Tools

Image Generator

15+ image models with reference images, inpainting, and per-model controls.

The Image Generator is a chat-style workspace: your prompt bar sits at the bottom, results flow upward, and older generations load as you scroll. Click any image to open it full-screen, where you can Download it, delete it, or Publish to Community.

Generating

Type a prompt and pick your options from the pills on the prompt bar:

  • Model: the pill shows a per-image credit guide for the selected model.
  • Aspect ratio: each model lists the ratios it supports (Nano Banana 2 and 2 Lite offer 14). The ratio pill is hidden for models whose shape comes from a size preset or the input image.
  • Count: 1 to 4 images per generation.
  • Settings: per-model options such as resolution, size, quality, seed, negative prompt, guidance, steps, and output format.

Reference images

Models that support references let you steer a generation with existing images. Three ways to add one:

  1. The image button on the prompt bar.
  2. Paste an image straight from your clipboard.
  3. Drag a file from your desktop, or any previously generated image from your history.

Multiple references are allowed. GPT Image 2 switches into edit mode automatically when a reference is attached and uses up to 4 references. Seedream 5 Pro uses up to 10.

Inpainting (GPT Image 2)

With GPT Image 2 and a reference image attached, Settings shows a Mask upload: a PNG with transparency, matching the reference's dimensions. Transparent areas get repainted from your prompt; opaque areas are preserved. When a mask is attached, only the first reference image is used.

Models

ModelCredits per imageReferences
Nano Banana 2 (default)90 / 134 / 202 / 302 at 0.5K / 1K / 2K / 4KYes
Nano Banana 2 Lite46 / 68 / 102 / 154 at 0.5K / 1K / 2K / 4KYes
Nano Banana80Yes
Nano Banana Pro270 at 1K or 2K, 480 at 4KYes
Seedream 5 Pro135, or 270 at 2048×2048Yes (up to 10)
Seedream 5 Lite70No
Seedream 4.580No
Seedream 4 Edit / Text To Image60Edit: yes
Seedream 360No
Flux Kontext Pro80Yes
Flux 2 Flex240, rising with references (500 with 4)Yes
GPT Image 212 to 1,741 by quality and size (see below)Yes (up to 4) + mask
GPT Image 1160 at 1:1, 240 at other ratiosNo
Runway100 at 16:9 or 9:16, 160 at 1:1Yes
Qwen Image 2 / Edit70Yes
Z Image20No

The model pill shows the exact credits for your current settings, image count and references, so what you see is what you're charged. Multiply the table by your image count. Reference images add about 1 credit each on the Nano Banana models. Failed generations are refunded automatically.

GPT Image 2 is priced by quality and size. Auto quality is billed as High.

SizeLowMediumHigh
1024×102412106422
1536×1024 or 1024×15361282330
2048×115215123495
2048×204827219880
3840×2160 (4K)534331,741

Model highlights:

  • Nano Banana 2: resolution tiers from 0.5K to 4K (1K by default), a Thinking Level control (High does better on complex prompts), and Search Grounding, which pulls real-time facts from Google Search into the generation.
  • Seedream 5 Pro and Seedream 5 Lite: an Image Size preset in Settings, up to 2048×2048 (XL) on Pro and 3072×3072 (HD) on Lite. Seedream 5 Pro takes up to 10 reference images.
  • GPT Image 1: renders at Medium quality.
  • GPT Image 2: sizes up to 3840×2160, Quality (Auto, Low, Medium, High), PNG/JPEG/WebP output, Background (Auto or Opaque; transparent backgrounds aren't supported), and Moderation (Auto or Low).
  • Z Image: the budget workhorse for fast drafts.

Tips

  • Start cheap. Iterate composition with Z Image or a Lite model, then re-run the winning prompt on a premium model at high resolution.
  • Drag a generated image back into the prompt bar as a reference to iterate on it.
  • Deleting an image removes all images from the same generation.
  • A generation that's still running can be cancelled from the queue panel in the bottom-right corner; its credits are refunded.

On this page