日本語版はまだないため、英語で表示しています。

Best AI model for image generation (November 2026): GPT-Image-2.6, Nano Banana 2.1, Midjourney V8.2, FLUX 3, and the open models

Quick answer

The best AI image model depends on the job: OpenAI's GPT-Image-2.6 (November 4, 2026) makes 4K images, SVG, and animated GIFs; Google's Nano Banana 2.1 costs $0.0336 for a 1K image and goes to 4K; Midjourney V8.2 is the pick for a distinctive look, from $10 a month; and FLUX 3 Image Klein, Qwen-Image-2.1, and MiniMax H3 Image have open weights you can run locally. Every model, price, and licence line is from its maker's own page, read on November 18, 2026.

Publish my HTML page free

Free plan: no credit card, 1 published page, 50 form entries a month.

On this page

You need images from an AI model and want to know which one to use this month. The answer changes with the job: a product shot, a poster with text on it, a logo as SVG, a hundred thumbnails through an API, or a model you run on your own GPU.

This page lists the current image models from each maker's own model page, pricing page, and terms, read on November 18, 2026 (Japan time). Where a maker's page doesn't state a fact, the table says so.

The best AI image models right now: comparison table

Current AI image generation models, read on November 18, 2026. Prices are US dollars.

MakerModel and datePriceMaximum resolutionEditing and text, as the maker statesAPI or appCommercial use, per the maker's terms
OpenAIGPT-Image-2.6 Sunburst and Flare, November 4, 2026$30 per million image output tokens, $8 per million image input tokens3840 px on the long edge, up to 8,294,400 pixels (3840x2160)"adjustable complexity, SVG generation, and animated GIF generation"; markup-guided editing; transparent backgroundsOpenAI APIThe customer "owns all Output"
GoogleNano Banana 2.1, October 6, 2026$0.0336 per 1K image, $0.0504 per 2K, $0.113 per 4K4K (4096x4096)"improved visual quality, text rendering, multi-turn consistency, and Google Search grounding"; up to 14 reference imagesGemini API"Google won't claim ownership over that content."
MidjourneyV8.2, July 24, 2026Plans: $10, $30, $60, or $120 a month2048 x 2048 in HD at 1:1A new Edit Model, Personalizationmidjourney.com and Discord. Midjourney's pages mention no API"You own all Assets You create"; companies above $1,000,000 a year in revenue need Pro or Mega
Black Forest LabsFLUX 3 Image. BFL's pricing page doesn't give a date$0.048 per 1K image, $0.10 per 2K, $0.607 per 4K4K"edit a finished image one box at a time"BFL APIBFL's self-hosting licence: "we make no ownership claim over those outputs"
Black Forest LabsFLUX 3 Image Klein, October 15, 2026$0.009 per 1K image, $0.014 per 2K, $0.036 per 4K4K"text-to-image, editing and multi-reference generation in a model you can run on your own hardware"BFL API, or the weights on Hugging FaceCommercial self-hosting needs a BFL licence tier
xAIGrok Imagine Image 2.0. xAI's docs don't give a date"Starting at $0.02 / image"2KGeneration, editing, and multi-image editingxAI APIEnterprise terms: the customer owns Output
IdeogramIdeogram 4.5. Ideogram's pages don't give a datePlans: Plus $20, Pro $60 a month2K"crisp typography", Precise Edit, reference-image editsIdeogram's app, and an APIIdeogram's terms claim no ownership of what you make
Alibaba (Qwen)Qwen-Image-2.1, repository created September 14, 2026Royalty-free licenceQwen's card doesn't state a maximumTransparent RGBA output, up to 10 reference imagesWeights on Hugging FaceQwen Research License: "FOR NON-COMMERCIAL PURPOSES ONLY"; commercial use needs a separate licence from Qwen. Qwen-Image and Qwen-Image-2512 are Apache 2.0
MiniMaxMiniMax H3 Image, repository created November 3, 2026Royalty-free licenceMiniMax's card doesn't state a maximum"state-of-the-art open-source model for image generation and editing"Weights on Hugging FaceMiniMax H3 Image Community License: commercial use allowed; above $1 million in annual revenue, or to host it for others, you need MiniMax's written authorization

How the prices compare

Price per image through each maker's API, read on November 18, 2026.

Model1K image2K image4K image
FLUX 3 Image Klein$0.009$0.014$0.036
Grok Imagine Image 2.0From $0.02xAI's page gives a starting price onlyNot offered
Nano Banana 2.1$0.0336$0.0504$0.113
FLUX 3 Image$0.048$0.10$0.607
GPT-Image-2.6 Sunburst and FlareBilled by token: $30 per million image output tokens

Two notes on reading it:

  • OpenAI bills by token. Sunburst and Flare have the same token rates, and OpenAI adds: "Equal token rates don't mean equal cost per image." Sunburst is for "detailed visual work" and Flare for "fast iteration". Batch jobs are half price.
  • Midjourney and Ideogram bill by plan. Midjourney is $10, $30, $60, or $120 a month ($8, $24, $48, or $96 a month billed yearly) Midjourney's plans page: "For unlimited image generations with Relax Mode, subscribe to the Standard, Pro, or Mega Plan." Ideogram Plus is $20 a month ($180 billed yearly) with 2,400 priority credits, and Pro is $60 ($504 billed yearly) with 7,500.

Which image model to pick for which job

Each row uses only what the maker's own page states.

Your jobPickThe stated fact
A logo or icon as a vector fileGPT-Image-2.6SVG generation
A short animationGPT-Image-2.6Animated GIF generation
A poster, ad, or thumbnail with words on itIdeogram 4.5, Nano Banana Pro, or Nano Banana 2.1Ideogram: "crisp typography". Google calls Nano Banana Pro the "premium choice" to "render high-fidelity text"
The largest imageGPT-Image-2.6, Nano Banana 2.1, or FLUX 3 Image3840x2160, 4096x4096, and 4K
Thousands of images through an API at the lowest priceFLUX 3 Image Klein$0.009 per 1K image
One character or product across many imagesNano Banana 2.1 or Qwen-Image-2.1Up to 14 and up to 10 reference images
A distinctive, art-directed lookMidjourney V8.2The V8.2 release improved aesthetics and Personalization
A cutout with no backgroundGPT-Image-2.6 or Qwen-Image-2.1Transparent backgrounds; transparent RGBA output
Images that cite current factsNano Banana 2.1Google Search grounding
Private work on a subscriptionMidjourney Pro or MegaStealth Mode "keeps your images and videos private"
A model on your own GPUFLUX 3 Image Klein, Qwen-Image-2.1, or MiniMax H3 ImageOpen weights on Hugging Face

Best AI model for image generation and editing

Editing is now part of the main model at every maker:

  • OpenAI. The API has v1/images/generations and v1/images/edits, and GPT-Image-2.6 adds markup-guided editing: you mark the region to change.
  • Google. Nano Banana 2.1 keeps "multi-turn consistency", so you edit by asking for the next change, with up to 14 reference images.
  • Midjourney. V8.2 shipped a new Edit Model that replaces Omni Reference, Character Reference, and Retexture.
  • Ideogram. Precise Edit and reference-image edits in Ideogram 4.5.
  • xAI. Editing and multi-image editing in grok-imagine-image-2.0.
  • Open weights. MiniMax describes H3 Image as built "for image generation and editing", and Qwen-Image-2.1 takes up to 10 reference images.

Best open-source AI image models to run locally

Open-weight image models on Hugging Face, read on November 18, 2026.

ModelMakerRepository dateLicenceNotes from the card
MiniMax-H3-ImageMiniMaxNovember 3, 2026MiniMax H3 Image Community License: royalty-free, commercial use with conditions"state-of-the-art open-source model for image generation and editing". First on Hugging Face's trending text-to-image list on November 18
FLUX 3 Image KleinBlack Forest LabsOctober 13, 2026BFL's own licence; commercial self-hosting needs a licence tier"A 5.6B parameter open weights model"
Qwen-Image-2.1AlibabaSeptember 14, 2026Qwen Research License: non-commercial onlyA 7B visual generation component, transparent RGBA, up to 10 reference images
Qwen-Image-2512 and Qwen-ImageAlibabaDecember 30, 2025 and August 2, 2025Apache 2.0Open download, no sign-in
FLUX.1 [schnell]Black Forest LabsJuly 31, 2024Apache 2.0Download after you accept the access form on Hugging Face

For commercial work with no licence to buy, the Apache 2.0 models are the direct route: Qwen-Image, Qwen-Image-2512, and FLUX.1 [schnell]. For the newest open models, read the licence file in the repository before you ship: each of the three newest has its own.

Best free AI image model

  • Free for commercial use: Qwen-Image and Qwen-Image-2512 are Apache 2.0 with an open download on Hugging Face, and FLUX.1 [schnell] is Apache 2.0 behind Hugging Face's access form.
  • Free with conditions: MiniMax H3 Image's licence is royalty-free and allows commercial use; you display "MiniMax-H3-Image" in a commercial product, and need MiniMax's written authorization above $1 million in annual revenue or to host it for others.
  • Free for research only: Qwen-Image-2.1's Qwen Research License is royalty-free "FOR NON-COMMERCIAL PURPOSES ONLY"; commercial use needs a separate licence from Qwen.

Licences read on Hugging Face on November 18, 2026.

  • In a chat app: for what each chat plan includes, see ChatGPT vs Claude vs Gemini vs Grok subscriptions.
  • Lowest paid price: FLUX 3 Image Klein at $0.009 for a 1K image, and Google's Nano Banana 2 Lite, which Google lists as its cheapest image model.

Recraft's pricing page states one more rule worth knowing before you use a free plan for work: "Images generated on the Free plan are owned by Recraft", while "Paid plans grant you full ownership and commercial rights".

Can you use AI images commercially?

Each maker's terms answer it differently:

  • OpenAI: under the Services Agreement, the customer "owns all Output".
  • Google: the Gemini API terms (effective March 23, 2026) say "Google won't claim ownership over that content." Every image carries a mark: "All generated images include a SynthID watermark."
  • Midjourney: "You own all Assets You create", and a company with more than $1,000,000 a year in revenue must be on the Pro or Mega plan.
  • Black Forest Labs: for self-hosted weights, "Our license allows you to use model outputs commercially, and we make no ownership claim over those outputs."
  • Recraft: paid plans only.
  • Open weights: the licence file decides. Qwen-Image, Qwen-Image-2512, and FLUX.1 [schnell] are Apache 2.0. MiniMax H3 Image allows commercial use with a name display, and needs MiniMax's authorization above $1 million in annual revenue. Qwen-Image-2.1 is non-commercial without a separate licence from Qwen.

Ownership under a maker's terms is separate from copyright law in your country. For AI-written pages, see commercial use of AI-generated HTML.

Put the images on a page

Images sell when they sit on a page: a product shot above a buy button, a poster above a sign-up form. Ask ChatGPT, Gemini, or Claude for one HTML page built around your images, drop the file into Convika's publish tool, and click Get my preview link for a private link that works for 3 minutes, with no account. Publish it from a free account to keep it at one address, with a form that saves what visitors send.

Frequently asked questions

What is the best AI model for image generation right now?

It depends on the job. GPT-Image-2.6 (November 4, 2026) makes 4K images, SVG, and animated GIFs. Nano Banana 2.1 makes 4K images from $0.0336 at 1K. Midjourney V8.2 is the pick for a distinctive look. FLUX 3 Image Klein is the lowest-priced API at $0.009 per 1K image. Read on November 18, 2026.

What is the best AI model for image generation and editing?

GPT-Image-2.6 has markup-guided editing, Nano Banana 2.1 keeps consistency across turns with up to 14 reference images, Midjourney V8.2 has a new Edit Model, and Ideogram 4.5 has Precise Edit. Among open weights, MiniMax H3 Image is built for generation and editing.

What is the best open-source AI image model?

The newest open-weight models on Hugging Face are MiniMax H3 Image (November 3, 2026), FLUX 3 Image Klein (October 2026), and Qwen-Image-2.1 (September 2026), each under its maker's own licence. Qwen-Image, Qwen-Image-2512, and FLUX.1 [schnell] are Apache 2.0.

Which AI image model can I run locally?

FLUX 3 Image Klein, a 5.6B-parameter model, Qwen-Image-2.1, MiniMax H3 Image, and the Apache 2.0 models Qwen-Image and FLUX.1 [schnell] all have weights on Hugging Face. GPT-Image-2.6, Nano Banana 2.1, and Midjourney V8.2 run only on their makers' servers.

What is the best free AI image generator?

Qwen-Image, Qwen-Image-2512, and FLUX.1 [schnell] are Apache 2.0 on Hugging Face, free to run and to use commercially (licences read on November 18, 2026). MiniMax H3 Image is royalty-free with conditions above $1 million in annual revenue. Through an API, FLUX 3 Image Klein costs $0.009 for a 1K image.

Which AI image model is best for text in images?

Ideogram describes Ideogram 4.5 as having "crisp typography". Google calls Nano Banana Pro its "premium choice" to "render high-fidelity text", and lists improved text rendering for Nano Banana 2.1. For text that must stay exact and editable, GPT-Image-2.6 generates SVG.

How much does AI image generation cost?

Per image through an API: $0.009 (FLUX 3 Image Klein, 1K), from $0.02 (Grok Imagine Image 2.0), $0.0336 (Nano Banana 2.1, 1K), and $0.048 (FLUX 3 Image, 1K). By plan: Midjourney from $10 a month, Ideogram Plus $20 a month.

Does Midjourney have an API?

Midjourney's plans and documentation pages mention no API. You use it on midjourney.com or in Discord, on a plan from $10 to $120 a month.

Sources

Each page was read on November 18, 2026, between 03:40 and 04:10 Japan time; dates in brackets are the page's own.

Next step

How to password protect an HTML page or website

A JavaScript password script leaves the password in the page source. To password protect an HTML page or link, encrypt the file or let a server check it.

Related articles

More ways to send your page

Skipbox

Cheap & fast GitHub Actions runners, unlimited for $39 a month

Your own machine keeps Docker layers and package caches between jobs. Switch by changing one runs-on line.

Start a free Skipbox trial⁠

Share this article

Put your images on a page people can open

Ask your AI for one HTML page with the images in it, then drop the file in. Convika gives you a link, keeps the form entries, and shows who visited.

Publish my HTML page free

Free plan: no credit card, 1 published page, 50 form entries a month.