TL;DR
As of June 2026, the Gemini app generates images with two models: Nano Banana 2 (Gemini 3.1 Flash Image), the fast default everyone gets, and Nano Banana Pro (Gemini 3 Pro Image), the slower, studio-grade option you reach by choosing the “Thinking” model. The old reputation — “Gemini can’t do legible text or accurate detail” — is out of date. Nano Banana Pro renders readable headlines, paragraphs, and multilingual signage, outputs up to 4K, and holds up to five characters consistent across a series. The remaining hard limits are real people (public figures are blocked) and free-tier quota.
This guide gives you the model picker, the current limits, a five-slot prompt structure, and the failure modes worth planning around.
Which model you are actually using
Open the Gemini app, tap Create images, and you get a model toggle:
| Option in the picker | Underlying model | Best for | Speed |
|---|---|---|---|
| Default (“Fast”) | Nano Banana 2 (Gemini 3.1 Flash Image) | Quick drafts, iteration, social cards | Seconds |
| ”Thinking” / Pro | Nano Banana Pro (Gemini 3 Pro Image) | Posters with text, infographics, final assets | Slower; it plans the scene first |
Nano Banana 2 shipped to the app on February 26, 2026 and replaced the older Imagen-based generator. It brings near-Pro quality at Flash speed and is what free users see by default. Nano Banana Pro, built on Gemini 3 Pro, “plans” the composition before rendering — that is why its text and layout are dramatically better, and why it takes longer.
Free vs paid: the limits that actually bite
Image quota is the thing that stops most people, not capability. As of June 2026 (Google adjusts these often because demand is high):
| Plan | Price/mo | Image model | Rough daily images |
|---|---|---|---|
| Free | $0 | Nano Banana 2 only | ~20/day |
| Google AI Pro | $19.99 | Nano Banana 2 + limited Nano Banana Pro | Higher quota |
| Google AI Ultra | $99.99 | Full Nano Banana Pro | Highest quota; visible watermark removed |
Two things to know:
- “Gemini Advanced” no longer exists by that name. Google renamed the paid consumer tier to Google AI Pro ($19.99/mo) in early 2026; the old “Google One AI Premium” branding is gone. If a tutorial still says “Gemini Advanced,” it predates the rename.
- Free users get a small daily allotment of Nano Banana Pro “Thinking” generations, then quietly drop back to Nano Banana 2. If your text suddenly looks worse mid-session, you have likely used up the Pro quota.
Aspect ratios, resolution, and watermarks
- Aspect ratios: Nano Banana Pro supports a real range — 1:1, 4:3, 3:4, 16:9, 9:16, plus cinematic 1.85:1 and 2.39:1. Ask for the ratio explicitly in the prompt; do not assume the default.
- Resolution: 1K, 2K, or 4K (up to 4096x4096). Request 4K only for print or large-display work; it is slower and eats quota.
- Watermarks: every image carries an invisible SynthID watermark plus C2PA Content Credentials metadata, which survive cropping and compression. A visible Gemini sparkle appears on Free and Google AI Pro images; only Google AI Ultra removes it. Plan around the visible mark if the asset goes to a client.
The five-slot prompt recipe
The single biggest quality lever is structure. Fill all five slots; a missing slot usually produces a generic result.
Generate an image of [subject], in [style], with [lighting],
[composition], [mood]. Aspect ratio [16:9]. [resolution].Worked example:
Generate an image of a ceramic pour-over coffee setup on a
walnut counter, in soft editorial product-photography style,
with warm window light from the left, shot at a low 30-degree
angle with shallow depth of field, calm and premium mood.
Aspect ratio 4:3. 2K.Make the style slot specific. A named reference beats an adjective every time:
- “Studio Ghibli watercolor” beats “anime.”
- “Annie Leibovitz portrait lighting” beats “professional lighting.”
- “flat editorial illustration, muted palette” beats “nice graphic.”
Text in images: now a strength, not a trap
This is the headline change. The old advice — “never ask Gemini for text, it produces gibberish” — applied to Imagen. Nano Banana Pro renders legible text reliably, including short taglines, full paragraphs, and non-English scripts. To get clean type:
- Use the “Thinking” (Pro) model — Nano Banana 2 is decent at text but Pro is the one to trust for headlines.
- Quote the exact words:
add the text "Spring Sale — 30% off" in bold white, top-center. - Keep each text block short and specify placement and weight. Long paragraphs work but proofread at 100%.
- For a poster, generate the full thing with text in one pass rather than adding it downstream.
If you are stuck on the free tier with Nano Banana 2 and the text is shaky, fall back to the old workflow: generate text-free art, then set type in Slides, Canva, or Figma.
Consistency across a series
For a multi-image set (a carousel, a slide deck, a character that recurs), Nano Banana Pro can hold up to five characters consistent and keep the fidelity of up to fourteen reference objects in a single workflow. Practical method:
- Generate the hero image first and download it.
- Upload it as a reference and prompt the next scene: “same character, same outfit, now in a kitchen.”
- Keep the style and lighting clause byte-for-byte identical across prompts. That clause is what carries continuity.
This is a genuine improvement over the previous generation, where character continuity was weak enough that silhouettes were the safe workaround.
What still gets blocked
- Real public figures and recent political events are refused. Reframe to “a person who looks like…” or an original character.
- Copyrighted characters are blocked. Build original designs.
- Violence, sexual content, and other sensitive themes trip the safety filter. If a prompt is refused with no detail, reword the most charged phrase first.
If the generator refuses or hangs entirely, see Gemini image generation unavailable for the specific error states and fixes.
Common mistakes
- Stuffing style words. “Cinematic, dramatic, vibrant, photoreal, ultra-detailed” cancels itself out. Use two adjectives and one specific reference.
- Assuming Gemini still can’t do text. That was Imagen. On Nano Banana Pro, ask for the exact words in quotes.
- Generating a whole series in one prompt. Quality drops on multi-asset prompts. One image per prompt, reuse the reference.
- Burning 4K quota on drafts. Iterate at 1K, render the winner at 2K or 4K.
- Shipping the first generation. Zoom to 100% and check hands, fine type, and perspective before approving.
FAQ
- Do I need a paid plan to generate images?: No. Free Gemini generates with Nano Banana 2 (around 20 images/day as of June 2026). Paid plans add a quota of the higher-quality Nano Banana Pro “Thinking” model.
- What is “Nano Banana”?: It is Google’s nickname for its Gemini image models. The app default is Nano Banana 2 (Gemini 3.1 Flash Image); the premium one is Nano Banana Pro (Gemini 3 Pro Image).
- Is “Gemini Advanced” the plan I need?: That name was retired. The paid consumer tier is now Google AI Pro ($19.99/mo) or AI Ultra ($99.99/mo).
- Can it really write legible text now?: Yes, on Nano Banana Pro. Quote the exact words and keep blocks short. Nano Banana 2 is weaker at type; use Pro for anything with a headline.
- How does it compare to Midjourney?: Gemini is faster, free to start, and now strong at in-image text and multilingual signage. Midjourney still has a higher ceiling for painterly, art-directed finals. Use Gemini for quick branded assets, Midjourney for finished art. See the Midjourney beginner guide.
- Can I tell if an image was AI-made?: Yes. Every Gemini image carries an invisible SynthID watermark and C2PA Content Credentials; you can upload an image back into the Gemini app to verify it.