Image Tools

Use these tools when you encounter images you cannot read natively, or when you need to generate new images.

image_read

Send an image to a vision-capable model and get a text description.

Local file (most common):

tai tool image_read --image_path /path/to/image.png --prompt "Describe this image"

URL:

tai tool image_read --image_path https://example.com/photo.jpg --prompt "What is shown?"

With a specific vision provider:

tai tool image_read --image_path /path/to/image.png --prompt "Describe" --provider llm.my-openai:gpt-4o

Images are automatically resized (preserving aspect ratio) before sending to the vision model. Supported formats: PNG, JPEG, GIF, WebP.

image_generate

Generate a new image from a text prompt (text-to-image). For editing an existing image, use image_edit instead.

Basic usage (always specify output):

tai tool image_generate --prompt "A serene mountain landscape at sunset" --output landscape.png

With specific provider, model and size:

tai tool image_generate --prompt "A futuristic city skyline" --provider llm.my-openai --model gpt-image-1 --dimensions 1792x1024 --output output/city.png

Transparent background (for icons, stickers, product shots):

tai tool image_generate --prompt "A cute fox mascot" --background transparent --output_format png --output fox.png

WebP output with quality control:

tai tool image_generate --prompt "A landscape painting" --output_format webp --quality high --output painting.webp

If output is omitted, the image is saved to a default path in the working directory.

image_edit

Edit or transform an existing image based on a text prompt (image-to-image). Use for style transfer, background replacement, adding/removing elements, or any modification that requires a reference image.

Basic usage:

tai tool image_edit --image_path /path/to/photo.png --prompt "Change the background to a beach scene" --output edited.png

With URL image:

tai tool image_edit --image_path https://example.com/photo.jpg --prompt "Make it look like a watercolor painting" --output watercolor.png

With specific provider and model:

tai tool image_edit --image_path /path/to/original.png --prompt "Remove the person in the foreground" --provider llm.my-openai --model gpt-image-1 --dimensions 1024x1024 --output result.png

With mask (edit only the masked region):

tai tool image_edit --image_path /path/to/photo.png --mask /path/to/mask.png --prompt "Replace with a garden" --output edited.png

Transparent background edit:

tai tool image_edit --image_path /path/to/product.png --prompt "Remove background" --background transparent --output_format png --output cutout.png

If output is omitted, the image is saved to a default path in the working directory.

image_providers

List available image providers filtered by capability.

List image generation providers (default):

tai tool image_providers

List image editing providers:

tai tool image_providers --capability image_editing

List vision (image reading) providers:

tai tool image_providers --capability vision

Returns a list of providers with their available models and connector IDs that can be passed to image_generate, image_edit, or image_read.

Constraints

Use only the parameters listed above for each tool. The supported first-class parameters are: prompt, output, provider, dimensions, model, background, output_format, output_compression, quality, mask (edit only), and extra.

Do not pass n, style, or response_format — they are unsupported and will be ignored or cause errors.

For provider-specific parameters not covered above (e.g. moderation, input_fidelity), pass them through the extra JSON object.