Skip to content

Nano Banana vs FLUX.2 Pro for Product Images

Nano Banana 2.1, Nano Banana Pro or FLUX.2 [pro] for product images: official prices, reference limits, sizes, editing, text, and a $3 test to run.

Get an API key, pay as you go from $10One key for video, image and voice models. Each call shows its price before it runs.
Get an API key
By Robin MFounder of Aitachyon

Published

Updated

Reading time7 min

Read inEnglishFrançais

You have a product photo and need more out of it: clean packshots, a few lifestyle scenes, a version where the label still reads. The image models people suggest for that are Google's Nano Banana 2.1 and Nano Banana Pro, and Black Forest Labs' FLUX.2 [pro]. This is a comparison of Nano Banana vs FLUX for product images, built only on what the two labs publish, checked on 1 November 2026: price per image, reference-image limits, output sizes, editing and text. It ends with a test you can run on your own product for under $3.

My short answer, argued below: start edits from a reference photo on Nano Banana 2.1, move to FLUX.2 [pro] when a color must match a hex code or when you need a model snapshot that will not change under you, and pay for Nano Banana Pro only for packaging text and 4K hero images. Then let your own test overrule me.

What the names mean on 1 November 2026

"Nano Banana" is a family. On Google's Gemini API pricing page, the current model is Nano Banana 2.1 (gemini-nano-banana-2.1), an update to Nano Banana 2 (gemini-3.1-flash-image) at half its price in 1K and 2K. Nano Banana Pro is gemini-3-pro-image. The original Nano Banana (gemini-2.5-flash-image) is deprecated and shuts down on 15 March 2027, so start nothing new on it.

Google's image generation guide calls 2.1 "the primary high-efficiency workhorse model for image generation and conversational editing" and recommends it for all new projects. Pro is the one "optimized for professional asset production".

FLUX.2 [pro] is Black Forest Labs' production model. It has two API endpoints: flux-2-pro, a fixed snapshot kept for reproducibility, and flux-2-pro-preview, where improvements land first, per the release notes. The same notes show the lab shipped FLUX 3 Image on 1 October 2026. This comparison sticks to FLUX.2 [pro].

Price per image, from the official pages

Google bills images as output tokens, converted to a price per image. Black Forest Labs bills by the megapixel: its pricing page lists FLUX.2 [pro] at $0.03 for the first megapixel and $0.015 for each additional one, and the API pricing doc lists editing from $0.045, because a reference image counts as megapixels too. Google's 2K is 2048x2048 and BFL's guide counts 2048x2048 as 4MP, which lines the columns up:

Model (list price)1K, 1024x10242K, 2048x20484K, 4096x4096
Nano Banana 2.1$0.0336$0.0504$0.113
Nano Banana Pro$0.134$0.134$0.24
FLUX.2 [pro], text to image$0.03$0.075not offered
FLUX.2 [pro], edit with one 1MP reference$0.045$0.09not offered

Three things move these numbers.

  • Thinking. Gemini 3 image models reason before they draw, Google's guide says this cannot be disabled in the API, and thinking bills as text output ($7.50 per million tokens on 2.1, $12 on Pro). Your invoice will sit above the per-image figure.
  • Batch. Google's Batch API halves the price for a turnaround of up to 24 hours: $0.0168 per 1K image on 2.1, $0.067 on Pro. BFL's doc says a batch simply multiplies the base cost by the number of images.
  • Reference size. FLUX.2 [pro] bills input megapixels. A photo straight off a phone carries many more than you need, so resize references to about 1MP before you send them.

Now a real catalog: 50 SKUs, one packshot and one lifestyle scene each, three candidates per shot, is 300 edits at 1K. Nano Banana 2.1: 300 x $0.0336 = $10.08, or $5.04 in batch. FLUX.2 [pro] with one 1MP reference: 300 x $0.045 = $13.50. Nano Banana Pro: 300 x $0.134 = $40.20, or $20.10 in batch, plus $0.0011 per reference image. The gap is a few tens of dollars; retries and rejected shots cost more, and the test at the end measures them.

Reference images and output sizes

For product work, the reference limit decides what fits in one call: front, back, a label close-up, the cap, a second SKU for a range shot.

  • Nano Banana 2.1: up to 10 object images "with high-fidelity" plus up to 4 character images, 14 in total. Output at 1K (the default), 2K or 4K. The 512px size exists on Nano Banana 2 but not on 2.1.
  • Nano Banana Pro: up to 6 object images, 5 character images and 3 style references. Output at 1K, 2K or 4K.
  • FLUX.2 [pro]: up to 8 reference images through the API. The prompting guide sets a 9MP budget for input plus output: 8 references at 1MP output, 7 at 2MP, fewer as the output grows. Output runs from 64x64 to 4MP, in multiples of 16, with up to 2MP recommended.

Two practical consequences. Only Nano Banana renders 4096x4096, so a print-size hero image is a Google job ($0.113 on 2.1, $0.24 on Pro). And the multiple-of-16 rule means FLUX.2 cannot output the 1080x1350 Instagram portrait directly, since neither number divides by 16. Ask for 1088x1360, which is exactly 4:5, and scale down.

Get an API key, pay as you go from $10One key for video, image and voice models. Each call shows its price before it runs.
Get an API key

Editing from a reference photo

This is the job that matters for products: the bottle has to stay your bottle.

Google describes editing as a conversation. You send the image and describe the change, and "the model will match the original image's style, lighting, and perspective". There is no mask parameter: the guide calls its method semantic masking, with prompts of the form "change only the [specific element]". Multi-turn chat is the recommended way to iterate, which suits instructions like "move the bottle left, keep everything else".

Black Forest Labs puts generation and multi-reference editing in the same model. Its FLUX.2 launch post claims "sharper textures, and more stable lighting suitable for product shots" and editing at up to 4 megapixels: a vendor claim to check in your own test. Two documented features have no equivalent I could find in Google's guide:

  • Hex colors. The prompting guide says to tie the code to an object ("The car is #FF0000" works better than "use red #FF0000 in the image"). That is how you get a backdrop or a box in your exact brand color.
  • A pinned snapshot. flux-2-pro does not change, so a shot made in March matches one made in October.

Text on packaging

Both labs claim reliable text. Google's Nano Banana Pro announcement calls it "the best model for creating images with correctly rendered and legible text directly in the image" and says it can write in several languages or translate existing text. The guide adds that 2.1 renders text better than Nano Banana 2. Black Forest Labs says complex typography with "legible fine text now work reliably in production", and its prompting guide asks for the exact words in quotes, then placement, style, size and a hex color for brand text.

My opinion, for real products: do not ask any model to write your label. Edit around the label already in your photo. When text has to be generated (a headline on a lifestyle shot, a localized version), check every character, units and numbers included. A claim about legibility does not promise that "200 ml" stays "200 ml".

A test to run on your own product

One product, four jobs, three models, three candidates each: 36 images.

  1. Packshot. Your photo, resized to about 1MP, as the only reference. Product on plain white, soft shadow, 1K.
  2. Lifestyle scene. Same reference, the product in a setting your buyer knows, 1K.
  3. Label check. A scene where the real label faces the camera and must stay readable, at 2K (4MP on FLUX.2).
  4. Brand color. A backdrop or prop in your exact hex code, 1K. Tie the hex to the object for FLUX.2, and give Nano Banana the same sentence to see what it does with it.

At list prices, Nano Banana 2.1 costs 9 x $0.0336 + 3 x $0.0504 = $0.45. FLUX.2 [pro] edits cost 9 x $0.045 + 3 x $0.09 = $0.68. Nano Banana Pro costs 12 x $0.134 = $1.61. About $2.74 in all, plus Gemini's thinking tokens.

Score every image pass or fail on five checks: shape and proportions match the photo; label text correct to the character; color within tolerance of the hex, checked with an eyedropper; no invented badges, claims or ingredients; usable without retouching. Then divide each model's spend by its passes. If, say, Nano Banana 2.1 passes 6 of its 12, a usable image costs $0.4536 / 6, about $0.076. If FLUX.2 [pro] passes 9 of 12, it costs $0.675 / 9 = $0.075, and the tie goes to whichever fails in ways you can fix. Route each job type to its winner and write the routing down, so nobody picks a model by habit.

Running both without two accounts

Aitachyon, the studio I run, has Nano Banana 2.1, Nano Banana Pro (for 4K) and FLUX.2 [pro] behind one API key and a hosted MCP server, billed from one balance. On 1 November 2026 a call costs $0.10 per image on Nano Banana, $0.09 to $0.135 on FLUX.2 [pro] and $0.40 to $0.72 for Nano Banana Pro at 4K. That is three times the lab price on images, as the MCP page states, so running the test on your own Google and Black Forest Labs accounts is cheaper. The case for one key is one ledger and one place to switch models.

If you would rather not run the test at all, product photos is the done-for-you version: $49 for 10 images made from your own product photos, studio packshots and lifestyle scenes in square, 4:5 and 16:9, with three test shots to approve within one business day or a full refund.

Robin MFounder of Aitachyon
How the studio works

One key for every model.

Video, image and voice models behind one API key and one MCP server. Pay as you go from $10: each call shows its price before it runs, and a failed job is refunded to the cent.

Get an API key

How the API and the MCP work