Skip to main content

Gemini image overview

Common ways to call Gemini image models on Omniall AI: Aspect ratio / resolution via extra_body.google.image_config (Chat) or generationConfig.imageConfig (native), e.g. aspect_ratio / aspectRatio, image_size / imageSize (1K / 2K / 4K). Pick model from the Model Square.

Banana 2.1 (gemini-nano-banana-2.1)

Supports text-to-image, image-to-image, plus PDF and short video reference generation.

Chat: PDF URL

Chat: video URL

Notes:
  • Pass video_url as a string, not { "url": "..." }
  • file requires both filename and file_data
  • URLs must be publicly downloadable; default download limit ~64MB; prefer short MP4 / small PDF
  • Full curl / native examples: Gemini native format and Chat image generation

Optional: Thinking levels

gemini-nano-banana-2.1 supports official Thinking. For typical integrations you can omit this config and use the default level. Usage may include reasoning_tokens / thoughtsTokenCount for thinking cost, regardless of whether thought text is returned. Native — under generationConfig.thinkingConfig:
Chat — under extra_body.google.thinking_config (snake_case required):
Notes: In Chat use thinking_config / thinking_level — camelCase thinkingConfig is rejected. When parsing the response, walk all parts and skip items with thought: true before reading the image. Other model examples: gemini-3.1-flash-image-preview, gemini-3-pro-image-preview (per Model Square).