reference pictures, no Edit mode. Describe what you want in the frame, and use Eroq One when a face or product must match your pictures.

Should I write tags or sentences for Nano Banana 2.1?

Sentences. It reads natural language well, and a sentence carries relationships ("asleep on", "draped over") that tags cannot. Name the style in the first few words.

Tagsnano-bananagoogleimage-modelspromptsdrafts

Mach das mit den Modellen hinter diesem Artikel – starte mit 50 Gratis-Credits oder sieh dir alle Engines und ihre Preise an.