Castaly · Blog · Pricing & billing

AI Image Models Compared: Price and Features of All Ten Castaly Models

GPT Image, Gemini, Seedream and Wan side by side: flat prices per image, how token-billed models estimate and settle, and which model fits which job.

A bare room with worn floorboards and one tall window holding back the sea, a school of fish circling behind the glass, an image made with Seedream 5.0 Lite.

Castaly has ten image models, and they bill in two different ways. Four charge one flat price per image whatever its size. Six bill by tokens, so the size and quality you pick change the price, and the figure the Studio shows before you generate is an estimate rather than the final charge.

This guide compares all ten on price and on what the Studio lets you do with them.

The short answer

One image at the Studio's default settings:

Model Billing Default setting Price
wan2.7-image Flat 2K $0.0212
gemini-2.5-flash-image Flat 1K $0.0330
seedream-5-0-lite-260128 Flat 2K $0.0332
wan2.7-image-pro Flat 2K $0.0530
gemini-3.1-flash-lite-image Tokens 1K $0.028 – $0.057
gemini-3.1-flash-image Tokens 512 $0.038 – $0.076
gpt-image-2.5-flare Tokens 1024 × 1024, medium up to $0.056
gpt-image-2.5-sunburst Tokens 1024 × 1024, medium up to $0.056
gpt-image-2 Tokens 1024 × 1024, auto up to $0.159
gemini-3-pro-image Tokens 1K $0.114 – $0.227

For token-billed models the range runs from the image tokens alone to the Studio's estimate. The sections below explain where each end comes from.

Flat-price models

Four models charge the same for every image, at any size they support:

Model Price per image Sizes Images per request
wan2.7-image $0.0212 1K, 2K up to 12
gemini-2.5-flash-image $0.0330 1K, 2K, 4K 1
seedream-5-0-lite-260128 $0.0332 2K, 3K, 4K, custom 1, or a group of up to 15
wan2.7-image-pro $0.0530 1K, 2K, 4K up to 12

Because the price does not depend on size, choose the largest size you will actually use. A 4K Seedream image costs the same $0.0332 as a 2K one.

A bare room with worn floorboards and one tall window holding back the sea, a school of fish circling behind the glass and water trickling onto the floor.
seedream-5-0-lite-260128 · 2560 × 1440 — $0.0332 at today's price.

One caveat from our own runs: a request for 2K to Gemini 2.5 Flash Image came back at 1344 × 768. If you need a large image from a flat-price model, Seedream 5.0 Lite and Wan 2.7 Image Pro returned the size they were asked for.

GPT Image: quality sets the price

GPT Image 2, GPT Image 2.5 Flare and GPT Image 2.5 Sunburst bill the tokens of the output image, plus a few tokens for your prompt. The two GPT Image 2.5 models share one price: $31.98 per million output tokens and $5.33 per million input tokens. GPT Image 2 is $22.62 and $3.77.

The Studio estimates the output tokens from the size and quality:

Model Size Low Medium High
GPT Image 2.5 Flare / Sunburst 1024 × 1024 $0.0065 $0.0564 $0.2249
GPT Image 2.5 Flare / Sunburst 1536 × 1024 $0.0053 $0.0441 $0.1758
GPT Image 2 1024 × 1024 $0.0046 $0.0399 $0.1591
GPT Image 2 1536 × 1024 $0.0038 $0.0312 $0.1243

Quality is the big lever: high costs four times medium. On GPT Image 2, quality starts at auto, which lets the provider pick, so the estimate assumes high.

The estimate is a hold, not the charge. The final charge uses the tokens the provider reports for the finished image, and our recorded runs often came in well under the estimate:

Run Settings Output tokens reported At today's price
GPT Image 2.5 Flare 1536 × 1024, medium 343 about $0.011
GPT Image 2.5 Sunburst 1536 × 1024, quality not set 1,372 about $0.045
GPT Image 2 1536 × 1024, quality not set 158 about $0.004
A child's picture-book illustration of a small fox carrying a paper lantern through a snowy birch forest at dusk.
gpt-image-2.5-flare · 1536 × 1024 · medium — 343 output tokens and 47 input tokens, about $0.011 at today's price against a $0.044 estimate.

What the 2.5 models add over GPT Image 2 in the Studio: a background option that includes transparent, PNG, JPEG or WebP output with adjustable compression, and a moderation setting. All three accept up to 16 reference images and return up to 10 images per request. Flare is documented as the everyday default for generation and editing, and Sunburst as the higher-fidelity, higher-control option.

Gemini: size sets the price

The three token-billed Gemini models bill each image at the token count Google publishes for its size, at these output rates:

Model Output rate 512 1K 2K 4K
gemini-3.1-flash-lite-image $25.35 / M — $0.028 — —
gemini-3.1-flash-image $50.70 / M $0.038 $0.057 $0.085 $0.128
gemini-3-pro-image $101.40 / M — $0.114 $0.114 $0.203

Those figures are the image alone: 747, 1,120, 1,680 and 2,520 tokens on Flash, and 1,120 or 2,000 on Pro, where 1K and 2K cost the same. Gemini also writes some text with the image, billed at the same output rate, so the Studio's estimate reserves twice the image tokens and refunds what the response does not use.

A flower-shop cold room where every bloom along one steel shelf is turning to blown glass as a figure in an acid-green coat carries a bucket toward the open door.
gemini-3.1-flash-image · 2K · 16:9 — 1,680 image tokens and 411 text tokens reported, about $0.106 at today's price against a $0.170 estimate.

In our recorded runs, a 2K Gemini 3 Pro Image response carried 1,120 image tokens, 110 to 134 text tokens and about 200 thinking tokens, and a 1K Gemini 3.1 Flash Lite response carried 1,120 image tokens and 414 text tokens. Expect the final charge to land between the image-only price and the estimate.

What else differs

Model Reference images Extras in the Studio
gpt-image-2 up to 16 custom sizes
gpt-image-2.5-flare, -sunburst up to 16 transparent background, output format and compression, custom sizes
seedream-5-0-lite-260128 up to 14 groups of up to 15 related images, custom sizes
wan2.7-image, wan2.7-image-pro up to 9 up to 12 images per request
Gemini models yes ratio presets; Flash Lite offers 10 ratios at 1K

Which model for which job

  • Many variations of one idea. Wan 2.7 Image returns up to 12 images in one request at $0.0212 each.
  • A large final image on a fixed budget. Seedream 5.0 Lite at 4K, or Wan 2.7 Image Pro at 4K, at their flat price.
  • A cut-out with a transparent background. GPT Image 2.5 Flare or Sunburst, the only models with a transparent background option.
  • A consistent set, such as storyboard frames. Seedream's group mode returns up to 15 related images from one prompt.
  • Cheap exploration on a token model. GPT Image at low quality is under a cent an image; move to medium or high once the composition is right.
A luxury mechanical watch with a deep green dial resting on wet black slate under dramatic rim lighting.
wan2.7-image-pro · 2K — $0.0530 at today's price.

Open the image Studio →

FAQ

What is the cheapest AI image model on Castaly?

Wan 2.7 Image at $0.0212 per image. GPT Image at low quality can settle lower, but its estimate depends on size and its final charge on the tokens reported.

Why does the Studio show a higher price than I was charged?

For token-billed models the Studio holds an estimate sized for the full image budget at that size and quality. The final charge uses the tokens the provider reports, and the difference goes back to your balance.

Does a bigger image cost more?

On Wan, Seedream and Gemini 2.5 Flash Image, no. On GPT Image and the token-billed Gemini models, yes: more pixels mean more output tokens.

Are Flare and Sunburst priced differently?

No. Both are $5.33 per million input tokens and $31.98 per million output tokens. The final charge depends on how many tokens each image uses.