# Choose the right AI image model before you write the prompt

Compare GPT Image 2, Nano Banana, Seedream and Z-Image Turbo: what each model does best, supported settings, prices, examples and prompt tips.

## Supported models

- [GPT Image 2](https://www.gptimagehub.com/models/gpt-image-2): GPT Image 2 is the best default choice on GPT Image Hub when you need high-quality generation or editing, many aspect-ratio choices, and up to 16 reference images. Use it for polished commercial visuals, product mockups, editorial assets, and prompts where instruction following matters more than the lowest possible iteration latency.
- [GPT Image 2 Saver](https://www.gptimagehub.com/models/gpt-image-2-saver): GPT Image 2 Saver runs OpenAI's gpt-image-2 through a lower-cost provider route that bills per image, so GPT Image Hub charges a flat 5 credits (about $0.005) whether you pick low, medium, or high quality at 1K, 2K, or 4K. In the Hub's tests it took 23 to 43 seconds per image, rendered on-image text accurately, and handled multi-image reference edits. Choose it for high-volume drafts, social content, and trying prompts cheaply; choose standard GPT Image 2 when you need up to 16 reference images.
- [GPT Image 2.5 Flare](https://www.gptimagehub.com/models/gpt-image-2-5-flare): GPT Image 2.5 Flare is OpenAI's default GPT Image 2.5 model for fast, high-quality everyday image generation. OpenAI positions it above GPT Image 2 in quality at roughly half the latency, billed at the same token rates; in GPT Image Hub's own tests a 1536×1024 image at high quality used 1,372 output tokens and 27 seconds, against 5,488 tokens and 78 seconds for GPT Image 2 at the same size. Choose it for most new work; choose Sunburst when editing precision matters most.
- [GPT Image 2.5 Sunburst](https://www.gptimagehub.com/models/gpt-image-2-5-sunburst): GPT Image 2.5 Sunburst is the GPT Image 2.5 model OpenAI recommends when precision matters most. It supports the same xhigh and max quality settings, custom resolutions, and transparent backgrounds as Flare, and GPT Image Hub runs it at xhigh by default; in Hub tests it took 47–108 seconds per image and placed every label and line of text exactly as prompted. Choose it for packaging, typography-heavy layouts and intricate illustration; choose Flare when speed matters more.
- [Nano Banana](https://www.gptimagehub.com/models/nanobanana): Nano Banana is the fast Gemini image option on GPT Image Hub. Use it when you want quick creative iteration, light image edits, and reference-image guided outputs at 1K. It is a good exploration model before moving a winning prompt into GPT Image 2, Nano Banana 2, or Nano Banana Pro for larger exports.
- [Nano Banana 2](https://www.gptimagehub.com/models/nanobanana2): Nano Banana 2 is the balanced Gemini choice on GPT Image Hub. It supports 1K, 2K, and 4K tiers in the app, keeps the fast Flash-style workflow, and allows up to 14 reference images. Choose it when Nano Banana is too limited but Nano Banana Pro is more than the task needs.
- [Nano Banana 2.1](https://www.gptimagehub.com/models/nanobanana2-1): Nano Banana 2.1 is Google's update to Nano Banana 2 and the model Google now recommends for new projects. Its image output costs half as much as Nano Banana 2, so on GPT Image Hub it is 25 credits at 1K, 41 at 2K, and 91 at 4K. In the Hub's tests it rendered a menu with every price correct and turned one concept from a reference sheet into a finished sticker. It thinks before drawing, so 2K and 4K images can take several minutes.
- [Nano Banana Pro](https://www.gptimagehub.com/models/nanobananapro): Nano Banana Pro is the premium Gemini image option on GPT Image Hub. Use it for detailed, high-resolution images, complex visual reasoning, branded scenes, product concepts, and outputs where typography or real-world knowledge may matter. It supports up to 14 reference images and 1K, 2K, and 4K tiers in the app.
- [Nano Banana 2 Lite](https://www.gptimagehub.com/models/nanobanana2-lite): Nano Banana 2 Lite is Google's fastest and cheapest Gemini image model. On GPT Image Hub it renders 1K images for 25 credits, and in the Hub's tests most images finished in 12 to 26 seconds. Google notes that it is not optimized for many reference images or long edit sequences, so choose Nano Banana 2.1 for text-heavy layouts, 2K or 4K output, and multi-image edits.
- [Seedream 4.0](https://www.gptimagehub.com/models/seedream-4-0): Seedream 4.0 is ByteDance's image generation and editing model. On GPT Image Hub it costs 25 credits per image at any size, 1K, 2K or 4K, which makes its 4K output one of the cheapest on the site. In the Hub's tests 1K and 2K images took about 20 seconds and a 4K landscape took 137 seconds. Choose it for photographic scenes and large exports on a budget; choose Seedream 4.5 or 5.0 for stronger text and design work.
- [Seedream 4.5](https://www.gptimagehub.com/models/seedream-4-5): Seedream 4.5 is ByteDance's image model with stronger typography and design control than 4.0. On GPT Image Hub it renders 2K and 4K images for 33 credits each; it has no 1K tier because it needs at least about 3.7 megapixels. In the Hub's tests it rendered clean poster text, replaced a headline and background in a reference edit, and produced a 4K product photo in 28 seconds.
- [Seedream 5.0](https://www.gptimagehub.com/models/seedream-5-0): Seedream 5.0 on GPT Image Hub is ByteDance's lowest-priced 5.0 model: 15 credits per image at 2K or 4K. BytePlus lists the 5.0 family as Pro, Lite, and Flash, and the Hub prices this model as 5.0 Flash. In the Hub's tests it produced a detailed 4K festival illustration in 129 seconds and a three-view character sheet with correct labels in 46 seconds.
- [Seedream 5.0 Pro](https://www.gptimagehub.com/models/seedream-5-0-pro): Seedream 5.0 Pro is ByteDance's most capable Seedream model, aimed at design work with accurate typography. On GPT Image Hub it costs 37 credits at 1K and 73 at 2K, following BytePlus's price step at 2.61 megapixels; it has no 4K tier. In the Hub's tests it rendered a Kyoto festival poster with a large title and a dated subtitle exactly as prompted. It can be slow: that 2K poster took 290 seconds, while a 1K image took 46 seconds.
- [Seedream 5.0 Lite](https://www.gptimagehub.com/models/seedream-5-0-lite): Seedream 5.0 Lite is the lighter model in ByteDance's Seedream 5.0 family. On GPT Image Hub it renders 2K images in eight aspect ratios, from 1:1 to 21:9, for 36 credits each, at BytePlus's full list price. In the Hub's tests it produced natural portraits and interiors in 25 to 33 seconds and relit a portrait from a reference image while keeping the person and scene.
- [Z-Image Turbo](https://www.gptimagehub.com/models/z-image-turbo): Z-Image Turbo is Alibaba Tongyi's fast text-to-image model. On GPT Image Hub it renders 1K and 2K images for 15 credits each, at Alibaba's full list price, in any of the Hub's 16 aspect ratios. In the Hub's tests a 2K product photo took 11 seconds and 1K scenes took 42 to 70 seconds. It does not accept reference images, so use another model for edits.

Compare all models: https://www.gptimagehub.com/models/compare