Every AI Image Model in Nenis Compared: Which One Should You Use?

Nenis has eight image models from Google, OpenAI, Krea, and ByteDance. This guide compares what you actually get in Nenis: credit prices, sizes, reference photos, results per run, and the job each model is best at.

If you want one rule before the tables:

Start with Nano Banana 2.1. Switch models for a specific reason: a precise edit, a transparent background, ten results at once, or the lowest price per image.

The eight models at a glance

Credits are per image, with no reference photos, at each model's own sizes.

A range is not a quality score. GPT Image 2.5's range runs from Low at 1K to Max at 4K. Seedream charges the same 5 credits at every size it offers, from 2K up.

Nano Banana 2 is also still in the menu until Google retires it on October 29, 2026. Use Nano Banana 2.1 instead: it is newer and cheaper at every size the two share. See the Nano Banana 2.1 guide.

The fast answer by task

Google: Nano Banana 2.1, Pro, and Lite

Nano Banana 2.1

The Google model to start with. It costs 5, 7, or 11 credits at 1K, 2K, and 4K, accepts up to 14 reference photos, and returns up to four results. It has the widest choice of formats in Nenis, 14 of them up to 8:1, plus optional web and image search and a Quick or Careful planning setting.

Reference photos add about 0.2 credits each. Read the Nano Banana 2.1 guide.

Nano Banana Pro

Nano Banana Pro costs 16 credits at 1K or 2K and 28 at 4K. It accepts 14 reference photos and is the only Google model that returns up to ten results in one run.

Google's own scores put 2.1 ahead of Pro, at under half the price. Choose Pro when you need ten results at once, or when your existing Pro prompts already give you exactly what you want. Read the Nano Banana Pro guide.

Nano Banana 2 Lite

The cheapest Google model: 3 credits at 512px and 4 at 1K. It accepts 14 reference photos and returns up to four results, but it has no search and stops at 1K.

Use it for throwaway drafts, moodboards and quick reference tests. Read the Nano Banana 2 Lite guide.

OpenAI: GPT Image 2.5 Flare and Sunburst

Flare and Sunburst share one price list and the same controls: 1K, 2K, and 4K, five quality levels from Low to Max, up to 16 reference photos, up to ten results, and transparent backgrounds.

Each reference photo adds about 1.4 credits, because OpenAI bills every photo it reads.

• Flare is the fast one. Start here for new images, quick edits, and many variations. Read the Flare guide.
• Sunburst is slower and more careful. Use it when a near miss is expensive: replacing one product, changing clothing without changing the person, or keeping packaging exact. Read the Sunburst guide.

Krea: Krea 3

Krea 3 is built for editing. Describe what should change and it leaves the rest of the photo as it was: the same label in a new colour, the same room at night. It costs 4 credits at 1K and 14 at 2K, and each reference photo adds about 3.5 credits. It takes up to eight photos and returns up to four results.

A 1K image from a prompt costs the same as on Nano Banana 2 Lite. An edit with several photos costs more than on any other model here: 18 credits at 1K with four photos. Planning offers Quick and Careful at the same price; Careful is slower. On edits there is also Ultra, which makes several versions and keeps the best, for 38 credits with one photo. Read the Krea 3 guide.

ByteDance: Seedream 5.0 Lite and 4.5

Both Seedream models cost 5 credits per image at every size, 2K and up. They accept 14 reference photos at no extra charge and return up to ten results, in eight formats.

• Seedream 5.0 Lite offers 2K, 3K, and 4K. Choose it when the brief needs reasoning: several subjects with clear relationships, information graphics, or an edit where the model must infer what you meant. Read the Seedream 5.0 Lite guide.
• Seedream 4.5 offers 2K and 4K. It remains a stable choice for reference-led edits, especially when an existing 4.5 workflow already works. Read the Seedream 4.5 guide.

Neither offers 1K. For many large images at a low, predictable price, start with Seedream.

Compare by size and reference photos

Nenis adds the exact parts together and rounds up once per image, so a photo worth 0.2 credits does not cost you a whole credit. The estimate in Create always shows the final price before you generate.

When you upload several photos, give each one a numbered job:

Image 1: exact subject or product.
Image 2: composition only.
Image 3: lighting only.

Keep: [details that must not change].
Ignore: text, logos and people in images 2 and 3.

For more, read How to use reference images in AI.

Models that have left Nenis

• GPT Image 2, GPT Image 1.5 and GPT Image 1 Mini were replaced by GPT Image 2.5 on September 11, 2026. Flare covers everyday and cheap work; Sunburst covers careful editing.
• Seedream 4 was replaced by Seedream 5.0 Lite on September 11, 2026, at 5 credits.
• Nano Banana 2 leaves on October 29, 2026. Its replacement is Nano Banana 2.1.

Your earlier images stay in your gallery whichever model made them.

The recommendation

• Start with Nano Banana 2.1 for almost everything. It is cheap, fast, and broad.
• Draft at 1 credit with GPT Image 2.5 Flare on Low, or at 3 to 4 with Nano Banana 2 Lite.
• Use Krea 3 or GPT Image 2.5 Sunburst when an edit must change one thing and nothing else, and Krea 3 on Ultra when it also has to get exact words right.
• Use Seedream 5.0 Lite for large images and big batches at 5 credits each.
• Use GPT Image 2.5 for transparent backgrounds and up to 16 reference photos.
• Keep Nano Banana Pro for ten results at once or for proven Pro workflows.

Draft cheaply. Switch models for a reason. Stop when the image is ready.

Nenis also has the Upscaler and Background Remover. They are tools for finished images, not generation models, so they are not compared here.

Catalogue note

Models, controls, and credit prices were checked against the Nenis model catalogue on October 21, 2026. Model makers can change their models and prices; Nenis always shows the final credit estimate before you create.