Nano Banana

Overview

Nano Banana is Google's name for Gemini's native image generation. This selection page is for video creators who need a YouTube, TikTok, or Shorts thumbnail or cover, a storyboard or reference still, or a first frame to hand to a video model. The model ids below output an image, and text if you leave the default response on. A still can be an input to Veo. That handoff is a workflow. It is not a flag on this model.

The image generation guide, footer "Last updated 2026-10-09 UTC," read 2026-10-20, says to use Nano Banana for image generation and editing. Text-to-image and image-editing are the capability tags for that sentence. Video-to-image is the same guide's still-from-video feature on Nano Banana 2.1, Gemini 3.1 Flash Image, and Gemini 3.1 Flash Lite Image, including what it calls high-quality video thumbnails. It does not emit a clip, so t2v and i2v are false. The guide points video work at the Veo guide. Audio generation is "Not supported" on every card checked, and image generation does not take audio inputs, so native_audio is false. Reference images are documented. A shot-list field is not, so multi_shot is false. open_source is false. These are Google API ids. A weight file was not on any page opened here.

Nano Banana 2.1, gemini-nano-banana-2.1, is the guide's "primary high-efficiency workhorse" and, on the card, an update to Nano Banana 2. The card says Stable, last updated 2026-10-06 UTC. Gemini 2.5 Flash Image launched on 2025-08-26 under the nickname nano-banana (aka), and the guide calls that id the legacy pioneer. The pricing page says it is deprecated and "will be shut down on October 2, 2026." Accessed 2026-10-20, the warning still says "will be," and the Standard price is still printed. This check did not call the API. Do not read the endpoint as already dark, and do not read it as confirmed still serving.

Imagen is shut down. The guide says: "Imagen is shut down and no longer available through the Gemini API. Use Nano Banana for image generation and editing." The same page calls Veo Google's video model.

This page is not a Flow or CapCut playground. It does not sell a try-on credit and it does not generate an image. Google Flow is the app that names the image model beside Veo. One prompt people already paste, the figurine still, is the recipe on /trends/ai-3d-figurine. That effect is not the model.

When not to use this page. A pure illustration portfolio, a Midjourney-style art contest, or a long bake-off among image models belongs on an image directory or on the contest pages that judge stills as art. This site does not open Midjourney or Flux model pages. If the deliverable is a moving Short with sound, start on a video model.

Specs

Quote the model id. "Nano Banana" covers five public names. Cards and the image guide were read 2026-10-20.

Public name	Model id	Status observed 2026-10-20
Nano Banana 2.1	gemini-nano-banana-2.1	Stable. Card updated 2026-10-06 UTC. Primary workhorse. An update to Nano Banana 2.
Nano Banana 2. Gemini 3.1 Flash Image	gemini-3.1-flash-image	Stable. Card updated 2026-10-09 UTC. Announced 2026-02-26. Previous-generation workhorse. New projects are pointed at 2.1.
Nano Banana 2 Lite	gemini-3.1-flash-lite-image	Stable. Card updated 2026-09-03 UTC. Latest update June 2026. 1K only.
Nano Banana Pro. Gemini 3 Pro Image	gemini-3-pro-image	Stable. Card updated 2026-09-03 UTC. Announced 2025-11-20.
Gemini 2.5 Flash Image	gemini-2.5-flash-image	Card still lists Stable, plus Deprecated gemini-2.5-flash-image-preview. Pricing page: "will be shut down on October 2, 2026." The card does not print that line. Knowledge cutoff June 2025.

Two migration sentences disagree. Do not merge them. The pricing page says "migrate to Gemini 3.1 Flash Image or Gemini 3.1 Flash Lite Image." The image guide's legacy row says the "recommended migration path is to transition to Nano Banana 2 Lite." The new-project line that names 2.1 does not rewrite the 2.5 warning.

Default fields for 2.1, from the 2.1 card:

Field	Observed value
Inputs / output	Text, image, video, and PDF in. Image and text out.
Audio	Generation not supported. Audio inputs are not supported.
Search grounding	Supported.
Thinking	minimal, medium, high. Default medium.
Batch	Supported. Flex and Priority on the card: not supported. No Flex or Priority table in this price section.
Resolutions	1K, 2K, 4K. Default 1K. 0.5K is not supported.
Tokens	Input 131,072. Output 32,768.
References on the card	Up to 14 images. Up to 4 characters. Up to 10 objects.

Video and PDF inputs are on the 2.1, Nano Banana 2, and Lite cards. The guide says video inputs are only supported for those three. Pro and Gemini 2.5 Flash Image take image and text. Nano Banana 2 adds 0.5K and, on its card, aspects 1:4, 4:1, 1:8, and 8:1. Those aspects were not copied onto 2.1. Lite is 1K only ("2K and 4K are unsupported"), with no search grounding, input 65,536 tokens, and output 4,096. Lite thinking is minimal and high, default minimal. Pro thinking is supported without the 2.1 level names. Gemini 2.5 Flash Image has one resolution column, 1:1 at 1024x1024, and no thinking or search grounding.

9:16 sizes come from the image-generation table. The token column is not a price. If you omit aspect_ratio, the model matches the input or makes a 1:1 square. A 1K square is 1024x1024. 16:9 at 1K on 2.1 is 1376x768, also 1120 tokens.

Model	9:16 size	Tokens on that row
Nano Banana 2.1, 1K / 2K / 4K	768x1376 / 1536x2752 / 3072x5504	1120 / 1680 / 2520
Gemini 3.1 Flash Image, 0.5K / 1K	384x688 / 768x1376	747 / 1120
Gemini 3 Pro Image, 1K and 2K	768x1376 and 1536x2752	1120 at both
Gemini 3 Pro Image, 4K	3072x5504	2000
Gemini 2.5 Flash Image	768x1344	1290

Lite has no separate size table. It lists 9:16 and 16:9 among 14 aspect ratios, at 1K only. The other twelve names were not copied. The legacy 9:16 height is 1344, not 1376.

Reference caps on that page ("Use up to 14 reference images"):

Slot	Lite	2.1 and Gemini 3.1 Flash Image	Pro
Objects	Up to 14	Up to 10	Up to 6
Characters	N/A	Up to 4	Up to 5
Style	N/A	N/A	Up to 3

A second paragraph on the same page says gemini-2.5-flash-image works best with up to 3 images, Pro supports 5 with high fidelity and up to 14 in total, and 2.1 plus gemini-3.1-flash-image support 4 characters and 10 objects. Keep that paragraph beside the table. Lite is "Not optimized for multiple reference inputs or multi-turn sequential editing." Its sub-2 second latency line is a card claim, not a measurement from this check.

SynthID applies to every generated image: "All generated images include a SynthID watermark." C2PA was on the Lite card only ("SynthID watermarking (Always On) + C2PA."). Do not paste C2PA onto 2.1, Pro, Nano Banana 2, or Gemini 2.5 Flash Image. Neither mark is a caption.

Prompt notes

No prompt was run for this page. The default response is text and image. Set the aspect for 9:16 or 16:9. On 2.1, 0.5K is outside the card. Thinking defaults to medium. No extra per-image thinking surcharge was printed on the 2.1 image equivalents. The Standard output cell still quotes "$7.50 (text and thinking)" per 1M. A PDF input is not a shot list. A prompt-language allow-list was not on the cards. Not observed. Arena scores were not opened. Not observed. A There's An AI For That blurb says "Released: August 25, 2025" and "built on Gemini 3.6 Flash." The developer post is "AUG. 26, 2025," and the 2.1 card says an update to Nano Banana 2 (Gemini 3.1 Flash Image). Do not copy the other date or the 3.6 line. Vidirect's model index, HTTP 200, "15 models tracked," had no "banana" in the stripped text.

Where to run

A direct request to ai.google.dev on 2026-10-20 returned HTTP 302 toward OAuth. Reader copies of the same URLs returned HTTP 200.

1. Pricing, free-tier column, and the deprecation warning: ai.google.dev/gemini-api/docs/pricing. Footer "Last updated 2026-10-19 UTC." Accessed 2026-10-20.
2. Aspects, SynthID, the Imagen shutdown, and the Veo pointer: image generation. The video link on that page is https://ai.google.dev/gemini-api/docs/video.
3. Cards: 2.1, Nano Banana 2, Lite, Pro, Gemini 2.5 Flash Image.
4. AI Studio chat for 2.1, the pricing Try href: prompts/new_chat?model=gemini-nano-banana-2.1. Unsigned visits were HTTP 302, HTTP 302, then HTTP 200 on Google sign-in, and the continue address keeps the model id. The same new-chat pattern exists for Pro, Nano Banana 2, and Lite. https://aistudio.google.com?model=gemini-nano-banana-2.1 is the welcome page, not the chat. The 2.5 Try href is that welcome form, ?model=gemini-2.5-flash-image. The 2025-08-26 post's preview id is the one the card marks Deprecated. Do not use it.
5. Flow about, direct HTTP 200: flow.google.com/about. Rendered text: "Nano Banana Powerful image generation and precise editing control." The free included list names "Nano Banana Pro" beside "Gemini Omni Flash" and "Veo 3.1." Those names are not one id. Directory page: /tools/google-flow. Credits stay there.
6. CapCut models: www.capcut.com/tools/models. This host got HTTP 307 to www.capcut.cn, then HTTP 200 on the CN homepage. That geo redirect is not a dead English page. A reader copy returned HTTP 200, heading "AI Image & Video Models," card "Nano Banana 2.1 online," dated "Oct 8, 2026," next to Kling, VEO, and Seedream. Reader copies also returned HTTP 200 for Nano Banana Pro and Nano Banana 2. The Pro extract's generate button goes to https://www.capcut.com/ai-design. The Nano Banana 2 page has a "Coming soon" block pointing at itself. The unversioned slug /tools/nano-banana is the reader-reported 404 and is not linked. The 307 and the 404 are different facts.
7. fal 2.1, HTTP 200, text-to-image, 9:16 and 16:9, resolutions 1K, 2K, 4K: google/nano-banana-2.1. Siblings at HTTP 200: nano-banana-2, nano-banana-pro, pro/edit (image-to-image), and fal-ai/nano-banana. Search https://fal.ai/models?q=nano-banana returned HTTP 308 to explore, then HTTP 200, and that HTML links those paths. fal-ai/nano-banana-2-lite had an HTML shell and llms.txt HTTP 404. No Lite price was copied.

Naming posts, each HTTP 200: 2025-08-26, Pro on 2025-11-20, Nano Banana 2 on 2026-02-26.

Multi-channel pricing

Accessed 2026-10-20. Pricing footer: "Last updated 2026-10-19 UTC." Amounts are per 1M tokens unless the cell says per image. Every image row's Free Tier cell is "Not available." "Google AI Studio usage is free of charge in all available regions" is a different sentence, not that cell. Lines that say "5,000 free search requests per month" are search allowances, not free images. Equivalents below are the page's own strings. This page does not multiply tokens into a second SKU.

2.1, gemini-nano-banana-2.1. Standard input "$1.50 (text/image/video)." Output "$7.50 (text and thinking)" and "$30.00 (images)." "Equivalent to $0.0336 per 1K image, $0.0504 per 2K image, and $0.113 per 4K image." Batch input "$0.75 (text/image/video)." Output "$3.75 (text and thinking)" and "$15.00 (images)." "Equivalent to $0.0168 per 1K image, $0.0252 per 2K image, and $0.0567 per 4K image."

Nano Banana 2, gemini-3.1-flash-image. Standard input "$0.50 (text/image)." Output "$3 (text and thinking)" and "$60.00 (images)." "Equivalent to $0.045 per 0.5K image, $0.067 per 1K image, $0.101 per 2K image, and $0.151 per 4K image." Batch input "$0.25 (text, image)." Output "$1.50 (text and thinking)" and "$30.00 (images)." "Equivalent to $0.022 per 0.5K image, $0.034 per 1K image, $0.050 per 2K image, and $0.076 per 4K image."

Lite, gemini-3.1-flash-lite-image. Standard input "$0.25 (text/image/video)." Output "$1.50 (text and thinking)" and "$30.00 (images)." "Equivalent to $0.0336 per 1K resolution image." Batch input "$0.125 (text/image/video)." Output "$0.75 (text and thinking)" and "$15.00 (images)." "Equivalent to $0.0168 per 1K resolution image."

Pro, gemini-3-pro-image. Standard input "$2.00 (text/image), equivalent to $0.0011 per image." Output "$12.00 (text and thinking)" and "$120.00 (images)." "Equivalent to $0.134 per 1K/2K image and $0.24 per 4K image." Batch input "$1.00 (text), $0.0006 (image)." Output "$6.00 (text and thinking)," "$0.067 per 1K/2K image," and "$0.12 per 4K image."

Gemini 2.5 Flash Image. Standard input "$0.30 (text / image)." Output "$0.039 per image." Batch input "$0.15 (text / image)." Output "$0.0195 per image."

Footnotes on the page tie some of those equivalents to square sizes. For 2.1, Standard image output is $30 per 1,000,000 tokens. 1024x1024 is 1120 tokens and $0.0336. 2048x2048 is 1680 tokens and $0.0504. 4096x4096 is 3780 tokens and $0.113. The 9:16 1K and 2K rows are also 1120 and 1680 tokens. The 9:16 4K row is 2520 tokens, not 3780. A 9:16 dollar was not printed. Do not compute one. Nano Banana 2's Standard footnote walks 747 tokens at 0.5K through 2520 tokens and $0.151 at 4K, and it does not restate the Batch token rate. Lite's footnote is 1024x1024 at 1120 tokens and $0.0336. Pro's footnote says image input is 560 tokens or $0.0011, 1K through 2K output is 1120 tokens and $0.134, and 4K output is 2000 tokens and $0.24. The 2.5 footnote says images up to 1024x1024 consume 1290 tokens and $0.039. The 9:16 row is also 1290 tokens. A separate 9:16 dollar was not printed.

2.1, Nano Banana 2, and Lite have no Flex or Priority table, and their cards say both are not supported. Pro and 2.5 cards say the same. The pricing page still prints Flex and Priority for those two. Flex matches Batch on both. Pro Priority input is "$3.60 (text/image)." Output is "$21.60 (text and thinking)" and "$216.00 (images)," with no per-image equivalent. Do not compute one. 2.5 Priority input is "$0.54 (text / image)." Output is "$0.0702 per image." Do not call Flex or Priority supported.

The warning on that page: "Gemini 2.5 Flash Image ( gemini-2.5-flash-image ) is deprecated and will be shut down on October 2, 2026; migrate to Gemini 3.1 Flash Image or Gemini 3.1 Flash Lite Image to avoid service disruption." The 2025-08-26 post already used "$0.039 per image" at 1290 tokens. That post is historical, not a second current SKU.

fal, from llms.txt on 2026-10-20. These are host prices. google/nano-banana-2.1: text/thinking "$1.764 input, $8.823 output" per 1M, image tokens "$1.764 input, $35.294 output" per 1M. No per-image dollar was in the fetched file. Not observed. Those rates are not Google's $1.50 and $30.00. fal-ai/nano-banana-2: "$0.08 per image," with 2K and 4K at "1.5 times and 2 times," 0.5K at "0.75 times," web search plus $0.015, and high thinking plus $0.002. fal-ai/nano-banana-pro and its edit route: "$0.15 per image," 4K at "double," web search plus $0.015. fal-ai/nano-banana and its edit route: "$0.039 per image." That number matches Google's 2.5 Standard cell and remains fal's sentence. The multiplier sentences were not turned into new dollar SKUs.

CapCut dollars were not observed. "Free" in the Pro title, and a models-index phrase "Nano Banana 2.1 free options," are not the API free tier. Flow's "$4.99/mo" is the Google AI Plus sticker beside Flow credits on /tools/google-flow, verified there 2026-10-17. It is not a per-image rate. On 2025-09-01, @GeminiApp at 2025-09-01T22:41:29.000Z said image creation and editing was "free for everyone in the Gemini app." That sentence is dated. It is not the 2026-10-20 API column. Leave Veo seconds on Veo. Do not divide a Flow credit into a still.

Workflow and platform fit

Covers and thumbnails. Build the still, then finish on captions and covers. Video-to-image on 2.1, Gemini 3.1 Flash Image, and Lite is the guide's route to a thumbnail from footage. Pro and Gemini 2.5 Flash Image do not list video input. Set 9:16 yourself. At 1K that is 768x1376 on 2.1, on Gemini 3.1 Flash Image, and on Pro, and 768x1344 on the legacy id. 16:9 at 1K on 2.1 is 1376x768. Lite cannot supply a 4K master. SynthID rides on the file. The caption and the crop stay in CapCut. Cover layout, if any, stays on captions and covers. Canva: a Nano Banana id inside Canva was not observed.

Storyboard stills. The control is the reference table, not a timeline. On 2.1 that is 4 characters and 10 objects, with no style slot. Pro has the style slot and the separate "5 with high fidelity, up to 14" sentence. Lite is a weak multi-reference editor on its own card. Faceless and UGC ads can use the still as a cover or a beat. They do not add a shot list.

First frame. Pass the image to image-to-video. When the video model is Google's, the next page is Veo. Duration and the per-second price stay there. Veo lists 9:16 and 16:9, and those ratios are on the 2.1 size table. A shared pixel size was not printed on either page. Do not treat 768x1376 as a Veo frame. This still has no soundtrack.

Shorts and TikTok. The fit is the vertical cover, the reference board, or the first frame, at a printed image price. The moving 9:16 file is a video model. Batch rows are the printed path for a set of covers, and they are paid. The free-tier cell is Not available. The $0.0336 line is the printed 2.1 1K equivalent, not a 9:16 4K price.

Pros

• New work has one named default, gemini-nano-banana-2.1.
• 9:16 sizes are printed for 2.1 at 1K, 2K, and 4K, and 16:9 at 1K is 1376x768.
• Video-to-image is a documented thumbnail path from footage on 2.1, Gemini 3.1 Flash Image, and Lite.
• Character, object, and, on Pro, style slots are numbered.
• SynthID is on every generated image. C2PA is written on the Lite card, and only there in this pass.
• Standard and Batch image equivalents are printed for each SKU, including a shared 1K/2K price on Pro.
• The Imagen shutdown and the Veo pointer are on the same guide.

Cons

• Text-to-video, image-to-video, native audio, and multi-shot are false. A Short still needs another model and an editor.
• The API free tier is "Not available" on every image row. The 2025 app sentence and the AI Studio "free of charge" sentence are other claims.
• Five names and two 2.5 migration sentences bill the wrong row if the brief says only "Nano Banana."
• 9:16 is not the default. The legacy 9:16 height is 1344, not 1376.
• The printed $0.113 follows 4096x4096 at 3780 tokens. The 9:16 4K row is 2520 tokens, and no dollar was printed for it.
• On 2026-10-20 the shutdown line still said "will be," and the API was not called. Live or dark is not established.
• Pro Flex and Priority dollars sit on a card that says those modes are not supported. Priority has no per-image equivalent.
• fal's 2.1 file has no per-image dollar. CapCut dollars were not observed. The unversioned CapCut slug 404'd.
• Lite is 1K only and is not the multi-reference editor. Arena scores were not opened. Not observed.

Alternatives

• Veo is the video model the image guide names next. Use it when the still has to move or the clip needs generated sound. It does not replace the still. Pixel sizes were not matched across the pages.
• Kling and Seedance are other video models on this directory. A Nano Banana cover is not one of their clips. Their billing unit on this page is not observed.
• Runway Gen and /tools/runway are on-site video-model alternatives. Nano Banana dollars on Runway: not observed.
• Luma Ray and MiniMax Hailuo are video models. Use them for the moving shot. Keep this page on the still.
• Imagen is shut down on the Gemini API. Use Nano Banana. This page does not open an Imagen slug. A directory entry that still lists Imagen is not the API status.
• CapCut's models index lists Kling, VEO, and Seedream beside the Nano Banana 2.1 card. Those are CapCut product surfaces.

Midjourney and Flux are not opened. This directory has no model page for them. An art-contest comparison belongs on an image directory.

Related tools

• /tools/google-flow holds the plan stickers. Flow's about page names Nano Banana for images and names Nano Banana Pro on the free included list, beside Veo 3.1. The Google AI Plus sticker is not a per-image rate for 2.1.
• /tools/capcut is the timeline for captions, the crop, and a later beat. The models index and the Pro and Nano Banana 2 pages are the host surfaces above. Dollars there were not observed. HTTP 307 is the geo redirect. The unversioned /tools/nano-banana path is the 404 and is not linked.
• /tools/canva. Cover layout, if any, stays on captions and covers. A Nano Banana id inside Canva was not observed.
• /tools/krea — Nano Banana on Krea: not observed in the 2026-10-20 source pack.
• /tools/runway is an on-site video-model alternative, with Runway Gen. Nano Banana dollars on Runway: not observed.
• /trends/ai-3d-figurine is one still effect run in a Nano Banana chat. This page does not restate the recipe.
• Image-to-video is the handoff matrix. Veo is the Google video page.

FAQ

Is Nano Banana a video model?

The cards output an image and text. Audio generation is not supported, so a thumbnail from this line has no soundtrack. Text-to-video and image-to-video are false. Video-to-image, on 2.1, Gemini 3.1 Flash Image, and Lite, makes a new still from footage. The image guide sends clip generation to Veo. You can pass the still to image-to-video or Veo. That step does not change the flags on this page.

Which id is the default, and is the API free tier open?

For new projects the image guide says gemini-nano-banana-2.1. The Try link is the new-chat URL with that id. Pro, Lite, Nano Banana 2, and Gemini 2.5 Flash Image are the other Stable or legacy ids in the table above. The Free Tier cell is "Not available" on all five, on the pricing page last updated 2026-10-19 UTC. The AI Studio "free of charge" sentence, the 2025-09-01 Gemini app sentence, CapCut's "Free" title, and the search-request allowance are not that cell.

What is the 9:16 size, and does it match Veo?

On 2.1 the image page prints 768x1376 at 1K, 1536x2752 at 2K, and 3072x5504 at 4K, and only when the aspect is set. The default matches the input or makes a square. Pro's 1K 9:16 row is also 768x1376. Gemini 2.5 Flash Image's row is 768x1344. Lite has no separate pixel table and is 1K only. Veo lists 9:16 and 16:9. A matching pixel size was not on either page. Not observed.

Has Gemini 2.5 Flash Image shut down, and which migration line do I follow?

The pricing page, updated 2026-10-19 UTC and read 2026-10-20, says the id "will be shut down on October 2, 2026" and still prints "$0.039 per image." The card still lists it as Stable. This check did not call the API, so this page does not say the endpoint is dark and does not say it is live. Pricing says migrate to Gemini 3.1 Flash Image or Gemini 3.1 Flash Lite Image. The image guide's legacy row says Nano Banana 2 Lite. For a new thumbnail, the guide's new-project line is 2.1, which is a separate sentence from both.

Is SynthID the same as C2PA, and where are the arena scores?

SynthID is the all-images watermark. C2PA was stated on the Lite card only. Arena pages were not opened. Not observed. Flow's about wording is not a rank. A third-party "Gemini 3.6 Flash" line conflicts with the official 2.1 card and is not a spec.