CTA keyword: NANO · from the @compedge.ai x @am_roshh carousel
Google made Nano Banana 2.1 generally available on 6 Oct 2026. Google says it improves text rendering, prompt adherence, character consistency across turns and wide or panoramic images, at the same Flash speed. The Decoder reports it costs about half as much per image as Nano Banana 2. Below: 12 prompts that lean on exactly those gains, and a checklist for developers moving off the old model.
What changed
- The model. Nano Banana 2.1 (model id
gemini-nano-banana-2.1) is an update to Nano Banana 2 (gemini-3.1-flash-image). Google's Gemini API changelog lists better visual quality, prompt adherence, multi-turn character consistency, text rendering, and wide and panoramic ratios (1:4, 4:1, 1:8, 8:1) at 1K, 2K and 4K. - The price. The Decoder, citing Google's pricing, puts a 1K image at 3.36 cents (was 6.70), 2K at 5.04 cents (was 10.10) and 4K at 7.56 cents (was 15.10).
- References. The Decoder reports it takes up to 14 reference images at once and keeps up to 4 characters and 10 objects consistent. It can also use Google Search and offers minimal, medium and high thinking levels.
- Where. The Decoder reports it is rolling out in the Gemini app, AI Mode in Search, Google AI Studio, Flow, Stitch, Google Ads and the Gemini Enterprise Platform. Rollouts take time, so you may not see it everywhere yet.
- The old model. Google has deprecated
gemini-3.1-flash-imageand says to migrate togemini-nano-banana-2.1. The Decoder reported a 29 Oct 2026 shutdown; when we checked on 7 Oct, Google's deprecations page said no shutdown date was announced. Either way, plan to move now. - Honest note. The Decoder's own test found Nano Banana Pro often still produces better looking images. 2.1 is the faster, cheaper option, not the best one.
The prompts and the checklist below are our own, not Google's.
How to use these prompts
- Swap everything in [BRACKETS] for your own details.
- Put any words that must appear in the image inside "double quotes". Keep them short.
- Say the ratio you want in plain words (for example "4:1 wide banner"). In the API, also set the aspect ratio in your request, as Google's image generation guide shows.
- Generate 2 or 3 versions and pick the best. Check every letter before you publish.
Part 1: Text in images (4 prompts)
1. Poster with an exact headline
A bold event poster for [EVENT].
At the top, the headline
"[HEADLINE, 2 TO 5 WORDS]" in
large clean sans serif letters,
spelled exactly as written. Below
it, smaller text "[DATE] ·
[CITY]". Background: [SCENE OR
STYLE]. High contrast so the text
is easy to read. No other words
anywhere in the image.
2. Product label or packaging
A studio photo of a [PRODUCT, e.g.
coffee bag] on a plain [COLOR]
background. The front label reads
"[BRAND NAME]" in large letters
and "[ONE SHORT LINE]" underneath,
both printed cleanly and spelled
exactly as written. Soft daylight,
sharp focus on the label.
3. Simple infographic
A clean vertical infographic
titled "[TITLE]". Three numbered
steps, each with a small flat icon
and a short label: 1 "[STEP ONE]",
2 "[STEP TWO]", 3 "[STEP THREE]".
Flat design, [2 OR 3 COLORS], lots
of white space. Spell every label
exactly as written and add no
other text.
4. Shop sign or storefront
A photo of a small [TYPE OF SHOP]
storefront at golden hour. Above
the door, a painted sign reads
"[SHOP NAME]" in [STYLE, e.g. hand
lettered gold] letters, spelled
exactly as written. A chalkboard
by the door says "[SHORT OFFER
LINE]". Realistic, street level
view.
Tip: if a word comes out wrong, ask again in the same chat: "Keep everything the same, but fix the spelling of [WORD] to "[CORRECT WORD]"."
Part 2: Banners and panoramic ratios (4 prompts)
Google lists 1:4, 4:1, 1:8 and 8:1 as new wide and tall ratios.
5. Website hero banner (4:1)
A 4:1 wide website banner for
[BUSINESS]. On the left third, the
text "[HEADLINE]" in large white
letters on a dark area. The right
two thirds show [SCENE THAT SHOWS
THE PRODUCT OR SERVICE]. Keep the
left side calm so the text stays
readable.
6. LinkedIn or YouTube style header (4:1)
A 4:1 wide profile header for a
[ROLE, e.g. freelance designer].
Abstract [COLOR] shapes flowing
from left to right, with the text
"[NAME] · [WHAT YOU DO]" placed
right of centre. Leave the bottom
left corner empty for a profile
photo.
7. Panoramic scene (8:1)
An 8:1 ultra wide panorama of
[PLACE, e.g. a city skyline at
dusk], one continuous scene from
left to right with no repeated
buildings. Even lighting across
the whole width. No text.
8. Tall story or sidebar strip (1:4)
A 1:4 tall vertical strip for a
[USE, e.g. blog sidebar ad]. Top:
the text "[OFFER]" in big letters.
Middle: [PRODUCT IMAGE]. Bottom:
the text "[CALL TO ACTION]". One
colour palette: [COLORS].
Part 3: The same character, every time (4 prompts)
The Decoder reports up to 14 reference images and up to 4 consistent characters per request. Upload your reference images first, then use these.
9. Lock the character
These images show the same
character, [NAME]. Learn their
face, hair, clothing and colours.
Create a clean front view of
[NAME] standing on a plain light
grey background, full body. Keep
every detail the same as the
reference images.
10. New scene, same character
Using the character [NAME] from
the reference images, show them
[ACTION] in [PLACE]. Keep their
face, hair and outfit exactly the
same. Change only the pose, the
background and the lighting.
11. Two characters together
The first reference images show
[NAME A]. The next ones show [NAME
B]. Create a scene where [NAME A]
and [NAME B] are [ACTIVITY]
together in [PLACE]. Both must
look exactly like their
references. [NAME A] is on the
left, [NAME B] on the right.
12. Comic strip with text
A 4 panel comic strip in [STYLE]
featuring [NAME] from the
reference images, who looks
identical in every panel. Panel 1:
[SCENE], speech bubble "[LINE]".
Panel 2: [SCENE], speech bubble
"[LINE]". Panel 3: [SCENE]. Panel
4: [SCENE], speech bubble
"[LINE]". Spell every bubble
exactly as written.
Tip: keep the chat going. Google says 2.1 is better at keeping a character consistent across turns, so ask for changes in the same conversation instead of starting over.
Developer migration checklist
Google has deprecated gemini-3.1-flash-image. The Decoder reported a 29 Oct 2026 shutdown, and Google's deprecations page said no date was announced when we checked on 7 Oct. Move now so a date never surprises you.
[ ] 1. FIND the old model id
search code, config, env:
gemini-3.1-flash-image
[ ] 2. SWITCH the model id to:
gemini-nano-banana-2.1
(one config value, not
hard coded everywhere)
[ ] 3. RERUN 10 real prompts
old vs new, side by side
check spelling, faces,
product shots
[ ] 4. TRY the new options
ratios 1:4, 4:1, 1:8, 8:1
up to 14 reference images
thinking: minimal, medium
or high
[ ] 5. CHECK cost per image
(per The Decoder)
1K: 3.36 cents
2K: 5.04 cents
4K: 7.56 cents
x your monthly volume
[ ] 6. SHIP behind a flag,
watch errors for a day
[ ] 7. REMOVE the old id
[ ] 8. RECHECK Google's
deprecations page weekly
for a shutdown date
Example: a team making 20,000 1K images a month paid about $1,340 at 6.70 cents. At 3.36 cents the same volume is about $672. Run your own numbers with your own volume.
Your test sheet: fill this in
PROJECT: ____________
OLD ID IN: ____________
NEW ID IN: ____________
10 TEST PROMPTS IN: ______
Old model passes: ___/10
New model passes: ___/10
Text spelled right: __/10
SIZE: [ ] 1K [ ] 2K [ ] 4K
IMAGES / MONTH: _________
COST / MONTH: _________
SWITCHED ON: _________
OLD ID REMOVED: _________
The news section reflects Google's Gemini API changelog of 6 Oct 2026 and The Decoder's report of 6 Oct 2026. Prices are as The Decoder reports them from Google's pricing. The prompts, tips, checklist and example math are CompEdge's own.