Эта страница есть на русском

Esta página también está en español

此页面也有中文版

Get started

Model comparisons

The best image generation APIs: 10 models in one table

Ten Nano Banana and GPT Image models in one table: resolutions, price per image, the task each model suits, and its average wait where measured.

Rendergrid editorsAbout the editors

Telegram

In brief

Rendergrid’s catalog lists ten image models: five Nano Banana models from Google and five GPT Image models from OpenAI. The table below sets out their resolutions, aspect ratios, prices per image, and the task each model suits, and each model’s recent average wait is listed where Rendergrid has measured it. This post ranks none of them: three example images show what a model returns for its own prompt, not which is better.

How do the ten models compare in one table?

The table has one row for each model in Rendergrid’s catalog. The models available to your account, with your own prices, are listed by GET /models.

Ten image models: maker, aspect ratios, resolutions with the price per image, and the task each suits
ModelMakerAspect ratiosResolutions and price per imageSuited toSource
Nano Banana ProGoogle10 and auto1K: $0.04, 2K: $0.04, 4K: $0.07Creating images and editing reference-based scenesCatalog, pricing, docs
Nano Banana 2Google14 and auto1K: $0.025, 2K: $0.04, 4K: $0.06Illustrations, product imagery, and reference-based compositionsCatalog, pricing, docs
Nano Banana 2.1Google14 and auto1K: $0.02, 2K: $0.03, 4K: $0.04Editing with up to 14 reference images in one requestCatalog, pricing, docs
Nano Banana FlashGoogle10 and auto1K only: $0.015Generating and editing images when 1K is enoughCatalog, pricing, docs
Nano Banana 2 LiteGoogle101K only: $0.015Workflows that set the format and use 1K outputCatalog, pricing, docs
GPT Image 2OpenAI101K only: $0.02Generating and editing images at a fixed 1KCatalog, pricing, docs
GPT Image 2.5OpenAI131K only: $0.025A fixed 1K with 9:21, 7:5, and 5:7 among its ratiosCatalog, pricing, docs
GPT Image 2 HDOpenAI101K: Pricing, 2K: Pricing, 4K: PricingIntegrations that need 2K or 4K outputCatalog, pricing, docs
GPT Image 2.5 SunburstOpenAI81K: Pricing, 2K: Pricing, 4K: PricingGenerating and editing images with precise handling of fine detailCatalog, pricing, docs
GPT Image 2.5 FlareOpenAI81K: Pricing, 2K: Pricing, 4K: PricingGenerating and editing images at 1K, 2K, or 4KCatalog, pricing, docs

The task column follows the short model descriptions in Rendergrid’s documentation. A model whose description says no more than its resolutions and aspect ratios, such as GPT Image 2.5 Flare, gets no more here.

Which resolutions and aspect ratios can you request?

Resolution is the model’s size level, not a number of pixels: the resolution field takes 1K, 2K, or 4K, and width and height are not accepted. Six models list all three. The other four, marked “1K only” in the table, are limited to 1K, and a 2K or 4K request to one of them is rejected with 400 before any charge. Leave resolution out and the image is made, and paid for, at the default: 2K for Nano Banana Pro, Nano Banana 2, and Nano Banana 2.1, and 1K for the other models.

The ten standard aspect ratios are 1:1, 16:9, 9:16, 4:3, 3:4, 2:3, 3:2, 5:4, 4:5, and 21:9. Every model takes them except GPT Image 2.5 Sunburst and GPT Image 2.5 Flare, which have no 5:4 or 4:5. The counts in the table include each model’s extra ratios. Leave aspect_ratio out and it is 1:1, not auto. Where a model has auto, the model chooses the shape, and without a reference image that shape is not guaranteed. The models documentation has the full rules.

What does an image cost on each model?

The price is per image and depends on the model and the resolution. It is charged when the request is accepted, and a failed request is free: the amount goes straight back to your balance. The table lists every price. For example, one Nano Banana Flash image at 1K costs $0.015, and one Nano Banana Pro image at 4K costs $0.07. To plan a budget by count, divide $10 by a price and round down.

Your account may have its own prices. In the GET /models response, read a resolution’s price from variants; the root price is the cheapest resolution’s, not your image’s.

For one model in detail, see the Nano Banana 2.1 price article and the Nano Banana Pro price article.

How long does each model take?

There is no promised time: it depends on the model and the load. Poll a task no more often than once every 5 seconds, and if your own waiting limit runs out, save the id and continue later. A model’s line below appears only when Rendergrid has measured it, and it gives a recent average, not a promise for your requests.

Nano Banana Pro: Average wait time over the last 7 days: 39 s. Nano Banana 2: Average wait time over the last 7 days: 34 s. Nano Banana 2.1: Average wait time over the last 7 days: 40 s. Nano Banana Flash: Average wait time over the last 7 days: 29 s. Nano Banana 2 Lite: Average wait time over the last 7 days: 32 s. GPT Image 2: Average wait time over the last 7 days: 43 s. GPT Image 2.5: Average wait time over the last 7 days: 35 s.

What do the three example images show?

Each example uses a different prompt and model, and the resolution and aspect ratio are not the same across them, so they do not compare the models. Each prompt was sent once, and the image is shown as it came back.

Prompt for the folded paper

Fill a square composition with broad folds of cream paper, like gentle waves. Directional side light separates bright planes from long soft shadows and reveals fine paper fibers. Avoid sharp cut-off corners. Limit the palette to cream and warm gray. No lettering, logos, people, or watermarks.

1K · 1:1 · one attempt

Nano Banana Flash

nano-banana-flash · 1K · 1:1 · $0.015

Nano Banana Flash, 1K, 1:1, one attempt. Cream paper folds fill the square, and their edges are irregular, with no sharp cut-off corners. Fine fibers are visible across the central fold, which curls into an open cone. Light from the right brightens the upper folds, and soft shadows deepen to gray-brown at the lower left. The palette is cream and gray-brown, and there is no lettering, logo, person, or watermark. Departure from the prompt: the central fold curls tighter than the gentle waves it asked for.

Prompt for the carafe and glass

Place a clear ribbed carafe and one drinking glass on a terracotta shelf. The glass sits slightly forward and to the side. Morning light passes through the glassware, casting soft bands on the wall. Show the complete carafe and keep the arrangement spacious. No lettering, logos, people, or watermarks.

2K · 1:1 · one attempt

Nano Banana 2.1

nano-banana-2.1 · 2K · 1:1 · $0.03

Nano Banana 2.1, 2K, 1:1, one attempt. A clear ribbed carafe stands left of center on a pitted terracotta shelf, complete from rim to base. A ribbed drinking glass sits slightly in front of it and to its right, as the prompt asked. Both vessels hold some water, and the glass is ribbed like the carafe; the prompt mentioned neither. Light from the left throws a large shadow of the carafe onto the cream wall, with pale bands of light across it, and a smaller shadow is cut off by the right edge. The arrangement is spacious, and there is no lettering, logo, person, or watermark. Departures: the prompt asked for soft bands on the wall, but the carafe’s shadow on the wall is a silhouette with a crisp edge, and it did not mention the darker area with a defined edge that covers the left of the wall or the dark shadow on the left end of the shelf.

Prompt for the poster

Design a portrait botanical poster on lightly textured off-white paper. One large dark green leaf silhouette runs diagonally beside the lettering. Use “A quiet garden” as the main heading and “Leaves and light” as a smaller line. Keep the type clear, with asymmetrical spacing and generous margins. Include only the specified lettering, with no logos, people, or watermarks.

1K · 2:3 · one attempt

GPT Image 2.5

gpt-image-2.5 · 1K · 2:3 · $0.025

GPT Image 2.5, 1K, 2:3, one attempt. Both lines of lettering match the prompt exactly: “A quiet garden” as the heading, broken onto two lines, and “Leaves and light” as the smaller line. There is no other text, logo, person, or watermark. The heading sits at the upper left and the smaller line at the lower right, with wide margins around both. Departures from the prompt: it asked for one large leaf silhouette, but the frame shows a sprig of six leaves on one stem, each with visible veins and shading rather than a flat silhouette; and it asked for generous margins, but the sprig’s right-hand leaf tip is about 5% from the right edge and its stem ends about 6% from the bottom edge.

None of the three shows people, logos, or a watermark, and each departs from its prompt in at least one respect, named in its caption.

Which model fits which task?

The tasks come from each model’s description in Rendergrid’s documentation, and the list ranks nothing.

  • Creating images and editing reference-based scenesNano Banana ProAt 1K, 2K, or 4K.
  • Illustrations, product imagery, and reference-based compositionsNano Banana 2In standard, panoramic, or tall formats.
  • Editing with many reference imagesNano Banana 2.1Up to 14 in one request.
  • Generating and editing images at 1KNano Banana FlashFor workflows that need no more.
  • A fixed format at 1KNano Banana 2 LiteYou set the aspect ratio yourself: there is no auto.
  • Generating and editing images at a fixed 1KGPT Image 2With the ten standard aspect ratios.
  • More aspect ratios at 1KGPT Image 2.5Thirteen in all, including 9:21, 7:5, and 5:7.

Where does this comparison not fit?

  • Three images are not a benchmark. Each prompt was sent once, to one model. Another attempt can look different, and nothing here is a rate, a speed, or a reliability figure.
  • Seven of the ten models made no image here. The three examples do not share one resolution and aspect ratio.
  • The table ranks nothing. It repeats the catalog and the documentation and does not judge a model’s images. Try your own prompts.
  • Waits are recent averages. A model’s wait appears only when it has been measured, and no time is promised.
  • Your account may differ. It may have its own prices or a shorter list, and GET /models returns them.

A failed request is free

Open the catalog and pick a model

Look through the catalog and open the page of the model that fits your task. If a request fails, the amount goes straight back to your balance.

Browse the models

Questions and answers

Which image generation model is the best?
This post ranks none of the ten models. The right choice depends on your task, the resolution and aspect ratio you need, and the price per image. Three example images do not make a ranking, so try your own prompts.
How many image models does Rendergrid list?
Ten: five Nano Banana models from Google and five GPT Image models from OpenAI. GET /models lists the models available to your account, with your prices.
What does an image cost?
It depends on the model and the resolution, and the table above lists every price. For example, one Nano Banana Flash image at 1K costs $0.015.
How long does an image take?
No time is promised: it depends on the model and the load. Poll a task no more often than once every 5 seconds and limit your waiting time. A model’s recent average is in the section on waits only when Rendergrid has measured it.
Do I pay for a failed request?
No. A failed request is free: the amount goes straight back to your balance.

Sources

  1. Rendergrid pricingThe price per image for every model and resolution.
  2. Models and parametersResolutions, aspect ratios, defaults, and GET /models.
  3. Tasks and statusesPolling, and why no waiting time is promised.
  4. Rendergrid model descriptionsThe short description of each model, behind the task column and list.
  5. The three example imagesOne attempt each, with the prompt printed beside the image.
  6. Rendergrid statisticsWhere a model’s average wait appears here, it is Rendergrid’s own recent measurement.

Related reading

All articles