What are the main AI image models and how do they differ?

The main AI image models in FLORA include GPT Image 2.5, Nano Banana 2.1 and Nano Banana Pro, Seedream 5 Pro, Ideogram 4.5, FLUX 3 Image, Qwen Image 3.0 Pro and Recraft V4.1. They differ in price, from under $0.01 to $0.36 per text-to-image run, in speed, in how well they print text and in the input shapes they accept.

At a glance

The options

1. FLORA

   The canvas and MCP server that put every model below behind one account. Each model is quoted in dollars before it runs, and every result is saved with the prompt and model that made it.

   • flora_list_models returns the live catalogue with estimated cost and time
   • Up to 20 generations per call, across any mix of models
   • Image, video, audio and 3D on one canvas

2. GPT Image 2.5 (Flare and Sunburst)

   Ids t2i-gpt-image-2-5-flare and t2i-gpt-image-2-5-sunburst, with i2i and is2i versions. Estimated $0.055 per text-to-image run and $0.064 per image-to-image run, about 25 to 35 seconds. Aspect ratios from 1:2 to 21:9 at 1K, 2K or 4K.

   Best for: Photographic scenes, products and layouts; FLORA's default image model

3. Nano Banana 2.1

   Ids t2i-elote-gateway, i2i-elote-gateway and is2i-elote-gateway. Estimated $0.052 per image and about 40 seconds; our run took 22 seconds.

   Best for: Edits to an existing image and multi-reference composites

4. Nano Banana Pro

   Ids t2i-gemini-3-pro, i2i-gemini-3-pro and is2i-gemini-3-pro. Estimated $0.18 per image and 80 to 90 seconds; our run took 29 seconds.

   Best for: Final images where detail matters more than cost

5. Seedream 5 Pro

   Ids t2i-gengateway-seedream-5-pro-t2i plus i2i and is2i versions. Estimated $0.17 per image and about 130 seconds. Our run returned 2304 x 1728 at 4:3 in 57 seconds.

   Best for: Large, high-resolution outputs straight from the model

6. Ideogram 4.5

   Ids t2i-ideogram-4-5-t2i, i2i-ideogram-4-5-i2i and is2i-ideogram-4-5-is2i. Estimated $0.12 for text-to-image and $0.072 for edits, about 20 seconds.

   Best for: Posters, labels and packaging with exact wording

7. FLUX 3 Image and FLUX 3 Image Klein

   Ids t2i-flux-3-image-t2i (estimated $0.05) and t2i-flux-3-image-klein-t2i (estimated $0.009, about 10 seconds). FLUX 3 Image also has i2i and is2i versions.

   Best for: Klein for cheap drafts and volume; FLUX 3 Image for clean text and finals

8. Qwen Image 3.0 Pro

   Id t2i-qwen-image-3-pro-t2i. Estimated $0.048 per image and about 30 seconds. Our run returned 1728 x 1296 at 4:3 in 21 seconds.

   Best for: Clean product and still-life images at a low price

9. Recraft V4.1 family

   Ids include t2i-recraft-v41 ($0.048), t2i-recraft-v41-flash-t2i ($0.00735, about 5 seconds) and t2i-recraft-v4-1-pro-vector-t2i ($0.36).

   Best for: Vector output and illustration styles

What happened when we ran image models in FLORA?

On October 16, 2026, we called flora_list_models from Claude Code with type image. It returned 237 image model entries, with FLORA's featured models first, each with an id, capabilities, estimated cost in USD and estimated seconds.

We then sent one prompt to four models at different price points in one flora_create_generations call, all at aspect_ratio 4:3: "Close-up of a hand-knit oatmeal wool sweater folded on a wooden chair seat, overcast window light, visible yarn texture, muted natural color, true-to-life detail."

FLUX 3 Image Klein cost $0.009 and took 14 seconds. Qwen Image 3.0 Pro cost $0.048 and took 21 seconds. GPT Image 2.5 Flare cost $0.055 and took 30 seconds. Nano Banana Pro cost $0.18 and took 29 seconds. The four came to $0.292.

The three higher-priced models all produced a folded crew-neck sweater on a wooden chair by a window in soft, muted light. Nano Banana Pro added folded sleeves and warmer, higher-contrast light; Qwen Image 3.0 Pro came out coolest and grayest. What went wrong: Klein's knit changes pattern partway across and the piece reads more like a rolled blanket than a sweater, with no neckline. At $0.009 it is still the right model for blocking out compositions, but rerun the final on a stronger model.

Tools called

flora_list_models, flora_create_generations, flora_list_generations

Models run

FLUX 3 Image Klein, Qwen Image 3.0 Pro, GPT Image 2.5 Flare, Nano Banana Pro

Cost

$0.292 for four images

Time

14 to 30 seconds per image, run in parallel

Cheap draft models or premium models: pros and cons

What do t2i, i2i and is2i mean?

They are the input shapes in FLORA model ids. A t2i model takes text only. An i2i model takes one reference image plus text, for edits and new scenes of the same subject. An is2i model takes several reference images, such as a product and a background. The same model often has all three, priced separately.

• t2i: text to image, for new images from a description.
• i2i: image to image, for edits and consistent subjects.
• is2i: images to image, for combining several references.

Which image model should I start with?

Start with GPT Image 2.5 Flare for photographic work: it is FLORA's default, about $0.055 per image and around 30 seconds in our runs. Move to another model when a job has a specific need.

• Exact text on packaging or posters: Ideogram 4.5 or FLUX 3 Image.
• Edits to an image you already have: an i2i model such as i2i-gpt-image-2-5-flare or Nano Banana 2.1.
• Many cheap drafts: FLUX 3 Image Klein or Recraft V4.1 Flash.
• Vector files: a Recraft vector model.

What else is in the catalogue?

Beyond the featured models, FLORA lists earlier versions and other families, so you can match a look made with an older model. The list includes GPT Image 1.5 and GPT Image 2, Nano Banana and Nano Banana 2, Seedream 3.0 to 5.0 Lite, Ideogram 3.0 and 4.0, the Flux 2 family, Imagen 4, Grok Imagine, Riverflow, Uni-1, Wan 2.2, Z-Image Turbo and Stable Diffusion 3.5. Upscalers, background removal, inpainting and outpainting models sit in the same list.

How do I check prices before I run anything?

Ask your agent to quote. flora_create_generations with quote set to true returns the dollar price of each request without starting it. Our quotes matched the final charges to the cent on every image model we ran.

Frequently asked questions